Instructions to use FINAL-Bench/Darwin-28B-NEG with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use FINAL-Bench/Darwin-28B-NEG with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="FINAL-Bench/Darwin-28B-NEG") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("FINAL-Bench/Darwin-28B-NEG") model = AutoModelForCausalLM.from_pretrained("FINAL-Bench/Darwin-28B-NEG", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use FINAL-Bench/Darwin-28B-NEG with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "FINAL-Bench/Darwin-28B-NEG" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Darwin-28B-NEG", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/FINAL-Bench/Darwin-28B-NEG
- SGLang
How to use FINAL-Bench/Darwin-28B-NEG 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 "FINAL-Bench/Darwin-28B-NEG" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Darwin-28B-NEG", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'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 "FINAL-Bench/Darwin-28B-NEG" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "FINAL-Bench/Darwin-28B-NEG", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use FINAL-Bench/Darwin-28B-NEG with Docker Model Runner:
docker model run hf.co/FINAL-Bench/Darwin-28B-NEG
โ ๏ธ DEPRECATED โ FINAL-Bench/Darwin-28B-Opus ๊ถ์ฅ
์ด ๋ชจ๋ธ์ ์ฌ์ฉ์ ๊ถ์ฅํ์ง ์์ต๋๋ค. ๋์:
FINAL-Bench/Darwin-28B-Opus๋ฅผ ์ฌ์ฉํ์ธ์.
๐จ DEPRECATION NOTICE (2026-04-24)
์ด ๋ชจ๋ธ์ Darwin V8 NEG(Native Entropy Gating) ๊ธฐ์ ์ 28B ์ค์ผ์ผ ์ ์ฉ ์คํ์ ์ฐ์ถ๋ฌผ์ ๋๋ค. ๋ด๋ถ ํ๊ฐ ๊ฒฐ๊ณผ, NEG๋ 27B ์ด์ ๋ํ ๋ชจ๋ธ์์ ์ฑ๋ฅ ํฅ์์ด ์๊ฑฐ๋ ์คํ๋ ค ์ญํจ๊ณผ๋ฅผ ๋ด๋ ๊ฒ์ผ๋ก ํ์ธ๋์ด ๊ณต์ ๋ฐฐํฌ๋ณธ์์ ์ ์ธ๋์์ต๋๋ค.
๐ ์ธก์ ๊ฒฐ๊ณผ (GPQA Diamond, 198 ๋ฌธ์ , Greedy, max_tokens 5010)
| ๋ชจ๋ธ | ์ ํ๋ | ฮ vs Darwin-28B-Opus |
|---|---|---|
| Darwin-28B-Opus (๊ถ์ฅ) | 148/198 = 74.75% | ๊ธฐ์ค๊ฐ |
| Darwin-28B-NEG (์ด ๋ชจ๋ธ, 9B-tuned) | 147/198 = 74.24% | โ0.51 pp |
| Darwin-28B-NEG (logit-shift penalty=2) | 14/24 = 58.33% | โ20.84 pp |
๐ฌ ๊ทผ๋ณธ ์์ธ
- Top-K ๊ฒ์ดํธ๋ Greedy ๋์ฝ๋ฉ์์ ์ํ์ ์ผ๋ก ๋ฌดํจ: argmax๋ ํญ์ top-k์ ํฌํจ๋๋ฏ๋ก top-k ๋ง์คํน์ด argmax๋ฅผ ๋ฐ๊พธ์ง ๋ชปํจ.
- ๋ํ ๋ชจ๋ธ์ ์ด๋ฏธ ์ ์บ๋ฆฌ๋ธ๋ ์ด์ ๋จ: Darwin-28B-Opus๋ ๋ถํ์คํ ์๊ฐ์๋ top-1์ด ๋๋ถ๋ถ ์ ๋ต. NEG์ "๋ถํ์คํ ๋ ๊ฐ์ " ์ฒ ํ์ด ์คํ๋ ค ์ ๋ต์ ๋ง์นจ.
- Entropy ๋ถํฌ: 28B์ ํ๊ท entropy๋ 0.268๋ก 9B ๋๋น ๋งค์ฐ ๋ฎ์ NEG ๊ฐ์ ์ฌ์ง ์์ฒด๊ฐ ์ ์.
โ ๊ณต์ NEG ์ ์ฉ ์ค์ผ์ผ ๊ท์น
โค 9B โ NEG ์ ์ฉ (๊ฒ์ฆ: Darwin-9B-NEG์์ +15%p ๊ฐ์ )
9B~27B โ ์คํ์ (์ถ๊ฐ ๊ฒ์ฆ ํ์)
โฅ 27B โ NEG ๊ธ์ง (28B์์ ์ญํจ๊ณผ ๊ฒ์ฆ๋จ)
๐ ๊ถ์ฅ ์ฌ์ฉ
Darwin 28B ์ถ๋ก ๋ชจ๋ธ์ ์ฐพ์ผ์ ๋ค๋ฉด ์๋๋ฅผ ์ฌ์ฉํ์ธ์:
- FINAL-Bench/Darwin-28B-Opus โ ๋์ผ base ๊ฐ์ค์น, NEG ์์, ๊ณต์ 74.75% GPQA Diamond
- Anserwise/AWAXIS-Think-28B โ Darwin-28B-Opus + ํ๊ตญ์ด LoRA ํนํ
9B๊ธ NEG ๋ชจ๋ธ์ ์ฐพ์ผ์ ๋ค๋ฉด:
- FINAL-Bench/Darwin-9B-NEG โ NEG ์ ์ฉ ์ ํจ (+15%p ๊ฒ์ฆ)
๐ ์ด ๋ชจ๋ธ์ ๊ทธ๋๋ก ๋๋์?
์ด ์ ์ฅ์๋ ์ฐ๊ตฌ ๊ธฐ๋ก ๋ฐ reproducibility ๋ชฉ์ ์ผ๋ก private ์ํ๋ก ๋ณด์กด๋ฉ๋๋ค. ๊ธฐ์ ๋ณด๊ณ ์ reference_darwin_v8_neg.md์ ์์ธ ์คํจ ๋ถ์์ด ๊ธฐ๋ก๋์ด ์์ต๋๋ค.
๐ ๋ผ์ด์ ์ค
Apache 2.0
Deprecated: 2026-04-24 Maintainer: FINAL-Bench
- Downloads last month
- 1
Model tree for FINAL-Bench/Darwin-28B-NEG
Base model
FINAL-Bench/Darwin-28B-Opus