diff --git a/cosmos_framework/model/generator/mot/maskless_attention.py b/cosmos_framework/model/generator/mot/maskless_attention.py
index 2e4d267b1..f8596a19b 100644
--- a/cosmos_framework/model/generator/mot/maskless_attention.py
+++ b/cosmos_framework/model/generator/mot/maskless_attention.py
@@ -432,7 +432,7 @@ def gapped_replay_unavailable_reason(device: torch.device) -> str | None:
     if torch.version.hip is not None:
         return "it needs NVIDIA CUDA, not ROCm"
     major, minor = torch.cuda.get_device_capability(device)
-    if (major, minor) != (9, 0) and major != 10:
+    if (major, minor) not in ((9, 0), (12, 0)) and major != 10:
         # Hopper and Blackwell use different FA4 kernels; the replay GPU tests cover
         # the segment isolation and untouched gap rows required by both paths.
         return f"it needs FlashAttention-4 on Hopper (sm90) or Blackwell (sm100/sm103), got sm{major}{minor}"
diff --git a/pyproject.toml b/pyproject.toml
index dc1ad0905..20d66f371 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -196,7 +196,7 @@ vllm = [
 # Match nvcr.io/nvidia/pytorch:25.11-py3: https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-26-01.html
 cu128 = [
   # FA4 CuTeDSL kernels; the cu13 extra is only needed by CUDA 13 environments.
-  "flash-attn-4==4.0.0b24",
+  "flash-attn-4==4.0.0b30",
   "flash-attn-3-nv==1.0.3+cu128.torch210; platform_machine == 'x86_64'",
   "flash-attn==2.7.4.post1+cu128.torch210; platform_machine == 'x86_64'",
   "natten==0.21.6.dev6+cu128.torch210; platform_machine == 'x86_64'",
@@ -231,7 +231,7 @@ cu128-train = [
 # Match nvcr.io/nvidia/pytorch:25.11-py3: https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-25-11.html
 cu130 = [
   # Match imaginaire4's FA4 installation on both x86_64 and aarch64.
-  "flash-attn-4[cu13]==4.0.0b24",
+  "flash-attn-4[cu13]==4.0.0b30",
   "flash-attn-3-nv==1.0.3+cu130.torch210; platform_machine == 'x86_64'",
   "flash-attn==2.7.4.post1+cu130.torch210; platform_machine == 'x86_64'",
   # GB300/B300 (sm_103a): use the custom build that includes sm_103a kernels (both arches).
@@ -267,7 +267,7 @@ cu130-train = [
 ]
 # torch 2.13 on CUDA 13.0. Same layout as cu130 but pinned to torch 2.13.
 cu130-torch213 = [
-  "flash-attn-4[cu13]==4.0.0b24",
+  "flash-attn-4[cu13]==4.0.0b30",
   #"flash-attn-3-nv==1.0.3+cu130.torch210; platform_machine == 'x86_64'", # TODO: torch213 build (cu130.torch210 wheels are cp313-only + torch210 ABI)
   #"flash-attn==2.7.4.post1+cu130.torch210; platform_machine == 'x86_64'", # TODO: torch213 build (cu130.torch210 wheels are cp313-only + torch210 ABI)
