Skip to content

fix missing parameter in config for qwen #365

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
sdpython merged 2 commits into from
Dec 17, 2025
Merged

fix missing parameter in config for qwen #365

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d1cf66f
fix config for qwen
xadupre Dec 17, 2025
b771e94
fixes
xadupre Dec 17, 2025
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
4 changes: 4 additions & 0 deletions onnx_diagnostic/ci_models/export_qwen25_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -280,6 +280,10 @@ def main(
).eval()
data = dict(model=model)
config = model.config
if not hasattr(config, "bos_token_id") or not config.bos_token_id:
config.bos_token_id = 151643
if not hasattr(config, "eos_token_id") or not config.eos_token_id:
config.eos_token_id = 151645
else:
print("-- random model")
data = get_untrained_model(model_id, second_input=second_input, verbose=1)
Expand Down
4 changes: 1 addition & 3 deletions onnx_diagnostic/helpers/log_helper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1921,9 +1921,7 @@ def first_err(df: pandas.DataFrame) -> pandas.Series:
return lambdas[formula]

if formula == "onnx_n_nodes_no_cst":
return lambda df: gdf(df, "onnx_n_nodes", 0) - gdf(
df, "op_onnx__Constant", 0
).fillna(0)
return lambda df: gdf(df, "onnx_n_nodes", 0) - gdf(df, "op_onnx__Constant", 0)
if formula == "peak_gpu_torch":
return lambda df: gdf(df, "mema_gpu_5_after_export") - gdf(df, "mema_gpu_4_reset")
if formula == "peak_gpu_nvidia":
Expand Down
13 changes: 12 additions & 1 deletion onnx_diagnostic/torch_export_patches/patches/_patch_transformers_qwen2_5.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -256,10 +256,21 @@ def qwen_sdpa_attention(
return attn_output

def qwen_version_selector(opset: int, *args: torch.Tensor) -> Tuple[str, torch.dtype]:
import onnx_ir

first_float_tensor = next(
a
for a in args
if a is not None and a.dtype in {torch.float16, torch.float32, torch.bfloat16}
if a is not None
and a.dtype
in {
torch.float16,
torch.float32,
torch.bfloat16,
onnx_ir.DataType.BFLOAT16,
onnx_ir.DataType.FLOAT16,
onnx_ir.DataType.FLOAT,
}
)
dtype = first_float_tensor.dtype
strategy = patched_Qwen2_5_VLVisionAttention.STRATEGY_FOR_ATTENTION()
Expand Down
Loading