research-awq

A guide to AWQ, a method for compressing large language-model weights to 4-bit values while preserving important weights. It covers when to use AWQ, installation options, supported hardware, and alternatives such as GPTQ or bitsandbytes.

In plain words
What is it for?
Preparing 7B–70B language models for deployment, installing AutoAWQ, loading pre-quantized models, and selecting between AWQ and other quantization methods.
Why use it?
It helps developers choose and set up a model-compression method when GPU memory or inference speed is a constraint.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/graycodeai/starling/research-awq
Any agent
npx skills add GrayCodeAI/starling --skill research-awq
Clone the repo
git clone --depth 1 https://github.com/GrayCodeAI/starling

Made for: Claude Code, Codex.

Per session 51 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,433 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin 86% copy Near-identical to another mod in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00051 $0.02433
Opus 5 $0.00026 $0.01216
Sonnet 5 $0.00010 $0.00487
Haiku 4.5 $0.00005 $0.00243

Measured 2d ago against content hash 650fa78a5cd8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

research-awq scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

Origin

This is a copy

86% identical to awq-quantization — 9 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.

categories/ai-ml/research-awq/SKILL.md · 310 lines

How it starts

The opening of the file, as written. The whole thing — 310 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AWQ (Activation-aware Weight Quantization)

4-bit quantization that preserves salient weights based on activation patterns, achieving 3x speedup with minimal accuracy loss.

When to use AWQ

Use AWQ when:

  • Need 4-bit quantization with <5% accuracy loss
  • Deploying instruction-tuned or chat models (AWQ generalizes better)
  • Want ~2.5-3x inference speedup over FP16
  • Using vLLM for production serving
  • Have Ampere+ GPUs (A100, H100, RTX 40xx) for Marlin kernel support

Use GPTQ instead when:

  • Need maximum ecosystem compatibility (more tools support GPTQ)
  • Working with ExLlamaV2 backend specifically
  • Have older GPUs without Marlin support

Use bitsandbytes instead when:

  • Need zero calibration overhead (quantize on-the-fly)
  • Want to fine-tune with QLoRA
  • Prefer simpler integration

Quick start

Installation

# Default (Triton kernels)
pip install autoawq

# With optimized CUDA kernels + Flash Attention
pip install autoawq[kernels]

# Intel CPU/XPU optimization
pip install autoawq[cpu]

Requirements: Python 3.8+, CUDA 11.8+, Compute Capability 7.5+

Load pre-quantized model

from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_name = "TheBloke/Mistral-7B-Instruct-v0.2-AWQ"

model = AutoAWQForCausalLM.from_quantized(
    model_name,
    fuse_layers=True  # Enable fused attention for speed
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Generate
inputs = tokenizer("Explain quantum computing", return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Quantize your own model

from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_path = "mistralai/Mistral-7B-Instruct-v0.2"

# Load model and tokenizer
model = AutoAWQForCausalLM.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Quantization config
quant_config = {
    "zero_point": True,      # Use zero-point quantization
    "q_group_size": 128,     # Group size (128 recommended)
    "w_bit": 4,              # 4-bit weights
    "version": "GEMM"        # GEMM for batch, GEMV for single-token
}

# Quantize (uses pileval dataset by default)
model.quantize(tokenizer, quant_config=quant_config)

# Save
model.save_quantized("mistral-7b-awq")
tokenizer.save_pretrained("mistral-7b-awq")

Read the full file on GitHub · 310 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 310 lines · 51 tokens per session scan A 650fa78a5cd8

Subscribe to this mod's changes

research-awq is a skill published in the GitHub repository GrayCodeAI/starling (2 stars, last pushed 3d ago), licensed MIT. It adds 51 tokens to every session and 2,433 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. It is 86% identical to awq-quantization, differing in 9 lines, and is treated as a copy.

Related

Other skills, from other repositories

awq-quantization

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper…

Orchestra-Research/AI-Research-SKILLs · 79 tokens

awq-quantization

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper…

ihatesea69/HieuNghi-AI-Skills · 79 tokens

gptq

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Orchestra-Research/AI-Research-SKILLs · 84 tokens

hqq-quantization

Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.

Orchestra-Research/AI-Research-SKILLs · 58 tokens

quantizing-models-bitsandbytes

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.

Orchestra-Research/AI-Research-SKILLs · 83 tokens

gguf-quantization

GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.

Orchestra-Research/AI-Research-SKILLs · 48 tokens