mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-05-20 19:07:00 +08:00
[Bugfix] Update starcoder2 to remap k/v scale names for kv_cache quantization (#11148)
This commit is contained in:
parent
1efce68605
commit
3989a79824
@ -37,7 +37,8 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
|
|||||||
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
|
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
|
||||||
from vllm.model_executor.layers.vocab_parallel_embedding import (
|
from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||||
DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
|
DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
|
||||||
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
|
from vllm.model_executor.model_loader.weight_utils import (
|
||||||
|
default_weight_loader, maybe_remap_kv_scale_name)
|
||||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||||
from vllm.sequence import IntermediateTensors
|
from vllm.sequence import IntermediateTensors
|
||||||
|
|
||||||
@ -345,6 +346,10 @@ class Starcoder2ForCausalLM(nn.Module, SupportsPP):
|
|||||||
weight_loader(param, loaded_weight, shard_id)
|
weight_loader(param, loaded_weight, shard_id)
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
|
name = maybe_remap_kv_scale_name(name, params_dict)
|
||||||
|
if name is None:
|
||||||
|
continue
|
||||||
|
|
||||||
if self.config.tie_word_embeddings and "lm_head.weight" in name:
|
if self.config.tie_word_embeddings and "lm_head.weight" in name:
|
||||||
continue
|
continue
|
||||||
if is_pp_missing_parameter(name, self):
|
if is_pp_missing_parameter(name, self):
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user