llama.cpp
ggml: support quant scales for fp8 and nvfp4
#28898
Open

ggml: support quant scales for fp8 and nvfp4 #28898

0cc4m wants to merge 3 commits into osimons/ocp_fp8_cuda from 0cc4m/fp8-scaled-mm
0cc4m
github-actions github-actions added model
github-actions github-actions added testing
github-actions github-actions added Vulkan
github-actions github-actions added ggml
github-actions github-actions added CUDA
github-actions github-actions added conversion
ORippler ORippler changed the base branch from master to osimons/ocp_fp8 33 days ago
0cc4m 0cc4m force pushed from 3d30cf09 to 973988c9 33 days ago
0cc4m 0cc4m changed the base branch from osimons/ocp_fp8 to osimons/ocp_fp8_cuda 33 days ago
0cc4m 0cc4m force pushed from 973988c9 to 1b434fb6 33 days ago
0cc4m 0cc4m force pushed from 1b434fb6 to 91888218 30 days ago
0cc4m
ggerganov ggerganov assigned ggerganov ggerganov 19 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 4bcf38c6 to d4b30c45 19 days ago
ORippler ORippler force pushed from c9021925 to c9afd59d 19 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from d4b30c45 to 902cdea8 18 days ago
ORippler ORippler force pushed from c9afd59d to 05b24373 18 days ago
ORippler ORippler force pushed from 05b24373 to 2297880d 18 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 902cdea8 to ac9720e7 18 days ago
ORippler ORippler force pushed from 2297880d to b370b462 18 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from ac9720e7 to 2de9bb7a 18 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 2de9bb7a to e50eed67 18 days ago
ORippler ORippler force pushed from b370b462 to 024aea46 18 days ago
ggerganov
0cc4m 0cc4m force pushed from 024aea46 to 672af957 17 days ago
ORippler ORippler force pushed from 672af957 to 5858f054 17 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from e50eed67 to 0bc3b72a 17 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 0bc3b72a to 93f06440 17 days ago
ORippler ORippler force pushed from 5858f054 to 675729f0 17 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 93f06440 to b7d3ef0d 17 days ago
ORippler ORippler force pushed from 675729f0 to 7ea0d0b6 17 days ago
ORippler ORippler force pushed from 7ea0d0b6 to cca75601 17 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from f8acf520 to 98d0007c 17 days ago
ORippler ORippler force pushed from cca75601 to 440d74a8 17 days ago
ORippler ORippler force pushed from 440d74a8 to 07b89858 17 days ago
ORippler ORippler force pushed from 07b89858 to 9a828bdd 16 days ago
ORippler ORippler force pushed from 9a828bdd to b8a4e2cb 16 days ago
ORippler
ORippler commented on 2026-10-07
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 021df05a to b757f0d5 9 days ago
ORippler ORippler force pushed from b8a4e2cb to 7be59da2 9 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from b757f0d5 to 31d720b8 5 days ago
ORippler ORippler force pushed from 7be59da2 to d3ea927f 5 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 31d720b8 to 03a892e2 5 days ago
ORippler ORippler force pushed from d3ea927f to 7cd37056 5 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 03a892e2 to e4595bb0 5 days ago
ORippler ORippler force pushed from 7cd37056 to a7ff904e 5 days ago
ORippler ORippler force pushed from a7ff904e to 6ceb1dfc 5 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from e4595bb0 to 28413b0b 5 days ago
ORippler ORippler force pushed from 6ceb1dfc to 9084618b 4 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 28413b0b to 46836d43 4 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 46836d43 to eea05fe1 3 days ago
ORippler ORippler force pushed from 9084618b to 5a3486df 3 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from eea05fe1 to 2d1b1324 3 days ago
ORippler ORippler force pushed from 5a3486df to b9bdd103 3 days ago
ORippler ORippler force pushed from b9bdd103 to 10028cf7 3 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 2d1b1324 to 1974d2a1 3 days ago
ORippler ORippler force pushed from 10028cf7 to 5bb7d7fa 3 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 1974d2a1 to 414c1a4a 3 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 414c1a4a to fd3d4118 3 days ago
ORippler ORippler force pushed from 5bb7d7fa to 22d59e91 3 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from fd3d4118 to f096911d 2 days ago
ORippler ORippler force pushed from 22d59e91 to 31ff37c6 2 days ago
ORippler ORippler force pushed from 31ff37c6 to 0de747f4 2 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from f096911d to 83eb3105 2 days ago
0cc4m ggml: add tensor scaling (per-tensor, per-channel, per-expert, per-ex…
1ac9a328
0cc4m add meta backend scale split
dd693dfc
0cc4m add optional activation scales to mul_mat and mul_mat_id tensors
a06027dc
ORippler ORippler force pushed from 0de747f4 to a06027dc 2 days ago
ORippler ORippler force-pushed the osimons/ocp_fp8_cuda branch from 83eb3105 to d35a3ede 2 days ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
Labels
Milestone