Skip to content

[Klaud Cold] Update DeepSeek-V4-Pro FP4 GB300 Dynamo-TRT AgentX MTP image to TensorRT-LLM 1.4.0rc1 / [Klaud Cold] 将 DeepSeek-V4-Pro FP4 GB300 Dynamo-TRT AgentX MTP 镜像更新至 TensorRT-LLM 1.4.0rc1 #4583

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
Klaud-Cold wants to merge 13 commits into from
Closed

[Klaud Cold] Update DeepSeek-V4-Pro FP4 GB300 Dynamo-TRT AgentX MTP image to TensorRT-LLM 1.4.0rc1 / [Klaud Cold] 将 DeepSeek-V4-Pro FP4 GB300 Dynamo-TRT AgentX MTP 镜像更新至 TensorRT-LLM 1.4.0rc1 #4583

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
a0cecc2
perf(dsv4): update GB300 Dynamo-TRT AgentX image to TensorRT-LLM 1.4.…
github-actions[bot] Oct 27, 2026
1fd1906
chore(perf-changelog): record DSv4 GB300 Dynamo-TRT AgentX TensorRT-L…
github-actions[bot] Oct 27, 2026
df494af
fix(dsv4): omit --publish-metrics for the pinned Dynamo fork on GB300…
github-actions[bot] Oct 27, 2026
af17588
Merge remote-tracking branch 'origin/main' into klaud/auto-abf776a96f…
github-actions[bot] Oct 27, 2026
d99b874
fix(dsv4): pin upstream Dynamo v1.6.1 for GB300 Dynamo-TRT AgentX
github-actions[bot] Oct 27, 2026
50e93b2
Merge remote-tracking branch 'origin/main' into klaud/auto-abf776a96f…
github-actions[bot] Oct 27, 2026
7abc13f
fix(dsv4): install Dynamo v1.6.1 from PyPI wheels for GB300 Dynamo-TR…
github-actions[bot] Oct 27, 2026
68cb251
fix(dsv4): disable SWA scratch reuse on GB300 Dynamo-TRT AgentX prefi…
github-actions[bot] Oct 27, 2026
aa17dda
Merge remote-tracking branch 'origin/main' into klaud/auto-abf776a96f…
github-actions[bot] Oct 27, 2026
66b9743
fix(dsv4): publish TRT-LLM metrics on GB300 Dynamo-TRT AgentX with Dy…
github-actions[bot] Oct 28, 2026
c310271
Merge remote-tracking branch 'origin/main' into klaud/auto-abf776a96f…
github-actions[bot] Oct 28, 2026
ab723eb
Merge remote-tracking branch 'origin/main' into klaud/auto-abf776a96f…
github-actions[bot] Oct 28, 2026
01fbe12
Merge main into klaud/auto-abf776a96f2b5e25-5f20d20eaa35dce7 [skip-sw…
github-actions[bot] Oct 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions ...benchmarks/multi_node/srt-slurm-recipes/dsv4/trtllm/gb300-fp4/agentx/disagg-variants.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,12 +6,14 @@ base:
model: {}
container: {}
frameworks:
tensorrt_llm: 1.3.0rc24
tensorrt_llm: 1.4.0rc1
dynamo:
install: true
source:
git: https://github.com/cquil11/dynamo.git
rev: 2cbbdc863c336c2cfffa2a3668e2a6661801b259
# Upstream v1.6.1: resolves interface-name DYN_TCP_RPC_HOST (ai-dynamo/dynamo#11870)
# and decouples TRT-LLM KV events from metrics (ai-dynamo/dynamo#12162).
# Prebuilt wheels: a source build's cudarc 0.19.8 rejects the image's CUDA 13.4.
pypi: "1.6.1"
request_plane: tcp
health_check:
max_attempts: 1440
Expand Down Expand Up @@ -56,6 +58,9 @@ base:
event_buffer_max_size: 0
free_gpu_memory_fraction: 0.8
host_cache_size: 193273528320
# 1.4.0rc1 always enforces the SWA scratch resize check (debug-only in 1.3.0rc24),
# which c388 prefill tripped; DeepSeek-V4 model defaults turn scratch reuse on.
enable_swa_scratch_reuse: false
Comment on lines +61 to +63

@claude claude Bot Oct 28, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 nit (optional): The new recipe comments record history, which AGENTS.md "Comments" forbids. Examples are "(debug-only in 1.3.0rc24)", "which c388 prefill tripped" and "Upstream v1.6.1: resolves ..." at lines 13-14. Fix: cut each block to one short technical line about why the value is set. For example: "# TRT-LLM 1.4 enforces the SWA scratch resize check; DeepSeek-V4 defaults enable reuse." Leave the version history and past results to git and perf-changelog.yaml.

Why this was flagged

AGENTS.md:57 says comments must not record history, meaning how code evolved, what it replaced, or past results. AGENTS.md:58 allows one short technical line per line or block. The comment added at disagg-variants.yaml:61-62 compares behaviour with the old 1.3.0rc24 image and names a past run failure (c388 prefill). The comment at disagg-variants.yaml:13-15 lists upstream PRs that fixed problems in the fork it replaces. Each of these blocks is three lines long. The base did not have these comments. The result is a convention slip, not a runtime fault.

Verification: nit. This is a repository-rule violation and nothing breaks at runtime. AGENTS.md:57 says: "No history: how code evolved, what it replaced, where it came from, or past results." At disagg-variants.yaml:61-62, the parenthetical "(debug-only in 1.3.0rc24)" describes the image this PR replaces, so it is history.

pool_ratio:
- 0.55
- 0.22
Expand Down
2 changes: 1 addition & 1 deletion inferencex-e2e/configs/nvidia-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6591,7 +6591,7 @@ kimik3-fp4-gb200-dynamo-vllm-agentic-mooncake-dcp16-agg:
dp-attn: false

dsv4-fp4-gb300-dynamo-trt-agentx:
image: nvcr.io#nvidia/tensorrt-llm/release:1.3.0rc24
image: nvcr.io/nvidia/tensorrt-llm/release:1.4.0rc1@sha256:3bedbe0321e7eb5430f3177d88894a3cf2dc2f6a087d0e85c13fc882667c75c0
model: deepseek-ai/DeepSeek-V4-Pro
model-prefix: dsv4
runner: cluster:gb300-nv
Expand Down
7 changes: 7 additions & 0 deletions inferencex-e2e/perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11556,3 +11556,10 @@
- "Disagg: 6 DEP2 prefill workers with a DEP8 decode at conc 280-360 or a DEP16 decode at conc 300; DeepGEMM MegaMoE, NIXL transfer plus a shared MooncakeStore DRAM pool (160 GB per rank), Dynamo KV router."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4468
scenario-type: [agentic-coding]

- config-keys:
- dsv4-fp4-gb300-dynamo-trt-agentx
description:
- "Update the TensorRT-LLM image from 1.3.0rc24 to 1.4.0rc1 and the recipe Dynamo pin to v1.6.1."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4583
scenario-type: [agentic-coding]
Loading