README.md

metadata

title: Darwin-35B-A3B-Opus API
emoji: 🧬
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
short_description: OpenAI-compatible FastAPI for Darwin-35B-A3B-Opus (INT4)

Darwin-35B-A3B-Opus API

Self-hosted OpenAI-compatible FastAPI server for FINAL-Bench/Darwin-35B-A3B-Opus.

• 35B MoE / 3B active — Qwen3.5-MoE based
• INT4 quantized (~18 GB) — fits on L4/A10G/L40S
• OpenAI-compatible endpoints + SSE streaming
• Bearer auth (configurable via API_KEYS secret)

Endpoints

• GET / — Landing page with examples
• GET /health — Health + load status
• GET /v1/models — List models
• POST /v1/chat/completions — Chat (OpenAI compat)

Configuration (HF Space secrets)

Secret	Required	Description
HF_TOKEN	optional	HF token for private/gated models
API_KEYS	optional	Comma-separated bearer keys (empty = public)
QUANT_MODE	optional	int4 (default), int8, bf16
MODEL_ID	optional	HF model id (default: FINAL-Bench/Darwin-35B-A3B-Opus)

Hardware

Recommended:

• L4 (24GB) — INT4 ✅
• A10G-small (24GB) — INT4 ✅
• L40S (48GB) — INT4 ✅ or INT8 ✅
• A100 (80GB) — any mode including BF16

Example

from openai import OpenAI
client = OpenAI(
    api_key="YOUR_KEY",
    base_url="https://final-bench-darwin-35b-a3b-opus-api.hf.space/v1",
)
resp = client.chat.completions.create(
    model="Darwin-35B-A3B-Opus",
    messages=[{"role":"user","content":"Explain GPQA"}],
    max_tokens=300,
)
print(resp.choices[0].message.content)