llama.cpp
Quant: OCP FP8 E4M3 support
#27512
Open
Overview
Commits
21
Changes
View On GitHub
Commits
  • WIP OCP FP8 E4M3 support
    ORippler committed 3 hours ago
  • Fix python type-checker
    ORippler committed 3 hours ago
  • Materialize input_scales also for FP8 when converting
    ORippler committed 3 hours ago
  • Fix conversion of https://huggingface.co/RedHatAI/Qwen3.8-27B-NVFP4
    ORippler committed 3 hours ago
  • Allow remaining FP8 tensors to be dequantized during conversion
    ORippler committed 3 hours ago
  • Allow tensor-strategy for FP8-only compressed-tensor ckpts
    ORippler committed 3 hours ago
  • Fix and test gguf-py behavior for FP8_E4M3
    ORippler committed 3 hours ago
  • Disallow QKV-fusion for FP8 temporarily
    ORippler committed 3 hours ago
  • Fix conversion of fused qkvz & FP8 for Qwen3-Next
    ORippler committed 3 hours ago
  • Avoid fusing FP8 for --fuse-gate-up-exps as we may have different ws
    ORippler committed 3 hours ago
  • Emit warnings when not fusing FP8 during conversion
    ORippler committed 3 hours ago
  • More FP8 conversion fixes
    ORippler committed 3 hours ago
  • Add supported for strided copies to/from FP8 in CPU
    ORippler committed 3 hours ago
  • remove unneeded scratch buffer from FP8 cpu cpy/dup path
    ORippler committed 3 hours ago
  • add explanatory comments and TODOs
    ORippler committed 3 hours ago
  • Add dedicated encoder and decoder tests for FP8
    ORippler committed 3 hours ago
  • Whitelist ggml_op_get_rows in metal instead of blacklisting
    ORippler committed 3 hours ago
  • Fix more conversion bugs
    ORippler committed 3 hours ago
  • Defer weight materialization during conversion
    ORippler committed 3 hours ago
  • Blacklist fp8 in metal backend instead of white-list
    ORippler committed 3 hours ago
  • Adjust indents
    ORippler committed 3 hours ago
FAQ Terms Privacy Refunds Impressum
Loading