How to use from
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 "usvsnsp/pythia-70m-ppo" \
    --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": "usvsnsp/pythia-70m-ppo",
		"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 "usvsnsp/pythia-70m-ppo" \
        --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": "usvsnsp/pythia-70m-ppo",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'
Quick Links

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Wandb Run: https://wandb.ai/eleutherai/pythia-rlhf/runs/gy2g8jj1

Model Evals:

Tasks Version Filter Metric Value Stderr
arc_challenge Yaml none acc 0.2253 ± 0.0122
none acc_norm 0.2278 ± 0.0123
arc_easy Yaml none acc 0.2551 ± 0.0089
none acc_norm 0.2567 ± 0.0090
lambada_openai Yaml none perplexity NaN ± NaN
none acc 0.0016 ± 0.0005
logiqa Yaml none acc 0.2028 ± 0.0158
none acc_norm 0.2028 ± 0.0158
piqa Yaml none acc 0.4946 ± 0.0117
none acc_norm 0.4924 ± 0.0117
sciq Yaml none acc 0.0140 ± 0.0037
none acc_norm 0.0140 ± 0.0037
winogrande Yaml none acc 0.5036 ± 0.0141
wsc Yaml none acc 0.6346 ± 0.0474
Downloads last month
9
Safetensors
Model size
96.1M params
Tensor type
F32
·
BOOL
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support