This repository provides quantized runtime packages of
TheDrummer/Behemoth-ReduX-123B-v1, packaged for vLLM using the compressed-tensors format.

TL;DR

This repo is quantized with multiple branches: W4A16-ASYM (AWQ W4A16 asymmetric) and W8A16 (INT8 weights / INT16 activations).

Load with vLLM using --quantization compressed-tensors.

Typical W4A16 recipe: group_size=128, keep lm_head in higher precision; uses the parent finetune’s chat template.

Revisions & Branches

The main branch is a placeholder landing branch (model card + links). All runnable artifacts live under per-revision branches.

main — placeholder / landing page
W4A16 — SYMMETRICAL - AWQ 4‑bit weights / 16‑bit activations builds and related assets (Will use Marlin Kernel in VLLM)
W4A16-ASYM — AWQ 4‑bit weights / 16‑bit activations builds and related assets
W8A16 — 8‑bit weights / 16‑bit activations builds

Quick links:

🔗 main
🔗 W4A16
🔗 W4A16-ASYM
🔗 W8A16

What’s in this repo (per revision)

Sharded quantized weights in .safetensors with an index (model.safetensors.index.json)
config.json including compressed-tensors metadata (e.g., weight_format, quantization, quantization_config)
Tokenizer artifacts (tokenizer.json, tokenizer.model, etc.)
Optional: chat_template.jinja (inherits the parent finetune’s chat format)

Exact files can differ by branch; see the Files and versions tab for each revision.

Quickstart — vLLM

Install vLLM (recent version recommended):

pip install vllm

Serve (adjust to your hardware):

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 vllm serve TheHouseOfTheDude/Behemoth-ReduX-123B-v1_Compressed-Tensors   --quantization compressed-tensors   --tensor-parallel-size 8   --max-model-len 32768   --gpu-memory-utilization 0.70   --dtype bfloat16

Query via Chat Completions:

bash

curl http://localhost:8000/v1/chat/completions   -H "Content-Type: application/json"   -d '{
    "model": "TheHouseOfTheDude/Behemoth-ReduX-123B-v1_Compressed-Tensors",
    "messages": [
      {"role":"system","content":"You are Behemoth-ReduX, helpful, precise, and safe."},
      {"role":"user","content":"Outline a retrieval pipeline for scientific PDFs."}
    ],
    "max_tokens": 512,
    "temperature": 0.7,
    "top_p": 0.95
  }'

Note: compressed-tensors is a vLLM runtime format. Loading this artifact directly in vanilla 🤗 Transformers is not supported; use vLLM for inference. If you need Transformers inference, use a different export (e.g., GPTQ/AWQ compatible with Transformers) or full-precision weights.

Prompting / Chat Template

This package follows the parent finetune’s chat format. If a chat_template.jinja is present in the branch, apply_chat_template will use it automatically.

Lineage

Finetuned parent: TheDrummer/Behemoth-ReduX-123B-v1
This repo: Quantized child of the finetune (compressed-tensors for vLLM)

Hardware & Tips (rule‑of‑thumb)

123B‑class models strongly prefer multi‑GPU deployments (e.g., 8× high‑VRAM).
Long contexts are KV‑cache heavy—tune --max-model-len and batch size.
Prefer BF16 on GPUs with native support; otherwise FP16.
Consider CUDA Graphs if stable in your stack.

License & Usage

This distribution inherits the licenses/policies of the finetuned parent model.
Use of the model constitutes acceptance of the upstream terms.

Changelog

v1 (current) — Quantized compressed‑tensors exports for Behemoth‑ReduX‑123B‑v1; added W4A16‑ASYM and W8A16 revision branches; model card set for Quantized classification.