Instructions to use ewin-reg/MiniCPM5-Vision-2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ewin-reg/MiniCPM5-Vision-2B with Transformers:

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

pipe = pipeline("image-text-to-text", model="ewin-reg/MiniCPM5-Vision-2B", trust_remote_code=True)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("ewin-reg/MiniCPM5-Vision-2B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ewin-reg/MiniCPM5-Vision-2B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ewin-reg/MiniCPM5-Vision-2B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ewin-reg/MiniCPM5-Vision-2B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ewin-reg/MiniCPM5-Vision-2B

SGLang

How to use ewin-reg/MiniCPM5-Vision-2B 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 "ewin-reg/MiniCPM5-Vision-2B" \
    --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": "ewin-reg/MiniCPM5-Vision-2B",
		"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 "ewin-reg/MiniCPM5-Vision-2B" \
        --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": "ewin-reg/MiniCPM5-Vision-2B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ewin-reg/MiniCPM5-Vision-2B with Docker Model Runner:
```
docker model run hf.co/ewin-reg/MiniCPM5-Vision-2B
```

ewin-reg commited on 3 days ago

Commit

f004bb0

verified ·

1 Parent(s): e1cc78b

fix: set vision_config.image_size=384 and interpolate_pos_encoding=True

Browse files

Files changed (1) hide show

modeling_minicpm5_v.py +4 -3

modeling_minicpm5_v.py CHANGED Viewed

@@ -82,7 +82,7 @@ class MiniCPM5VConfig(PretrainedConfig):
         if vision_config is None:
             vision_config = {
                 "hidden_size": 1152,
-                "image_size": 448,
                 "intermediate_size": 4304,
                 "num_attention_heads": 16,
                 "num_hidden_layers": 27,
@@ -189,13 +189,14 @@ class MiniCPM5VForConditionalGeneration(MiniCPM5VPreTrainedModel):
         self.llm.set_output_embeddings(new_embeddings)
     def encode_vision(self, pixel_values: torch.Tensor) -> torch.Tensor:
-        vision_outputs = self.vpm(pixel_values=pixel_values)
         hidden_states = vision_outputs.last_hidden_state
-        img_size = self.config.vision_config.image_size
         patch_size = self.config.vision_config.patch_size
         grid_dim = img_size // patch_size
         return self.resampler(hidden_states, grid_h=grid_dim, grid_w=grid_dim)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

         if vision_config is None:
             vision_config = {
                 "hidden_size": 1152,
+                "image_size": 384,
                 "intermediate_size": 4304,
                 "num_attention_heads": 16,
                 "num_hidden_layers": 27,
         self.llm.set_output_embeddings(new_embeddings)
     def encode_vision(self, pixel_values: torch.Tensor) -> torch.Tensor:
+        vision_outputs = self.vpm(pixel_values=pixel_values, interpolate_pos_encoding=True)
         hidden_states = vision_outputs.last_hidden_state
+        img_size = pixel_values.shape[-1]
         patch_size = self.config.vision_config.patch_size
         grid_dim = img_size // patch_size
         return self.resampler(hidden_states, grid_h=grid_dim, grid_w=grid_dim)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,