tacodevs/Behemoth-X-R1-123B-FP8

<div align="center"> <img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/hero.png" alt="Behemoth-X-R1-123B" style="width:100%; max-width:960px; border-radius:16px; box-shadow:0 0 60px rgba(236,72,153,0.35), 0 0 100px rgba(139,92,246,0.25);"/> </div>

<div align="center" style="margin-top:24px;">

<h1 style="font-size:3em; font-weight:900; background:linear-gradient(90deg,#ec4899 0%,#a855f7 50%,#06b6d4 100%); -webkit-background-clip:text; -webkit-text-fill-color:transparent; background-clip:text; margin:0;">Behemoth-X-R1-123B · FP8</h1>

<p style="font-size:1.2em; color:#a855f7; font-style:italic;">Single-GPU beast mode.</p>

<p> <img src="https://img.shields.io/badge/quant-FP8Dynamic-8B5CF6?style=for-the-badge" alt="quant"/> <img src="https://img.shields.io/badge/VRAM-~130GB-EC4899?style=for-the-badge" alt="vram"/> <img src="https://img.shields.io/badge/runson-1x_H200-06B6D4?style=for-the-badge" alt="gpu"/> </p>

</div>

About

FP8 dynamic quantization of `tacodevs/Behemoth-X-R1-123B`. Near-lossless quality, half the weight bytes, fits on a single H200.

• —Method: W8A8 dynamic quantization via llm-compressor
• —Format: compressed-tensors
• —Size: ~115 GB
• —Calibration: None needed (dynamic scheme)

Usage with vLLM

python -m vllm.entrypoints.openai.api_server \
  --model tacodevs/Behemoth-X-R1-123B-FP8 \
  --max-model-len 16384 \
  --gpu-memory-utilization 0.95 \
  --trust-remote-code

Fits on 1× H200 141GB with ~30k context window.

See the main model card

Full documentation, prompt format, prefill examples, credits, and everything else is on the source repo:

👉 tacodevs/Behemoth-X-R1-123B

License

Inherited from base: [Mistral Research License](https://mistral.ai/licenses/MRL-0.1.md) — non-commercial use only.