]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
models : fix the attn_factor for mistral3 graphs + improve consistency (#17945)
authorGeorgi Gerganov <redacted>
Fri, 12 Dec 2025 15:12:40 +0000 (17:12 +0200)
committerGitHub <redacted>
Fri, 12 Dec 2025 15:12:40 +0000 (17:12 +0200)
commit7bed317f5351eba037c2e0aa3dce617e277be1c4
tree8d672e0aa61bd9db5f4ca41379b2747dd058ce01
parentdcb7d17758a95bada4383869c8ce66ca60ac3efc
models : fix the attn_factor for mistral3 graphs + improve consistency (#17945)

* models : fix the attn_factor for mistral3 graphs

* cont : rework attn_factor correction logic

* cont : make deepseek2 consistent

* cont : add TODO

* cont : special-case DSv2

* cont : revert Mistral 3 Large changes

* cont : fix DS2 to use the original attn_factor

* cont : minor comments
convert_hf_to_gguf.py
src/llama-graph.cpp
src/llama-hparams.cpp
src/llama-hparams.h
src/llama-kv-cache.cpp
src/llama-model.cpp
src/models/deepseek2.cpp