   # torch2.13 build (natten 0.21.6, sm 80/86/89/90/100/103/110/120/121) — matches this group's torch/ABI.
diff --git a/uv.lock b/uv.lock
index 22310a24e..774e41e91 100644
--- a/uv.lock
+++ b/uv.lock
@@ -2816,7 +2816,7 @@ provides-extras = ["automl", "automl-prompts", "reasoner-eval", "guardrail", "in
 cu128 = [
     { name = "flash-attn", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==2.7.4.post1+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "flash-attn-3-nv", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==1.0.3+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
-    { name = "flash-attn-4", marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==0.21.6.dev6+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas-cu12", marker = "python_full_version == '3.13.*'", specifier = "==12.8.4.1" },
     { name = "nvidia-cuda-cupti-cu12", marker = "python_full_version == '3.13.*'", specifier = "==12.8.90" },
@@ -2841,7 +2841,7 @@ cu128 = [
 cu128-train = [
     { name = "flash-attn", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==2.7.4.post1+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "flash-attn-3-nv", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==1.0.3+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
-    { name = "flash-attn-4", marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==0.21.6.dev6+cu128.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas-cu12", marker = "python_full_version == '3.13.*'", specifier = "==12.8.4.1" },
     { name = "nvidia-cuda-cupti-cu12", marker = "python_full_version == '3.13.*'", specifier = "==12.8.90" },
@@ -2869,7 +2869,7 @@ cu128-train = [
 cu130 = [
     { name = "flash-attn", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==2.7.4.post1+cu130.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "flash-attn-3-nv", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==1.0.3+cu130.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
-    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'aarch64'", specifier = "==0.21.6.dev6+cu130.torch210.gb300", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==0.21.6.dev6+cu130.torch210.gb300", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas", marker = "python_full_version == '3.13.*'", specifier = "==13.1.0.3" },
@@ -2893,7 +2893,7 @@ cu130 = [
     { name = "torchvision", marker = "python_full_version == '3.13.*'", specifier = "==0.25.0+cu130", index = "https://download.pytorch.org/whl" },
 ]
 cu130-torch213 = [
-    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*'", specifier = "==0.21.6+cu130.torch213", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas", marker = "python_full_version == '3.13.*'", specifier = "==13.1.1.3" },
     { name = "nvidia-cuda-cupti", marker = "python_full_version == '3.13.*'", specifier = "==13.0.85" },
@@ -2916,7 +2916,7 @@ cu130-torch213 = [
     { name = "torchvision", marker = "python_full_version == '3.13.*'", specifier = "==0.28.0+cu130", index = "https://download.pytorch.org/whl" },
 ]
 cu130-torch213-train = [
-    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*'", specifier = "==0.21.6+cu130.torch213", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas", marker = "python_full_version == '3.13.*'", specifier = "==13.1.1.3" },
     { name = "nvidia-cuda-cupti", marker = "python_full_version == '3.13.*'", specifier = "==13.0.85" },
@@ -2944,7 +2944,7 @@ cu130-torch213-train = [
 cu130-train = [
     { name = "flash-attn", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==2.7.4.post1+cu130.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "flash-attn-3-nv", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==1.0.3+cu130.torch210", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
-    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b24" },
+    { name = "flash-attn-4", extras = ["cu13"], marker = "python_full_version == '3.13.*'", specifier = "==4.0.0b30" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'aarch64'", specifier = "==0.21.6.dev6+cu130.torch210.gb300", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "natten", marker = "python_full_version == '3.13.*' and platform_machine == 'x86_64'", specifier = "==0.21.6.dev6+cu130.torch210.gb300", index = "https://nvidia-cosmos.github.io/cosmos-dependencies/v1.5.0" },
     { name = "nvidia-cublas", marker = "python_full_version == '3.13.*'", specifier = "==13.1.0.3" },
@@ -4673,27 +4673,27 @@ wheels = [
 
 [[package]]
 name = "flash-attn-4"
