Skip to content

convert : allow vision target for DFlash/DSpark #29339

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

Merged
ngxson merged 1 commit into from
Sep 23, 2026
+2 −15
Merged

convert : allow vision target for DFlash/DSpark #29339

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
17 changes: 2 additions & 15 deletions conversion/qwen.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@
if TYPE_CHECKING:
from torch import Tensor

from .base import LazyTorchTensor, ModelBase, TextModel, gguf, logger
from .base import LazyTorchTensor, ModelBase, ModelType, TextModel, get_model_architecture, gguf, logger


@ModelBase.register("QWenLMHeadModel")
Expand Down Expand Up @@ -666,7 +666,7 @@ def set_vocab(self):
from . import get_model_class
with open(self.target_model_dir / "config.json", "r", encoding="utf-8") as f:
target_hparams = json.load(f)
target_arch = target_hparams["architectures"][0]
target_arch = get_model_architecture(target_hparams, ModelType.TEXT)
target_cls = get_model_class(target_arch)

if target_cls is not type(self):
Expand Down Expand Up @@ -841,13 +841,6 @@ def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Ca
return None
return super().filter_tensors(item)

_ROPE_PERMUTE_SUFFIXES = (
"self_attn.q_proj.weight",
"self_attn.k_proj.weight",
"self_attn.q_norm.weight",
"self_attn.k_norm.weight",
)

def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
if name == "model.d2t":
self._d2t = data_torch
Expand All @@ -856,12 +849,6 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
if self._n_vocab_draft == self.hparams["vocab_size"] and name.endswith("lm_head.weight"):
return

# interleaved-rope checkpoints (rope_is_neox_style = false) -> NeoX layout: per head, even dims first then odd
if not self.hparams.get("rope_is_neox_style", True) and name.endswith(self._ROPE_PERMUTE_SUFFIXES):
head_dim = self.hparams["head_dim"]
shape = data_torch.shape
data_torch = data_torch.reshape(-1, head_dim // 2, 2, *shape[1:]).transpose(1, 2).reshape(shape)

yield from super().modify_tensors(data_torch, name, bid)

def prepare_tensors(self):
Expand Down
Loading