Kangheng commited on
Commit
2ddcc49
·
verified ·
1 Parent(s): fc4a171

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +0 -40
README.md CHANGED
@@ -168,46 +168,6 @@ decoded = processor.decode(generate_ids[0, inputs["input_ids"].shape[-1] :], ski
168
  print(decoded)
169
  ```
170
 
171
- ### Requirements
172
-
173
- To run STEP3-VL-10B efficiently, we recommend setting up a Python environment (>=3.10) with **vLLM**:
174
-
175
- ```bash
176
- pip install vllm>=0.6.3
177
- ```
178
-
179
- ### vLLM Inference Example
180
-
181
- Below is a minimal example to load the model and generate a response using vLLM's chat API.
182
-
183
- ```python
184
- from vllm import LLM, SamplingParams
185
-
186
- # 1. Load the model
187
- # Ensure you have ~24GB VRAM for BF16 inference
188
- llm = LLM(
189
- model="stepfun-ai/Step3-VL-10B",
190
- trust_remote_code=True,
191
- gpu_memory_utilization=0.95
192
- )
193
-
194
- # 2. Prepare input (Supports local paths or URLs)
195
- messages = [
196
- {
197
- "role": "user",
198
- "content": [
199
- {"type": "image", "image": "[https://modelscope.oss-cn-beijing.aliyuncs.com/resource/demo.jpg](https://modelscope.oss-cn-beijing.aliyuncs.com/resource/demo.jpg)"},
200
- {"type": "text", "text": "Describe this image in detail."}
201
- ]
202
- }
203
- ]
204
-
205
- # 3. Generate
206
- sampling_params = SamplingParams(temperature=0.1, max_tokens=1024)
207
- outputs = llm.chat(messages=messages, sampling_params=sampling_params)
208
-
209
- print(f"Output: {outputs[0].outputs[0].text}")
210
- ```
211
 
212
  ## 📜 Citation
213
 
 
168
  print(decoded)
169
  ```
170
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
171
 
172
  ## 📜 Citation
173