-version = "4.0.0b24"
+version = "4.0.0b30"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "apache-tvm-ffi", version = "0.1.14.post1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "einops", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "nvidia-cutlass-dsl", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "quack-kernels", version = "0.5.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "nvidia-cutlass-dsl", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "quack-kernels", version = "0.6.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "torch", version = "2.10.0+cu128", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "torch", version = "2.10.0+cu130", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "torch", version = "2.13.0+cu130", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train')" },
     { name = "torch-c-dlpack-ext", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "typing-extensions", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/32/74/1303dfd7a38d91c0e9717c8367b30cdf6d249fb1a42feabceeda1ed86fbf/flash_attn_4-4.0.0b24.tar.gz", hash = "sha256:6c4b981ef433882871ded48317deaa18ea22f731ca4f8b9387804bfa2e8078e2", size = 358637, upload-time = "2026-07-29T09:02:12.753Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/10/a4/e6ced2048a081cfbb3be423bfc39e4046829533c74f259f1b02f5a66a615/flash_attn_4-4.0.0b30.tar.gz", hash = "sha256:fa3d1cc0318f6d4693366167be5d7aaf9199daf821f875af04db25ecfb62c28b", size = 380989, upload-time = "2026-09-09T08:20:47.785Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e5/82/861caf5303cde9ba1499904d63be94b1464a4c0932215f2e77500e5387ab/flash_attn_4-4.0.0b24-py3-none-any.whl", hash = "sha256:c1dcf0dfcf37c4496728547dcb3c1e66d7dcaa07cfedef0f26ccc4d74453951f", size = 383266, upload-time = "2026-07-29T09:02:11.189Z" },
+    { url = "https://files.pythonhosted.org/packages/87/af/5fec32fc6e41ac0584f174275821483d801df5e8239d917a2102c8c6e419/flash_attn_4-4.0.0b30-py3-none-any.whl", hash = "sha256:326585b153feb101c61571626bd88ac6b4977ae0301d472cc220c28c9c4f78c9", size = 406189, upload-time = "2026-09-09T08:20:46.166Z" },
 ]
 
 [package.optional-dependencies]
 cu13 = [
-    { name = "nvidia-cutlass-dsl", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, extra = ["cu13"], marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "nvidia-cutlass-dsl", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, extra = ["cu13"], marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero')" },
 ]
 
 [[package]]
@@ -4717,7 +4717,7 @@ dependencies = [
     { name = "numpy" },
     { name = "nvidia-cudnn-frontend" },
     { name = "nvidia-cutlass-dsl", version = "4.4.2", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version < '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "nvidia-cutlass-dsl", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "nvidia-ml-py" },
     { name = "packaging" },
     { name = "requests" },
@@ -8827,6 +8827,16 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/41/bc/83f5426095d93694ae39fe1311431b5d5a9bb82e48bf0dd8e19be2765942/nvidia_cuda_cupti_cu12-12.8.90-py3-none-win_amd64.whl", hash = "sha256:bb479dcdf7e6d4f8b0b01b115260399bf34154a1a2e9fe11c85c517d87efd98e", size = 7015759, upload-time = "2025-03-07T01:51:11.355Z" },
 ]
 
+[[package]]
+name = "nvidia-cuda-nvdisasm"
+version = "13.3.29"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8c/08/c9405b910ec88e7b6dc889450b1a7b5d379aa0981263290f056144f89d0b/nvidia_cuda_nvdisasm-13.3.29-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ea6f6f403a0aaa5d4240d1b26f7a70496719465f226c01ba8d1ad19b338b4fb6", size = 4744516, upload-time = "2026-05-26T16:35:48.845Z" },
+    { url = "https://files.pythonhosted.org/packages/87/96/4d77373913d602772a503e5c73475b043878b84d95906878f7eb34a0d886/nvidia_cuda_nvdisasm-13.3.29-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d412810102e838b4429a221a872c1ed77b398ac919d9818d0123bd3937a110dc", size = 4782826, upload-time = "2026-05-26T16:36:08.685Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/79/2162190499f5259ac7557720fc3236c9c1f696d50b5cb9ba134eb183d6e1/nvidia_cuda_nvdisasm-13.3.29-py3-none-win_amd64.whl", hash = "sha256:b5f96e500f76fb3123cf27085945d708bd8b94a6f7cf27fff5b6fbb0524c993c", size = 5015371, upload-time = "2026-05-26T17:14:25.697Z" },
+]
+
 [[package]]
 name = "nvidia-cuda-nvrtc"
 version = "13.0.88"
@@ -9444,7 +9454,7 @@ wheels = [
 
 [[package]]
 name = "nvidia-cutlass-dsl"
-version = "4.6.0.dev0"
