Instructions to use ce-amtic/ProcVLM-2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ce-amtic/ProcVLM-2B with Transformers:

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

pipe = pipeline("image-text-to-text", model="ce-amtic/ProcVLM-2B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("ce-amtic/ProcVLM-2B")
model = AutoModelForImageTextToText.from_pretrained("ce-amtic/ProcVLM-2B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ce-amtic/ProcVLM-2B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ce-amtic/ProcVLM-2B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ce-amtic/ProcVLM-2B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/ce-amtic/ProcVLM-2B

SGLang

How to use ce-amtic/ProcVLM-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 "ce-amtic/ProcVLM-2B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ce-amtic/ProcVLM-2B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "ce-amtic/ProcVLM-2B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ce-amtic/ProcVLM-2B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use ce-amtic/ProcVLM-2B with Docker Model Runner:
```
docker model run hf.co/ce-amtic/ProcVLM-2B
```

ce-amtic commited on 19 days ago

Commit

ef9ad41

verified ·

1 Parent(s): bb1445d

Update README.md

Browse files

Files changed (1) hide show

README.md +20 -12

README.md CHANGED Viewed

@@ -19,7 +19,7 @@ base_model:
 # ProcVLM-2B
-ProcVLM-2B is a procedure-grounded vision-language model for estimating progress rewards from robot manipulation observations. Given a task description and a recent window of video frames, the model reasons about the remaining atomic actions and predicts the current task completion percentage in a `<progress>...</progress>` tag.
 <p align="center">
   <a href="https://procvlm.github.io/">Homepage</a> |
@@ -42,7 +42,6 @@ ProcVLM-2B is designed for research on robot learning, progress reward modeling,
 - estimating task completion progress from robot videos;
 - producing dense progress rewards from sparse demonstrations;
-- visualizing progress over time for manipulation rollouts;
 - adapting progress prediction to a new environment with one-shot LoRA fine-tuning.
 This model is not intended to be used as a safety-critical controller without downstream validation.
@@ -122,8 +121,17 @@ Given the recent observation and the task "{task}", first infer the remaining at
 The model should answer with reasoning and a final progress tag, for example:
 ```text
-The drawer is already open and the bread is close to the drawer. The remaining action is to place the bread inside the drawer.
-Therefore, the estimated progress is <progress>62.5%</progress>.
 ```
 ## vLLM Batch Inference
@@ -177,14 +185,14 @@ ProcVLM can be adapted to a new environment with one successful task demonstrati
 If you use ProcVLM, please cite the paper:
 ```bibtex
-@misc{procvlm2026,
-  title         = {ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic Manipulation},
-  author        = {ProcVLM Authors},
-  year          = {2026},
-  eprint        = {2605.08774},
-  archivePrefix = {arXiv},
-  primaryClass  = {cs.RO},
-  url           = {https://arxiv.org/abs/2605.08774}
 }
 ```

 # ProcVLM-2B
+ProcVLM-2B is a procedure-grounded vision-language model for estimating progress rewards from robot manipulation observations. Given a task description and a recent window of video frames, the model reasons about the remaining atomic actions and predicts the current task completion percentage.
 <p align="center">
   <a href="https://procvlm.github.io/">Homepage</a> |
 - estimating task completion progress from robot videos;
 - producing dense progress rewards from sparse demonstrations;
 - adapting progress prediction to a new environment with one-shot LoRA fine-tuning.
 This model is not intended to be used as a safety-critical controller without downstream validation.
 The model should answer with reasoning and a final progress tag, for example:
 ```text
+To complete the task: Tower the blocks, the following steps are required:
+1. Grasp the green block.
+2. Place the green block onto the red block.
+Therefore, the estimated progress percentage is <progress>84.13%</progress>.
+```
+Or if the task is finished:
+```text
+The task requires: Tower the blocks. Images show no block outside the tower, no further steps required.
+Therefore, the estimated progress percentage is <progress>100.00%</progress>.
 ```
 ## vLLM Batch Inference
 If you use ProcVLM, please cite the paper:
 ```bibtex
+@misc{feng2026procvlmlearningproceduregroundedprogress,
+      title={ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic Manipulation},
+      author={Youhe Feng and Hansen Shi and Haoyang Li and Xinlei Guo and Yang Wang and Chengyang Zhang and Jinkai Zhang and Xiaohan Zhang and Jie Tang and Jing Zhang},
+      year={2026},
+      eprint={2605.08774},
+      archivePrefix={arXiv},
+      primaryClass={cs.RO},
+      url={https://arxiv.org/abs/2605.08774},
 }
 ```