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 "123sg/ResearchModel-Rivulet-s25" \
    --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": "123sg/ResearchModel-Rivulet-s25",
		"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 "123sg/ResearchModel-Rivulet-s25" \
        --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": "123sg/ResearchModel-Rivulet-s25",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'
Quick Links

ResearchModel-Rivulet

release overview

Release scope

This repository records the selected checkpoint from the rivulet-release-candidate training run. The package includes its configuration, weights, tokenizer metadata, training state, and release figures.

Evaluation matrix

Evaluation Score
Math Reasoning 0.704
Logical Reasoning 0.850
Common Sense 0.815
Reading Comprehension 0.802
Question Answering 0.692
Text Classification 0.850
Sentiment Analysis 0.834
Code Generation 0.786
Creative Writing 0.768
Dialogue Generation 0.751
Summarization 0.827
Translation 0.838
Knowledge Retrieval 0.756
Instruction Following 0.823
Safety Evaluation 0.810

Scores are reported on a zero-to-one scale and rounded to three decimal places.

Usage

Load the configuration and weights with a compatible Transformers runtime. Review trainer_state.json for the selected training step and tokenizer_config.json for tokenizer settings.

benchmark profile

Release notes

The checkpoint was selected by the workspace evaluation runner using its aggregate eval_accuracy result. No other checkpoint files belong in this repository.

training summary

Downloads last month
91
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support