20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Bugfix] Fix LoRA extra vocab size (#15047)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li 2025-03-19 00:40:29 +08:00 committed by GitHub
parent 179a619c21
commit 46c759c165
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 1 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
examples/offline_inference/audio_language.py
View File

@ -93,7 +93,6 @@ def run_phi4mm(question: str, audio_count: int) -> ModelRequestData:
max_num_seqs=2, max_num_seqs=2,
enable_lora=True, enable_lora=True,
max_lora_rank=320, max_lora_rank=320,
lora_extra_vocab_size=0,
limit_mm_per_prompt={"audio": audio_count}, limit_mm_per_prompt={"audio": audio_count},
) )

1
examples/offline_inference/vision_language.py
View File

@ -682,7 +682,6 @@ def run_phi4mm(questions: list[str], modality: str) -> ModelRequestData:
max_num_seqs=2, max_num_seqs=2,
enable_lora=True, enable_lora=True,
max_lora_rank=320, max_lora_rank=320,
lora_extra_vocab_size=0,
) )
return ModelRequestData( return ModelRequestData(

1
examples/offline_inference/vision_language_multi_image.py
View File

@ -342,7 +342,6 @@ def load_phi4mm(question: str, image_urls: list[str]) -> ModelRequestData:
limit_mm_per_prompt={"image": len(image_urls)}, limit_mm_per_prompt={"image": len(image_urls)},
enable_lora=True, enable_lora=True,
max_lora_rank=320, max_lora_rank=320,
lora_extra_vocab_size=0,
) )
placeholders = "".join(f"<|image_{i}|>" placeholders = "".join(f"<|image_{i}|>"

1
tests/models/decoder_only/vision_language/test_phi4mm.py
View File

@ -100,7 +100,6 @@ def run_test(
distributed_executor_backend=distributed_executor_backend, distributed_executor_backend=distributed_executor_backend,
enable_lora=True, enable_lora=True,
max_lora_rank=320, max_lora_rank=320,
lora_extra_vocab_size=0,
gpu_memory_utilization=0.8, # set to 0.8 to avoid OOM in CI gpu_memory_utilization=0.8, # set to 0.8 to avoid OOM in CI
enforce_eager=True, enforce_eager=True,
) as vllm_model: ) as vllm_model:

2
vllm/config.py
View File

@ -2324,7 +2324,7 @@ class LoRAConfig:
# Setting the maximum rank to 512 should be able to satisfy the vast # Setting the maximum rank to 512 should be able to satisfy the vast
# majority of applications. # majority of applications.
possible_max_ranks = (8, 16, 32, 64, 128, 256, 320, 512) possible_max_ranks = (8, 16, 32, 64, 128, 256, 320, 512)
possible_lora_extra_vocab_size = (0, 256, 512) possible_lora_extra_vocab_size = (256, 512)
if self.max_lora_rank not in possible_max_ranks: if self.max_lora_rank not in possible_max_ranks:
raise ValueError( raise ValueError(
f"max_lora_rank ({self.max_lora_rank}) must be one of " f"max_lora_rank ({self.max_lora_rank}) must be one of "