Mistral-7B Model Evolution Experiment (Alternative task set)
Collection
21 items • Updated
How to use from
SGLangUse 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 "yedi-hu/Model_Kinship_1-3" \
--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": "yedi-hu/Model_Kinship_1-3",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'Quick Links
This model is a checkpoint produced during iterative model merging experiments.
It is not intended as a final release but rather as an intermediate artifact that captures
the progression of merging steps. Researchers and practitioners can use it to study
the effects of incremental merging strategies, compare intermediate states, or build
upon this checkpoint in further experiments.
For more details, please refer to the paper: arXiv:2410.12613.
- Downloads last month
- 5
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "yedi-hu/Model_Kinship_1-3" \ --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": "yedi-hu/Model_Kinship_1-3", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'