[Misc] Update Phi-3-Vision Example (#5981)

Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
Roger Wang 2024-06-28 23:34:29 -07:00 committed by GitHub
parent 580353da93
commit 329df38f1a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -10,8 +10,10 @@ from vllm.multimodal.image import ImagePixelData
def run_phi3v(): def run_phi3v():
model_path = "microsoft/Phi-3-vision-128k-instruct" model_path = "microsoft/Phi-3-vision-128k-instruct"
# Note: The model has 128k context length by default which may cause OOM # Note: The default setting of max_num_seqs (256) and
# In this example, we override max_model_len to 2048. # max_model_len (128k) for this model may cause OOM.
# In this example, we override max_num_seqs to 5 while
# keeping the original context length of 128k.
llm = LLM( llm = LLM(
model=model_path, model=model_path,
trust_remote_code=True, trust_remote_code=True,
@ -19,7 +21,7 @@ def run_phi3v():
image_token_id=32044, image_token_id=32044,
image_input_shape="1,3,1008,1344", image_input_shape="1,3,1008,1344",
image_feature_size=1921, image_feature_size=1921,
max_model_len=2048, max_num_seqs=5,
) )
image = Image.open("images/cherry_blossom.jpg") image = Image.open("images/cherry_blossom.jpg")