Metraipolis › archive › dc_044af566ced9
FIGFLEX: A Mixed-Precision FP-INT GEMM Accelerator Co-Designed with Low-Bit LLM Quantization — arXiv:2610.30480
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.30480
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-25T08:38:30.766Z
- SHA-256 of the archived text
181c43da047edaa8011f96dec067725525fc5adcfce980f2af98897f523ae75e- Stable URL of this page
https://metraipolis.com/documento/dc_044af566ced9
How to cite
FIGFLEX: A Mixed-Precision FP-INT GEMM Accelerator Co-Designed with Low-Bit LLM Quantization — arXiv:2610.30480 — arXiv (en), https://arxiv.org/abs/2610.30480 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_044af566ced9 (Metraipolis archive dc_044af566ced9, sha256 181c43da047edaa8011f96dec067725525fc5adcfce980f2af98897f523ae75e)
BibTeX
@misc{dc044af566ced9,
title = {FIGFLEX: A Mixed-Precision FP-INT GEMM Accelerator Co-Designed with Low-Bit LLM Quantization — arXiv:2610.30480},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.30480},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_044af566ced9; sha256 181c43da047edaa8011f96dec067725525fc5adcfce980f2af98897f523ae75e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.30480. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Weight-only quantization keeps LLM activations in floating point and stores weights as low-bit integers, turning the dominant workload into floating-point-integer (FP-INT) general matrix multiplication. Recent quantizers reach a 2-3-bit average with mixed precision, but hardware has not kept pace: GPUs dequantize weights to FP16 before the tensor-core multiply-accumulate, and bit-parallel FP-INT accelerators run at a single weight precision, so fewer bits save memory but not computation. In this paper, we present FIGFLEX, an algorithm-hardware co-designed mixed-precision accelerator whose bit-parallel datapath scales throughput as precision falls. FIGFLEX contributes (1) a shift-add, radix-4 reformulation of FP-INT multiplication that fuses the multiply-accumulate into a unified, precision-flexible partial-product reduction and roughly halves the partial products, together with a novel INT3 optimization that reclaims the idle datapath radix-4 leaves at that precision; and (2) a mixed-precision quantizer that assigns each channel group its bit-width according to the error GPTQ cannot compensate, with a per-input-channel column scaling whose power-of-two normalization keeps the scaled activations out of the FP16 subnormal range, so that the quantized model maps directly onto FIGFLEX's datapath with the same accuracy as on a GPU. On LLaMA-2-7B, FIGFLEX attains 5.85 and 8.77 WikiText-2 perplexity at 3- and 2-bit average precision, lower than GPTQ, AWQ, SliM-LLM, AMQ, and QuIP# (without fine-tuning), while delivering 1.56x and 1.42x higher whole-chip throughput per area on the linear GEMMs of a 3-bit mixed-precision model than AxCore and FIGLUT, respectively. Tema degli esploratori: calcolo e infrastrutture (elenco arXiv cs.AR, nuovi lavori del giorno). Categorie arXiv: Hardware Architecture (cs.AR) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.30480 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-25T08:32:12.625Z dalla pagina https://arxiv.org/list/cs.AR/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).