A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 23:15 UTC
[Doc] Add note about context length in Phi-3-Vision example (#5887)
Cyrus Leung committed
Jun 27, 2024 at 14:20 UTC
6eabc6cb0e022a3f57fbf115ce53dc8bcba66840
1 file changed
+3
-1
examples/phi3v_example.py
+3
-1
@@ -9,6 +9,9 @@ from vllm.multimodal.image import ImagePixelData
9
10
def run_phi3v():
11
model_path = "microsoft/Phi-3-vision-128k-instruct"
12
+
13
+ # Note: The model has 128k context length by default which may cause OOM
14
+ # If that's the case, override `max_model_len` with a smaller value via args
15
llm = LLM(
16
model=model_path,
17
trust_remote_code=True,
@@ -16,7 +19,6 @@ def run_phi3v():
19
image_token_id=32044,
20
image_input_shape="1,3,1008,1344",
21
image_feature_size=1921,
19
- disable_image_processor=False,
22
)
23
24
image = Image.open("images/cherry_blossom.jpg")