Instructions to use Efficient-Large-Model/VILA15-3b-hf-preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Efficient-Large-Model/VILA15-3b-hf-preview with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Efficient-Large-Model/VILA15-3b-hf-preview", trust_remote_code=True)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Efficient-Large-Model/VILA15-3b-hf-preview", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Efficient-Large-Model/VILA15-3b-hf-preview with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Efficient-Large-Model/VILA15-3b-hf-preview"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Efficient-Large-Model/VILA15-3b-hf-preview",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Efficient-Large-Model/VILA15-3b-hf-preview

SGLang

How to use Efficient-Large-Model/VILA15-3b-hf-preview with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Efficient-Large-Model/VILA15-3b-hf-preview" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Efficient-Large-Model/VILA15-3b-hf-preview",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Efficient-Large-Model/VILA15-3b-hf-preview" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Efficient-Large-Model/VILA15-3b-hf-preview",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Efficient-Large-Model/VILA15-3b-hf-preview with Docker Model Runner:
```
docker model run hf.co/Efficient-Large-Model/VILA15-3b-hf-preview
```

Ligeng-Zhu commited on Feb 25, 2025

Commit

2e75984

verified ·

1 Parent(s): c5f696d

Upload files with `vila-upload`.

Browse files

Upload mm_utils.py
Upload siglip_encoder.py

Files changed (2) hide show

mm_utils.py +1 -1
siglip_encoder.py +6 -2

mm_utils.py CHANGED Viewed

@@ -26,7 +26,7 @@ import torch
 from PIL import Image
 from transformers import StoppingCriteria
-from llava.constants import DEFAULT_IMAGE_TOKEN
 def get_frame_from_vcap(vidcap, num_frames=10, max_fps=0.0, fps=None, frame_count=None, video_file_name=None):

 from PIL import Image
 from transformers import StoppingCriteria
+from .constants import DEFAULT_IMAGE_TOKEN
 def get_frame_from_vcap(vidcap, num_frames=10, max_fps=0.0, fps=None, frame_count=None, video_file_name=None):

siglip_encoder.py CHANGED Viewed

@@ -19,12 +19,16 @@ import torch.nn as nn
 import torch.nn.functional as F
 from accelerate.hooks import add_hook_to_module
 from einops import rearrange
-from s2wrapper import forward as multiscale_forward
 from transformers import AutoConfig, PretrainedConfig, PreTrainedModel, SiglipImageProcessor
 from transformers.image_processing_utils import BaseImageProcessor
-from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
 from transformers.models.siglip import SiglipVisionModel
 class VisionTower(nn.Module):
     def __init__(self, vision_tower, args, delay_load=False):

 import torch.nn.functional as F
 from accelerate.hooks import add_hook_to_module
 from einops import rearrange
 from transformers import AutoConfig, PretrainedConfig, PreTrainedModel, SiglipImageProcessor
 from transformers.image_processing_utils import BaseImageProcessor
 from transformers.models.siglip import SiglipVisionModel
+from s2wrapper import forward as multiscale_forward
+# from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
+def is_deepspeed_zero3_enabled():
+    return False
 class VisionTower(nn.Module):
     def __init__(self, vision_tower, args, delay_load=False):