Use Transformers v5 WeightRenaming for Transformers modeling backend (#31545)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-03-13 20:49:08 +00:00
committed by GitHub
parent d0b402974f
commit 0005d2a3c9
7 changed files with 162 additions and 89 deletions
+12 -12
View File
@@ -995,19 +995,10 @@ class SupportsQuant:
def __new__(cls, *args, **kwargs) -> Self:
instance = super().__new__(cls)
# find config passed in arguments
quant_config = cls._find_quant_config(*args, **kwargs)
if quant_config is not None:
# attach config to model for general use
instance.quant_config = quant_config
# find config passed in arguments and attach it to model for general use
instance.quant_config = cls._find_quant_config(*args, **kwargs)
# apply model mappings to config for proper config-model matching
if (hf_to_vllm_mapper := instance.hf_to_vllm_mapper) is not None:
instance.quant_config.apply_vllm_mapper(hf_to_vllm_mapper)
if instance.packed_modules_mapping is not None:
instance.quant_config.packed_modules_mapping.update(
instance.packed_modules_mapping
)
cls._maybe_apply_model_mapping(instance)
return instance
@@ -1026,6 +1017,15 @@ class SupportsQuant:
return None
def _maybe_apply_model_mapping(self):
"""Apply model mappings to config for proper config-model matching"""
if self.quant_config is None:
return
if (hf_to_vllm_mapper := self.hf_to_vllm_mapper) is not None:
self.quant_config.apply_vllm_mapper(hf_to_vllm_mapper)
if self.packed_modules_mapping is not None:
self.quant_config.packed_modules_mapping.update(self.packed_modules_mapping)
@runtime_checkable
class SupportsRealtime(Protocol):