+version = "4.6.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
     "python_full_version >= '3.13' and platform_machine == 'x86_64' and sys_platform == 'linux' and extra != 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero'",
@@ -9522,10 +9532,11 @@ resolution-markers = [
     "python_full_version >= '3.13' and sys_platform == 'darwin' and extra != 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra != 'group-16-cosmos-framework-libero'",
 ]
 dependencies = [
-    { name = "nvidia-cutlass-dsl-libs-base", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl-libs-base", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl-libs-cu12", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/95/80/9e563d5b9e670761c930505e68e3826711ce1d32db5a992db6fae4aa1961/nvidia_cutlass_dsl-4.6.0.dev0-py3-none-any.whl", hash = "sha256:98dfd40fabc6c0db610eeeaea403f0bb9e2aec0bc699ae0cdf475fa4a54710ca", size = 10234, upload-time = "2026-06-14T04:18:51.719Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/ec/14e6ecbfed31ec35bbd1bb6965ae61f879370906a8f9c2e851e292704ba4/nvidia_cutlass_dsl-4.6.2-py3-none-any.whl", hash = "sha256:06ac62deb182a852dfb053032cf945bf02b9aa1bf502a18b129d280d7babb26c", size = 10460, upload-time = "2026-08-05T14:39:41.657Z" },
 ]
 
 [package.optional-dependencies]
@@ -9677,7 +9688,7 @@ wheels = [
 
 [[package]]
 name = "nvidia-cutlass-dsl-libs-base"
-version = "4.6.0.dev0"
+version = "4.6.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
     "python_full_version >= '3.13' and platform_machine == 'x86_64' and sys_platform == 'linux' and extra != 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero'",
@@ -9758,40 +9769,83 @@ dependencies = [
     { name = "cuda-python", version = "12.9.4", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "cuda-python", version = "13.0.3", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'extra-16-cosmos-framework-train') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra != 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "numpy", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cuda-nvdisasm", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl-libs-core", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "protobuf", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "typing-extensions", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fd/d1/5f4647b12473308229518b528af0fcb4a5339d20c300820d1c68d523d04f/nvidia_cutlass_dsl_libs_base-4.6.2-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:42e5db1c338c6f14726afe01b451899e8b8a50f1f40934eb9d5bd8df303b1154", size = 3320170, upload-time = "2026-08-05T14:10:22.493Z" },
+    { url = "https://files.pythonhosted.org/packages/26/a7/2da08f68c683246f324a50ae0fb7a9b81bfde6633a56382ec22fc4930108/nvidia_cutlass_dsl_libs_base-4.6.2-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:f15edb860162f0601651115d761ec15dc57244808cc47a08b95b9aec83c68f3f", size = 2824919, upload-time = "2026-08-05T14:11:18.112Z" },
+    { url = "https://files.pythonhosted.org/packages/70/0e/5020da76c7dd3ed74acc6f435e53ca2df967762b2d8408f3e81f146e018a/nvidia_cutlass_dsl_libs_base-4.6.2-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:aae76fcc94c5483279f06848ee470b87417b9b1a4ad8f822a12e2654cbaf513a", size = 3321401, upload-time = "2026-08-05T14:11:32.779Z" },
+    { url = "https://files.pythonhosted.org/packages/12/ce/dfefb22eb438de1e0e6c6627f188449ad3f65448522ba80a12a67bb14715/nvidia_cutlass_dsl_libs_base-4.6.2-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:551097a99537ffdd239678088865ab8d1e273633f0961493509284d8445dd420", size = 2827063, upload-time = "2026-08-05T14:11:54.706Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/f3/4be98f2faa283471e355a42ecbc20956fb2c3b6dbc71861f483be277e99a/nvidia_cutlass_dsl_libs_base-4.6.2-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:e129db7155d56ba4478be098c0f819e37690a6ce5d43a9a9a83bf1300d32c57f", size = 3321894, upload-time = "2026-08-05T14:12:30.139Z" },
+    { url = "https://files.pythonhosted.org/packages/54/1b/cf4a0837054bb4b9dc36cdd86f9d2e7fead3629e3373aaa7f7ec1e5e1542/nvidia_cutlass_dsl_libs_base-4.6.2-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:24dfaddad6077fd0de14eecaf66a72762f2f5f30ae1872231f5bf8b243ac2eac", size = 2824987, upload-time = "2026-08-05T14:12:49.618Z" },
