Instructions to use FINAL-Bench/Darwin-4B-Chimera with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FINAL-Bench/Darwin-4B-Chimera with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="FINAL-Bench/Darwin-4B-Chimera")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("FINAL-Bench/Darwin-4B-Chimera")
model = AutoModelForCausalLM.from_pretrained("FINAL-Bench/Darwin-4B-Chimera")
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-4B-Chimera with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FINAL-Bench/Darwin-4B-Chimera"
# 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-4B-Chimera",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/FINAL-Bench/Darwin-4B-Chimera

SGLang

How to use FINAL-Bench/Darwin-4B-Chimera 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-4B-Chimera" \
    --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-4B-Chimera",
		"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-4B-Chimera" \
        --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-4B-Chimera",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use FINAL-Bench/Darwin-4B-Chimera with Docker Model Runner:
```
docker model run hf.co/FINAL-Bench/Darwin-4B-Chimera
```

SeaWolf-AI commited on 14 days ago

Commit

0f6879b

verified ·

1 Parent(s): af24be3

Upload folder using huggingface_hub

Browse files

Files changed (2) hide show

README.md +9 -11
model.safetensors +1 -1

README.md CHANGED Viewed

@@ -1,24 +1,22 @@
 ---
 license: gemma
 base_model: FINAL-Bench/Darwin-V9-Chimera-4B-SFT
-tags:
-- rft
-- self-distillation
-- chimera
-- korean
 pipeline_tag: text-generation
 ---
 # Darwin-V9-Chimera-4B-RFT
-RFT self-distillation of Darwin-V9-Chimera-4B-SFT via **learnable-edge** filtering.
 ## Method
-Self-rollout (n=8, temp 0.8) on KMMLU-domain problems, keep correct CoT **only from learnable-edge questions** (1-6 of 8 correct) — excluding already-stable (8/8) and unknown (0/8) items. LoRA r16, merged.
 ## Result (KMMLU 6-subject, 240 held-out, greedy)
-| | Baseline (V9-SFT) | This model | Delta |
-|---|---|---|---|
-| KMMLU | 42.9% | **47.5%** | **+4.6pp** |
-5 of 6 subjects improved (Chemistry/Ecology +12.5 each). Ablation: all-correct data was neutral/harmful; edge-filtering was the key. Small-sample eval (240); to be confirmed on a larger set.

 ---
 license: gemma
 base_model: FINAL-Bench/Darwin-V9-Chimera-4B-SFT
+tags: [rft, self-distillation, star, chimera, korean]
 pipeline_tag: text-generation
 ---
 # Darwin-V9-Chimera-4B-RFT
+RFT self-distillation of Darwin-V9-Chimera-4B-SFT via **learnable-edge** filtering + **STaR** accumulation (2 rounds).
 ## Method
+Self-rollout (n=8, temp 0.8) on KMMLU-domain train problems; keep correct CoT **only from learnable-edge questions** (1-6 of 8 correct), excluding already-stable (8/8) and unknown (0/8). Round 2 (STaR) rolls out from the improved model on fresh problems and accumulates edge data. LoRA r16, LR 2e-5, merged.
 ## Result (KMMLU 6-subject, 240 held-out, greedy)
+| Model | KMMLU | vs base |
+|---|---|---|
+| Baseline (V9-SFT) | 42.9% | - |
+| RFT round 1 (edge 909) | 47.5% | +4.6 |
+| **This (STaR round 2, edge 1679)** | **48.3%** | **+5.4** |
+Ablation: training on ALL correct rollouts was neutral/harmful; **edge-filtering (1-6/8) was the decisive lever** — empirically confirms arXiv 2607.00152. Small held-out (240); confirm on a larger set.

model.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27ae8ac33c5cc1ee62e53c1379b685e87b9dcfdf35f42f104f269321f5e290f2
 size 8044982080

 version https://git-lfs.github.com/spec/v1
+oid sha256:f1de5b2dfd9149e4646da71ad40e0f4e7a4c622c17f7766e10ac48223fbf9fbd
 size 8044982080