Instructions to use FINAL-Bench/Darwin-397B-ZTC with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FINAL-Bench/Darwin-397B-ZTC with Transformers:

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

pipe = pipeline("text-generation", model="FINAL-Bench/Darwin-397B-ZTC")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("FINAL-Bench/Darwin-397B-ZTC")
model = AutoModelForMultimodalLM.from_pretrained("FINAL-Bench/Darwin-397B-ZTC", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use FINAL-Bench/Darwin-397B-ZTC with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FINAL-Bench/Darwin-397B-ZTC"
# 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-397B-ZTC",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/FINAL-Bench/Darwin-397B-ZTC

SGLang

How to use FINAL-Bench/Darwin-397B-ZTC 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-397B-ZTC" \
    --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-397B-ZTC",
		"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-397B-ZTC" \
        --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-397B-ZTC",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use FINAL-Bench/Darwin-397B-ZTC with Docker Model Runner:
```
docker model run hf.co/FINAL-Bench/Darwin-397B-ZTC
```

Darwin-397B-ZTC

File size: 2,142 Bytes

6f06b4d

# Zero-Token Confidence (ZTC) — usage
#
# ZTC reads the model's own internal state ONCE, before generation, and returns
# the probability that the answer the model is about to produce will be correct.
# No extra tokens are generated. No second model is required.
#
#   probe file : ztc/ztc_probe_darwin397b.npz   (45 KB)
#   input      : final-layer hidden state of the last prompt token (4096-dim)
#   output     : score, and a calibrated probability in [0, 1]
#
# Reported performance on this model (PubMedQA, 539 items, 146 incorrect):
#   self-reported confidence  AUROC 0.7646
#   ZTC                       AUROC 0.8801      (permutation null z = 13.31)

import numpy as np
import torch
from transformers import AutoModel, AutoTokenizer

MODEL = "FINAL-Bench/Darwin-397B-ZTC"
PROBE = "ztc/ztc_probe_darwin397b.npz"


class ZTC:
    def __init__(self, path=PROBE):
        z = np.load(path)
        self.w = z["w"].astype(np.float32)
        self.mu = z["mu"].astype(np.float32)
        self.sd = z["sd"].astype(np.float32)
        self.s_mean = float(z["s_mean"]); self.s_std = float(z["s_std"])
        self.A = float(z["cal_A"]); self.B = float(z["cal_B"])

    def score(self, hidden):
        """hidden: (4096,) or (batch, 4096) final-layer state of the last prompt token."""
        h = np.asarray(hidden, dtype=np.float32)
        s = ((h - self.mu) / self.sd) @ self.w
        p = 1.0 / (1.0 + np.exp(-(self.A * (s - self.s_mean) / self.s_std + self.B)))
        return s, p


# --- one forward pass, zero generated tokens -------------------------------
tok = AutoTokenizer.from_pretrained(MODEL)
model = AutoModel.from_pretrained(MODEL, dtype=torch.bfloat16, device_map="auto").eval()
ztc = ZTC()

prompt = "Question: ...\nAnswer:"
b = tok(prompt, return_tensors="pt").to(next(model.parameters()).device)
with torch.no_grad():
    h = model(**b).last_hidden_state[0, -1].float().cpu().numpy()

s, p = ztc.score(h)
print("ZTC score %.3f  ->  P(correct) = %.3f" % (s, p))

# Gate the action, not the answer:
#   if p < THRESHOLD:  do not call the tool / escalate / answer "I don't know"
#   else:              generate as usual