+    { url = "https://files.pythonhosted.org/packages/41/e9/a402e079e926f1fbcf82e30dfcd87aa924aaf91b02db08d57e83b158dbb2/nvidia_cutlass_dsl_libs_base-4.6.2-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:334a9be3c5054286666b14f81c9278075c2007b947ba75f7ee44b94aadf26415", size = 3321794, upload-time = "2026-08-05T14:13:12.501Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/4c/0a6c6e785e8fb2bfc6a0941559d30fbaf6264ae77ee1a26a42e37fd6efdc/nvidia_cutlass_dsl_libs_base-4.6.2-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:896f8a87d0815e59066b0607930c9905a5cf8f5c7a01a423b0093bc876bac4b4", size = 2825075, upload-time = "2026-08-05T14:13:34.746Z" },
+]
+
+[[package]]
+name = "nvidia-cutlass-dsl-libs-core"
+version = "4.6.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cuda-python", version = "12.9.4", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "cuda-python", version = "13.0.3", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'extra-16-cosmos-framework-train') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra != 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "numpy", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cuda-nvdisasm", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "protobuf", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "typing-extensions", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7c/ef/10c0089234a32f1379e1f213f28f8c7521a9ff0154e3acc58f613d5ef3be/nvidia_cutlass_dsl_libs_core-4.6.2-py3-none-any.whl", hash = "sha256:571d4b46fca1bfbb123d0dc348eaa7fd80af0014ddffc07b849e8195b2364333", size = 772393, upload-time = "2026-08-05T14:09:50.76Z" },
+]
+
+[[package]]
+name = "nvidia-cutlass-dsl-libs-cu12"
+version = "4.6.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cuda-python", version = "12.9.4", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and sys_platform == 'linux' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "cuda-python", version = "13.0.3", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'extra-16-cosmos-framework-train') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra == 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and sys_platform != 'linux' and extra != 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra != 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "numpy", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cuda-nvdisasm", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl-libs-base", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "protobuf", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "typing-extensions", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/fa/4d/22e1cd5be088668029e8c53f7b6a92d16e919525fa8535536f53956748ae/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:516042afd42419aa085120950415ac2a49333a7f821756647ebc2e142049720b", size = 89043771, upload-time = "2026-06-14T04:28:11.009Z" },
-    { url = "https://files.pythonhosted.org/packages/b7/0c/45b8e0b3c8386ae88b82cd65ca249271f9a6978d60f208665228a358830b/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:29c01b026e41afe89db970466c1b6a16e60166e47a76e4c6f88148133c485ab0", size = 90621274, upload-time = "2026-06-14T04:29:18.612Z" },
-    { url = "https://files.pythonhosted.org/packages/9b/70/5d5e8dbb4911e890a98848da23cf624a022e320f93d79b637e50e5ddd0f0/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:d08feb9429bd872022d01fd790a6a9cded57a14f4ea2b2525ed62ee4a32f5f7f", size = 89045465, upload-time = "2026-06-14T04:26:05.71Z" },
-    { url = "https://files.pythonhosted.org/packages/0a/0b/1b2e7b5a1c1f59fed0302fc90f03e3f4b61f81b75607453c6726fe642a36/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:218b0cfeaf8c3184933635032f589c6c190513f3efe0b232b05b69d0b5be5c4b", size = 90624030, upload-time = "2026-06-14T04:30:22.411Z" },
-    { url = "https://files.pythonhosted.org/packages/59/07/5747add860c9747b56c2d11eb2831d6f6d831b27f715bdf33573258b0a6c/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:97683ee6231ec57c4f7bfd2ce53171d7fa3805cf8f307ddd238ed75707446cdb", size = 89046659, upload-time = "2026-06-14T04:27:34.296Z" },
-    { url = "https://files.pythonhosted.org/packages/d1/f2/f23740acf232540576439e2fbb5a231f74739047dbd1db5896cbd2eb6498/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:a33d83905cfca7a2db2034a2f658941024755d75c1b500c93aacbe6164ec0300", size = 90623701, upload-time = "2026-06-14T04:30:55.438Z" },
-    { url = "https://files.pythonhosted.org/packages/49/b0/cbb6474af79fcf94b9c6ad84f6656d6f9603a3ab399781ae5c83a8d8bf86/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:fe292f2352e78a5b93f7501811a035f7d8945fc1986426b0be8cee7a8cbb2e1f", size = 89046517, upload-time = "2026-06-14T04:27:05.881Z" },
