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 "palette-lab/songgot-m" \
--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": "palette-lab/songgot-m",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'Songgot (์ก๊ณณ)
A Korean-first tiny agentic model for tool calling on the device, trained from scratch by Hanish Keloth (Palette). Apache 2.0.
- Paper: https://hanishkeloth.github.io/songgot
- Code, data generators, scorer: https://github.com/hanishkeloth/songgot
- Demo: https://huggingface.co/spaces/Hanish/songgot
Try it on the device: https://hanishkeloth.github.io/songgot/app/ (runs in the browser, works offline after the first load).
Numbers
Kakao FunctionChat-Bench SingleCall (500 Korean items, 5 tool conditions), exact match on function name and arguments, scorer in the repo. Comparators run with identical tools and queries in their own documented formats.
| model | params | exact | 4_random | 4_close | 8_random | 8_close | all | name only |
|---|---|---|---|---|---|---|---|---|
| Songgot-M (2 epochs, v8 set) | 126M | 45.0 | 43.0 | 26.0 | 34.0 | 18.0 | 33.2 | 73.8 |
| Songgot (6B tokens, 3 epochs, v8 set) | 50M | 44.0 | 39.0 | 30.0 | 35.0 | 17.0 | 33.0 | 73.6 |
| Songgot-nano (1 epoch) | 39M | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| Needle 2 | 45M | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| FunctionGemma-270M | 270M | 3.0 | 5.0 | 1.0 | 1.0 | 1.0 | 2.2 | 36.2 |
| Qwen3-0.6B | 600M | 48.0 | 49.0 | 45.0 | 37.0 | 37.0 | 43.2 | 70.8 |
| Qwen3.5-0.8B | 800M | 51.0 | 48.0 | 41.0 | 52.0 | 34.0 | 45.2 | 73.6 |
Tokens per Hangul syllable on the same 100 queries: Songgot 0.90, Gemma 3 0.98, Qwen3 1.15, Needle 2 3.47.
Status (2026-09-11 05:18)
Weights in this repo are Songgot-M, 2 epochs, v8 set: 16 layers, hidden 768, about 126M parameters, pretrained on 8xH100 (Modal) on 6B tokens, post-trained on the v8 set (five teacher-synthesis rounds, the last with confusable sibling tools), post-trained on the v2 tool-calling set. Call accuracy on FunctionChat-Bench SingleCall 33.2 percent (name only 73.8). GGUF exports (f16, Q8_0, Q4_K_M) are in this repo.
Format
<|system|>
[{"name": "set_alarm", "description": "์๋์ ์ค์ ํฉ๋๋ค.", "parameters": {...}}]
<|user|>
๋ด์ผ ์์นจ 7์์ ์๋ ๋ง์ถฐ์ค
<|call|>
{"name":"set_alarm","arguments":{"time":"07:00"}}<|end|>
Tokenizer: SentencePiece BPE, 32k, byte fallback (tokenizer.model). Use sentencepiece directly; the special tokens live inside the vocabulary.
Data and provenance
fineweb-edu sample-10BT (ODC-By), Korean Wikipedia 20231101.ko (CC BY-SA 3.0; this model card carries the attribution and share-alike notice for that text), glaive-function-calling-v2 (Apache 2.0), template-generated Korean tool calls (Apache 2.0, in the repo). No closed-model outputs. FunctionChat-Bench was never used for training.
Limits
Single-call tool selection and argument extraction only. No multi-turn, no tool results, no free chat. Small models are finicky with rare tools and paraphrased values; validate every call in application code.
- Downloads last month
- 547
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "palette-lab/songgot-m" \ --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": "palette-lab/songgot-m", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'