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

[Model] Refactor Phi-4-multimodal to use merged processor and support V1 (#15477)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Isotr0py
2025-04-19 17:26:11 +08:00
committed by GitHub
parent d9737ca1c6
commit 83f3c3bd91
15 changed files with 818 additions and 1246 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/offline_inference/audio_language.py
View File

@@ -89,7 +89,7 @@ def run_phi4mm(question: str, audio_count: int) -> ModelRequestData:
engine_args = EngineArgs(
model=model_path,
trust_remote_code=True,
max_model_len=4096,
max_model_len=12800,
max_num_seqs=2,
enable_lora=True,
max_lora_rank=320,

5
examples/offline_inference/vision_language.py
View File

@@ -814,10 +814,13 @@ def run_phi4mm(questions: list[str], modality: str) -> ModelRequestData:
engine_args = EngineArgs(
model=model_path,
trust_remote_code=True,
max_model_len=4096,
max_model_len=5120,
max_num_seqs=2,
max_num_batched_tokens=12800,
enable_lora=True,
max_lora_rank=320,
# Note - mm_processor_kwargs can also be passed to generate/chat calls
mm_processor_kwargs={"dynamic_hd": 16},
limit_mm_per_prompt={"image": 1},
)

4
examples/offline_inference/vision_language_multi_image.py
View File

@@ -503,11 +503,13 @@ def load_phi4mm(question: str, image_urls: list[str]) -> ModelRequestData:
engine_args = EngineArgs(
model=model_path,
trust_remote_code=True,
max_model_len=10000,
max_model_len=4096,
max_num_seqs=2,
limit_mm_per_prompt={"image": len(image_urls)},
enable_lora=True,
max_lora_rank=320,
# Note - mm_processor_kwargs can also be passed to generate/chat calls
mm_processor_kwargs={"dynamic_hd": 4},
)
placeholders = "".join(f"<|image_{i}|>"