-    { url = "https://files.pythonhosted.org/packages/47/f5/be8d6d0bf4a5ba7e6a1825cf72150c27a81e041463b907b393f68693b9a9/nvidia_cutlass_dsl_libs_base-4.6.0.dev0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:19de039bb3496baa806cf22c8a68b989dd998a129847d3d8bb830d34ba777f08", size = 90623353, upload-time = "2026-06-14T04:31:30.134Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/25/8b73e70530040643dd218c5d72b6d0f33372eeeec7f3dd3e71d34ff4870e/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:420ab19a170f1a778e874ff598a49ea91b437f857343009f88e8297be6c27ede", size = 87012605, upload-time = "2026-08-05T14:16:37.363Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/c0/fc5b7738f5dc7c5b5d15ac2c4d045f84e4f7cc4d68469f397bb0d909b4cb/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:32bfc7caa5341076c7000adbb58da2a86b6ecfd8c480a264b36326a8359a7a50", size = 88453632, upload-time = "2026-08-05T14:17:01.159Z" },
+    { url = "https://files.pythonhosted.org/packages/42/26/033408d2f1488e941b3434c21e703ec11c42873166af1eb8348271565782/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:d245dfb3255d95d1ac75063b758acd3e40d5c6699dc98c80f488165260923e11", size = 87011888, upload-time = "2026-08-05T14:18:09.399Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/1b/4bc2842e0b497d9d0fda73976af7640ef742cdb5210e2d143b72c5ec3938/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:4ad5083bf8bcbda83973bdc38a03c2c8452d11fe0afc04b10fc362e23ddd7f8c", size = 88454141, upload-time = "2026-08-05T14:18:31.516Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/b7/019a8f74ad30ad9b6190dd5963d8921fc2b03777316e47260eb822624a0a/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:ea3b96dd6b627fe20637acfa7876ea1adfa8d81a2b106ffff0bb2d042c0c2731", size = 87013242, upload-time = "2026-08-05T14:18:57.703Z" },
+    { url = "https://files.pythonhosted.org/packages/80/6e/70c545a600b52b3f0fbbe01608aefc4e675c924ba886485ad72541f5c88e/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:7f510307369d522da8e7d666557b9b9e7df06b94748bd5dc0198d59dfd0d918a", size = 88454261, upload-time = "2026-08-05T14:21:14.733Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/54/6006a2b4fcd05f32be451fe8881968dcbed526438b4eae9a527534062c79/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:2f3b163060325777a3847954b3a599556b2bff500eca19806a81d1d635bb4149", size = 87012255, upload-time = "2026-08-05T14:21:44.507Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/7a/a08ac0ecdd88bb129a56d95b7c58ea6826e6f5fa00191ef7ecdff8085836/nvidia_cutlass_dsl_libs_cu12-4.6.2-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:84243743382948c83ca627c2dcb106db3844c31d45b8cc2af3d14bc9df0a535e", size = 88453568, upload-time = "2026-08-05T14:22:49.158Z" },
 ]
 
 [[package]]
 name = "nvidia-cutlass-dsl-libs-cu13"
-version = "4.6.0.dev0"
+version = "4.6.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "cuda-python", version = "13.0.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "numpy", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "nvidia-cutlass-dsl-libs-base", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "nvidia-cuda-nvdisasm", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
+    { name = "nvidia-cutlass-dsl-libs-base", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "protobuf", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "typing-extensions", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero')" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/82/5a/9016ee356450d487dc9b64546b727df138ccc81645c49660de5359971ef2/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:4120d2d2ba93c6dabfd193f813687c9a9bef85007512b57ae100ba993347e4dc", size = 88751712, upload-time = "2026-06-14T04:21:27.363Z" },
-    { url = "https://files.pythonhosted.org/packages/58/f0/5f97a440b57e9b1507f2a99754a64e69e47784966590a684b9ffacbaf026/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:67b8f6796fe8e083a97b0fb644a0c7762ce5feadc564dd112c64c5ce734d4241", size = 90211621, upload-time = "2026-06-14T04:23:50.136Z" },
-    { url = "https://files.pythonhosted.org/packages/20/08/9180353817833a96db628d29e1afe6988f30bd8cc379d8012912cbe63623/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:2e0b484347e7b09046edb9bd7f8a425fb2fd35333f051bab36065c00fb57b70f", size = 88753645, upload-time = "2026-06-14T04:19:51.911Z" },
