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

[LoRA][1/N]Remove LoRA extra vocab (#28382)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li
2025-11-12 03:06:21 +08:00
committed by GitHub
parent 8c32c6e4b4
commit 9d1c474704
65 changed files with 197 additions and 754 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
vllm/model_executor/models/commandr.py
View File

@@ -288,17 +288,12 @@ class CohereModel(nn.Module):
config = vllm_config.model_config.hf_config
cache_config = vllm_config.cache_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config
self.quant_config = quant_config
self.config = config
lora_vocab = (
(lora_config.lora_extra_vocab_size * (lora_config.max_loras or 1))
if lora_config
else 0
)
self.vocab_size = config.vocab_size + lora_vocab
self.org_vocab_size = config.vocab_size
self.vocab_size = config.vocab_size
self.embed_tokens = VocabParallelEmbedding(
config.vocab_size, config.hidden_size
)
@@ -424,17 +419,15 @@ class CohereForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsQuant):
super().__init__()
config = vllm_config.model_config.hf_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config
self.config = config
# currently all existing command R models have `tie_word_embeddings`
# enabled
assert config.tie_word_embeddings
self.unpadded_vocab_size = config.vocab_size
if lora_config:
self.unpadded_vocab_size += lora_config.lora_extra_vocab_size
self.quant_config = quant_config
self.logits_processor = LogitsProcessor(
self.unpadded_vocab_size, config.vocab_size, scale=config.logit_scale
config.vocab_size, scale=config.logit_scale
)
self.model = CohereModel(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")