Instructions to use Shangy/browsecomp-perception-e2e-rl with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Shangy/browsecomp-perception-e2e-rl with Transformers:

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

pipe = pipeline("text-generation", model="Shangy/browsecomp-perception-e2e-rl")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Shangy/browsecomp-perception-e2e-rl", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Shangy/browsecomp-perception-e2e-rl with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Shangy/browsecomp-perception-e2e-rl"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Shangy/browsecomp-perception-e2e-rl",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Shangy/browsecomp-perception-e2e-rl

SGLang

How to use Shangy/browsecomp-perception-e2e-rl 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 "Shangy/browsecomp-perception-e2e-rl" \
    --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": "Shangy/browsecomp-perception-e2e-rl",
		"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 "Shangy/browsecomp-perception-e2e-rl" \
        --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": "Shangy/browsecomp-perception-e2e-rl",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Shangy/browsecomp-perception-e2e-rl with Docker Model Runner:
```
docker model run hf.co/Shangy/browsecomp-perception-e2e-rl
```

BrowseComp perception e2e RL

Private archival release of the Qwen3-8B perception-surrogate checkpoint from the BrowseComp collaboration campaign.

The small model compresses long search observations for a frozen large-model agent. It was trained with complete-episode reward and faithfulness penalties, without length shaping. Revision iter149 is the recommended campaign endpoint: held-out accuracy was 0.780 versus a 0.717 zero-shot perception baseline, while the output-cap rate fell from 82% at iter9 to 60% at iter149.

Provenance

Revision: iter149
Source export: perc-e2e-rl-gb300-iter149
W&B project/run: perception-rl/qtizfliz
Experiment documentation: docs/experiments/perception-collab-gb300-campaign.md
Emergency source snapshot: ys-2020/miles@5ed731544
Archived on 2026-08-21 before cluster checkout

The reported score requires the matching perception harness and is not a standalone language-model benchmark.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for Shangy/browsecomp-perception-e2e-rl

Base model

Qwen/Qwen3-8B-Base

Finetuned

Qwen/Qwen3-8B

Finetuned

(1973)

this model