-    { url = "https://files.pythonhosted.org/packages/4e/0a/3941744fa039c98d72f26a5d19f883df8778492fe9e4bd27032b64635e3d/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:9893b9675671fae8fd37d01c01b20843ae59c6eb2d0eeb33f3915645fe953890", size = 90214126, upload-time = "2026-06-14T04:24:57.092Z" },
-    { url = "https://files.pythonhosted.org/packages/7f/89/aabeed4cb39c6dd611c99ec15abd405103ed624083af2874847a608d5c89/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:681c56fdbb5ec7adda3afe3cc6e4d41d7f7e080d17d35b1283bf330fdd922dda", size = 88753497, upload-time = "2026-06-14T04:21:58.966Z" },
-    { url = "https://files.pythonhosted.org/packages/61/e2/ec7e3b8b4a520effd41cabba33381d31c5c129fc542d26404917bfc0094f/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:d45056221b808a719815520fbbfa35798e6917ced7978cf0e601953bcd7cc9aa", size = 90211238, upload-time = "2026-06-14T04:22:36.405Z" },
-    { url = "https://files.pythonhosted.org/packages/50/43/ff318b716f5633c6c0fe9849b9afa4f03493d5738163240cc5e268ffb908/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:f635bb384d00ffaef938cd685aa511567139eb16f34012ace52c8ded153cfc2d", size = 88754108, upload-time = "2026-06-14T04:20:24.284Z" },
-    { url = "https://files.pythonhosted.org/packages/55/54/12c49f5290636feb8a8de0a29caf12a744c1d49ea385eefcfd6e584aaa17/nvidia_cutlass_dsl_libs_cu13-4.6.0.dev0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:3447be9af3c5de8f20d3d8cde325361ba9ee8c55dafad670a02ac6880432495b", size = 90211144, upload-time = "2026-06-14T04:25:33.513Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/73/889e76581ad00467f4d5b5ced211246db1f7473e2aae830550d55d6cc1ab/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:43197a6706d9c08c8b6eebd8841b9366c4f58847b6e159a594ae8912d8d40b77", size = 86713509, upload-time = "2026-08-05T14:25:28.345Z" },
+    { url = "https://files.pythonhosted.org/packages/60/9c/a151ba1fa6590bc1d24baba475e41649009ac3e88c39ffc7fe96986653e8/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:3a5cfc9877b78e3169c04a3071250e8589cf727682270861b9c23b9077a0c3ea", size = 88035526, upload-time = "2026-08-05T14:25:58.253Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/28/5341cc64b3c2c6c697686d64f8e35b3f3fab05055ef69f3380a577f12182/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:3a891048e8227e8c04c09ded2dfbc705bbda5b44a740c409794a67545d86dc1f", size = 86712079, upload-time = "2026-08-05T14:26:43.526Z" },
+    { url = "https://files.pythonhosted.org/packages/14/96/734d780c03b988b0622cd94fec61fecbc83333cd3ed36cede9019fd8fa61/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:e2a3c8f7ba5c98d5e2f6c5231926ceca0cf2c8f7d75dd2e6660394c689b89f7e", size = 88036498, upload-time = "2026-08-05T14:27:05.581Z" },
+    { url = "https://files.pythonhosted.org/packages/46/38/bbe5b8f1ef1a15e7bd1e2ea86686c38f15eeee0a4ba440a5d00d8222cb8e/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:b9543257622ce70d696c6a2bda7e9617ef5b1cc9f29bfe66f4b38d1d1e29a8cb", size = 86712108, upload-time = "2026-08-05T14:27:35.488Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/05/7874c37d50e112177a64a96dd6c7eaeacb1661ebb864e0fe5cfd9f706767/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:a299a623fad75eb752c1a4e447b924e9049ed396b037509c41a48da82d1f340a", size = 88036030, upload-time = "2026-08-05T14:27:58.991Z" },
+    { url = "https://files.pythonhosted.org/packages/33/35/a38b52a8f99d1549c31680deae6a44328f3840e76a1202d5ab0af5b91f58/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:5d215db842c47862232537a6737b0cbf4992318aed24ddb1d4f18c3a43022682", size = 86711155, upload-time = "2026-08-05T14:29:18.118Z" },
+    { url = "https://files.pythonhosted.org/packages/de/b7/75d9470796d4c8fb8e10165825a9cd1e95316bab45af1a6a3dd9555cfe70/nvidia_cutlass_dsl_libs_cu13-4.6.2-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:ad108ba4a6d35c0549fa3860ccbb5a90f005f240d2750ac9f8c3bdaa1f1864cc", size = 88036124, upload-time = "2026-08-05T14:31:27.349Z" },
 ]
 
 [[package]]
@@ -12671,7 +12725,7 @@ wheels = [
 
 [[package]]
 name = "quack-kernels"
-version = "0.5.3"
+version = "0.6.3"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
     "python_full_version >= '3.13' and platform_machine == 'x86_64' and sys_platform == 'linux' and extra != 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero'",
@@ -12751,16 +12805,16 @@ resolution-markers = [
 dependencies = [
     { name = "apache-tvm-ffi", version = "0.1.14.post1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "einops", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
