Research map / Architectures and efficiency / Quantization
AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs
WenXiang Lin, Juntao Huang, LuHan Zhang, Laiyi Li, Xiang Bao, mengyang zhang, Bing Wang, Shaohuai Shi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use AGoQ instead of Megatron-LM on LLaMA2-13B at 80K sequence | LLaMA2-13B · seq 80K, 64 GPUs | iteration time (ms): 149667 -> 111422 | Improved |
| use AGoQ instead of Megatron-LM and ZeRO-1 at 32K | LLaMA2-13B · seq 32K, 64 GPUs | iteration time (ms): 37635 -> 36568 | Improved |
| use AGoQ instead of COAT (24K) | OLMo-1B · seq 24k | memory (MB): 94100 -> 66852 | Improved |
| use AGoQ instead of COAT (32K) | OLMo-1B · seq 32k | iteration time (ms): 8861 -> 8076 | Improved |
| add gradient quantization on top of activation-free optimizer quantization (O+G vs O) (ablation) | AGoQ · LLaMA2-13B seq 12K GPU | peak memory (GB): 37.7 -> 35.3 | Improved |
| add activation quantization on top of O+G (ablation) | AGoQ · LLaMA2-13B seq 12K GPU | peak memory (GB): 35.3 -> 22.3 | Improved |
| add activation quantization on top of O+G (NPU) (ablation) | AGoQ · LLaMA2-13B seq 12K NPU | peak memory (GB): 40.5 -> 29.7 | Improved |
| add dynamic bit-width compensation (DBC) (ablation) | AGoQ · sciq (LLaMA2-7B) | accuracy (%): 71.20 -> 72.80 | Improved |
| add dynamic bit-width compensation (DBC) (ablation) | AGoQ · arc_e (LLaMA2-7B) | accuracy (%): 41.88 -> 41.58 | Hurt |
| add dynamic bit-width compensation (DBC) (ablation) | AGoQ · arc_c (LLaMA2-7B) | accuracy (%): 17.66 -> 18.34 | Improved |
| use layer-aware mixed-precision instead of uniform 4-bit activations (ablation) | AGoQ · LLaMA2-7B pretraining | convergence | Improved |
| fuse quantization/dequantization with GEMM (ablation) | AGoQ · LLaMA3-8B GEMM shapes | speedup: 1.07x average; 1.05x end-to-end | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- COAT — COAT: Compressing Optimizer states and Activations for Memory-Efficient FP8 Training (ICLR 2025) · better on 3 of 3 shared results
- ZeRO-1 — ZeRO++: Extremely Efficient Collective Communication for Large Model Training (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models (AAAI 2025 · similarity 0.91)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.90)
- GANQ: GPU-Adaptive Non-Uniform Quantization for Large Language Models (ICML 2025 · similarity 0.90)
- BiLLM: Pushing the Limit of Post-Training Quantization for LLMs (ICML 2024 · similarity 0.90)
- ShiftAddLLM: Accelerating Pretrained LLMs via Post-Training Multiplication-Less Reparameterization (NeurIPS 2024 · similarity 0.90)
- LoQT: Low-Rank Adapters for Quantized Pretraining (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
