biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Improve Mistral format checks. (#33253)

Browse Source 
Signed-off-by: Julien Denize <julien.denize@mistral.ai>
Signed-off-by: juliendenize <julien.denize@mistral.ai>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Julien Denize
2026-01-30 15:23:33 +01:00
committed by GitHub
parent a11bc12d53
commit ae5b7aff2b
8 changed files with 193 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/config/model.py
View File

@@ -565,6 +565,7 @@ class ModelConfig:
self.dtype,
is_pooling_model=self.runner_type == "pooling",
revision=self.revision,
config_format=self.config_format,
)
self.original_max_model_len = self.max_model_len
@@ -1844,9 +1845,10 @@ def _get_and_verify_dtype(
*,
is_pooling_model: bool,
revision: str | None = None,
config_format: ConfigFormat = "hf",
) -> torch.dtype:
config_dtype = ModelArchConfigConvertorBase.get_torch_dtype(
config, model_id, revision=revision
config, model_id, revision=revision, config_format=config_format
)
model_type = config.model_type