Instructions to use tsinghua-sigs-robot-lab/VeriLoop-E2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsinghua-sigs-robot-lab/VeriLoop-E2 with Transformers:

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

pipe = pipeline("text-generation", model="tsinghua-sigs-robot-lab/VeriLoop-E2")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("tsinghua-sigs-robot-lab/VeriLoop-E2")
model = AutoModelForMultimodalLM.from_pretrained("tsinghua-sigs-robot-lab/VeriLoop-E2", device_map="auto")
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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tsinghua-sigs-robot-lab/VeriLoop-E2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsinghua-sigs-robot-lab/VeriLoop-E2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsinghua-sigs-robot-lab/VeriLoop-E2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsinghua-sigs-robot-lab/VeriLoop-E2

SGLang

How to use tsinghua-sigs-robot-lab/VeriLoop-E2 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 "tsinghua-sigs-robot-lab/VeriLoop-E2" \
    --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": "tsinghua-sigs-robot-lab/VeriLoop-E2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "tsinghua-sigs-robot-lab/VeriLoop-E2" \
        --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": "tsinghua-sigs-robot-lab/VeriLoop-E2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsinghua-sigs-robot-lab/VeriLoop-E2 with Docker Model Runner:
```
docker model run hf.co/tsinghua-sigs-robot-lab/VeriLoop-E2
```

ConorWang commited on 28 days ago

Commit

b06bb80

verified ·

1 Parent(s): adfff8c

Upload README.md

Browse files

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -131,7 +131,7 @@ The README reports the **frozen release scores** for VeriLoop E2. Agentic benchm
 | **GPQA Diamond** | **[93.9](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond)** | GPT-5.6 Sol 94.1 | Claude Fable 5 92.6 | Kimi K3 93.5 | GLM-5.2 Max 91.2 | — | — |
 | **Apex 2025** | **[89.6](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025)** | GPT-5.5 80.0 | Claude Opus 4.8 81.0 | Kimi K3 66.0 | — | DeepSeek V4 Pro 28.0 | Gemini 3.1 Pro 61.0 |
-A dash means that the release figure does not include a public comparison point for that provider on that benchmark. Each linked VeriLoop E2 score above resolves directly to its benchmark-specific public evidence directory. External reference values mirror the frozen comparison set used in Figure 1; harness notes and protocol caveats are retained in the evaluation ledger rather than duplicated here. **SWE-Marathon v1.1 (45.0)** remains reported on the model page, but is intentionally not represented by a Hugging Face Native Benchmark `.eval_results` entry because a stable HF benchmark registration/task identifier is not currently available.
 ### Evaluation evidence
@@ -162,7 +162,7 @@ integrity metadata and provenance
 | GPQA Diamond | [https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond) |
 | Apex 2025 | [https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025) |
-The structured Hugging Face evaluation descriptors for these eight registered benchmarks are published in the model repository under [`/.eval_results/`](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2/tree/main/.eval_results).
 ---

 | **GPQA Diamond** | **[93.9](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond)** | GPT-5.6 Sol 94.1 | Claude Fable 5 92.6 | Kimi K3 93.5 | GLM-5.2 Max 91.2 | — | — |
 | **Apex 2025** | **[89.6](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025)** | GPT-5.5 80.0 | Claude Opus 4.8 81.0 | Kimi K3 66.0 | — | DeepSeek V4 Pro 28.0 | Gemini 3.1 Pro 61.0 |
+A dash means that the release figure does not include a public comparison point for that provider on that benchmark. Each linked VeriLoop E2 score above resolves directly to its benchmark-specific public evidence directory. External reference values mirror the frozen comparison set used in Figure 1; harness notes and protocol caveats are retained in the evaluation ledger rather than duplicated here. **Seven results currently map to Hugging Face Native Benchmark leaderboards:** AIME 2026, DeepSWE v1.1, GPQA Diamond, SWE-bench Pro, Terminal-Bench 2.1, Terminal-Bench 3.0, and Terminal-Bench 4.0. **Apex 2025 (89.6)** is retained as a structured model result with public evidence, but `MathArena/apex_2025` is currently a dataset used by the MathArena leaderboard rather than a Hugging Face Native Benchmark leaderboard, so no Hugging Face rank is expected from that entry. **SWE-Marathon v1.1 (45.0)** remains reported on the model page and is intentionally not represented by a Hugging Face Native Benchmark `.eval_results` entry because a stable Hugging Face benchmark registration/task identifier is not currently available.
 ### Evaluation evidence
 | GPQA Diamond | [https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond) |
 | Apex 2025 | [https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025) |
+The model repository publishes **eight structured evaluation descriptors** under [`/.eval_results/`](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2/tree/main/.eval_results). Seven of them currently map to Hugging Face Native Benchmark leaderboards; the Apex 2025 descriptor is retained for structured reporting and provenance, but does not currently produce a Hugging Face leaderboard rank.
 ---