AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
by MIT / MIT-IBM Watson AI Lab · open-source · Last verified 2026-03-17
Introduced AWQ (Activation-aware Weight Quantization), a hardware-friendly low-bit weight quantization approach that protects a small fraction (1%) of salient weights based on activation magnitudes, achieving better performance than GPTQ at 4-bit while being faster and more broadly applicable across model architectures.
https://arxiv.org/abs/2306.00978 ↗B+
B+—Good
Adoption: AQuality: A+Freshness: ACitations: B+Engagement: F
Specifications
- License
- MIT
- Pricing
- open-source
- Capabilities
- weight-quantization, activation-aware-compression, 4-bit-inference, hardware-efficient
- Integrations
- autoawq, huggingface, vllm, llm-compressor
- Use Cases
- model-compression, edge-deployment, efficient-llm-serving
- API Available
- No
- Tags
- awq, quantization, activation-aware, weight-quantization, efficiency
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
73.3Adoption
88
Quality
93
Freshness
81
Citations
78
Engagement
0
Put AI to work for your business
Deploy this paper alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.