]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
overview / pkg / ggml / sources / whisper.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 077c5d4)
sycl: fix UE4M3 parsing (llama/25608)
authorChris Lee <redacted>
Fri, 7 Aug 2026 05:28:53 +0000 (23:28 -0600)
committerGeorgi Gerganov <redacted>
Fri, 7 Aug 2026 18:59:49 +0000 (21:59 +0300)
commit56cb15472d5899256f1cfb9abe7b207bbdaa38fe
treebeb288ec82204f74d4cc27a75e7ae3880722278ftree
parent077c5d42d27f3f46658965d1f4afbb95e5ff1b1ecommit | diff
sycl: fix UE4M3 parsing (llama/25608)

The NVFP4 quantization format stores a scaling factor for every group of
16 weights, packed into a single UE4M3 byte.

The SYCL GPU code was converting these scale values using the E4M3 path,
but that's *signed*, and these are unsigned values.
ggml/src/ggml-sycl/common.hpp diff | blob | history
Packaging of ggerganov/whisper.cpp