-    { name = "nvidia-cutlass-dsl", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "torch", version = "2.10.0", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and sys_platform == 'darwin' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (sys_platform != 'darwin' and extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (sys_platform != 'darwin' and extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "torch", version = "2.10.0+cu128", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "torch", version = "2.10.0+cu130", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and sys_platform != 'darwin' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-train') or (python_full_version < '3.13' and extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero')" },
     { name = "torch", version = "2.13.0+cu130", source = { registry = "https://download.pytorch.org/whl" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train')" },
     { name = "torch-c-dlpack-ext", marker = "python_full_version >= '3.13' or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/df/d4/979d9afc6008adeba640b146fe4cb3aa61157581b2dad5649104af4079b5/quack_kernels-0.5.3.tar.gz", hash = "sha256:e295c3861ffebca44f6adafff9bf9bc55102a7178b3e1690b2c4cd8f7e0533ff", size = 370274, upload-time = "2026-06-26T03:51:23.488Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/87/6e/589b7e1ac366eaf2f526e350ce9785200f5ef021ad93a92823c5c12cbefc/quack_kernels-0.6.3.tar.gz", hash = "sha256:e307269931e18590f7555afb55debf28e266b4371960aad0528e4055ca8c79a3", size = 839016, upload-time = "2026-08-06T10:28:04.657Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/38/99/a697f7168dbd91d2e7e66df07187a0693cb3d350a58ba8a91271596adfa6/quack_kernels-0.5.3-py3-none-any.whl", hash = "sha256:d2590dbbdfedaf7e36682a0e1bc36a623897e137579664941578b16a2b127da4", size = 338690, upload-time = "2026-06-26T03:51:21.929Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/7c/c16475d19327376f2d5d64ada9f8eff0077fce0b54fd87408b4bb0c04b49/quack_kernels-0.6.3-py3-none-any.whl", hash = "sha256:dfa69468a3f71fb9f7192c087c82e85ad68d59155fa03e408a19c7069fb21b8f", size = 727601, upload-time = "2026-08-06T10:28:03.325Z" },
 ]
 
 [[package]]
@@ -17101,7 +17155,7 @@ dependencies = [
     { name = "numpy" },
     { name = "nvidia-cudnn-frontend" },
     { name = "nvidia-cutlass-dsl", version = "4.4.2", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version < '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "nvidia-cutlass-dsl", version = "4.6.0.dev0", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "nvidia-cutlass-dsl", version = "4.6.2", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "openai" },
     { name = "openai-harmony" },
     { name = "opencv-python-headless" },
@@ -17123,7 +17177,7 @@ dependencies = [
     { name = "pyyaml" },
     { name = "pyzmq" },
     { name = "quack-kernels", version = "0.3.9", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version < '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version < '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra != 'group-16-cosmos-framework-cu130-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-libero') or (extra == 'extra-16-cosmos-framework-train' and extra != 'group-16-cosmos-framework-cu128' and extra != 'group-16-cosmos-framework-cu128-train' and extra != 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-libero')" },
-    { name = "quack-kernels", version = "0.5.3", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
+    { name = "quack-kernels", version = "0.6.3", source = { registry = "https://pypi.org/simple" }, marker = "(python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu128-train') or (python_full_version >= '3.13' and extra == 'group-16-cosmos-framework-cu130') or (python_full_version >= '3.13' and extra != 'group-16-cosmos-framework-cu130-torch213' and extra != 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu128-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu128-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'extra-16-cosmos-framework-train' and extra == 'group-16-cosmos-framework-libero') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-torch213-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213' and extra == 'group-16-cosmos-framework-vllm') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-cu130-train') or (extra == 'group-16-cosmos-framework-cu130-torch213-train' and extra == 'group-16-cosmos-framework-vllm')" },
     { name = "regex" },
     { name = "requests" },
     { name = "sentencepiece" },