]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
Dflash support for nemotron-3.5 (#26905)
authorlnigam <redacted>
Tue, 11 Aug 2026 13:16:26 +0000 (18:46 +0530)
committerGitHub <redacted>
Tue, 11 Aug 2026 13:16:26 +0000 (18:46 +0530)
commitcc078b45b635b3a59aa9adc1b888150ab67798a8
tree9b0b59a9b919f645324ca24fc7b4d5d0a661f3a1
parent6e62ba538478202094edc6c100c782719e310aa3
Dflash support for nemotron-3.5 (#26905)

* conversion: skip untrained DFlash embeddings

* Add Nemotron DFlash support

* Add DFlash NVFP4 support

* Address review comments

* add missing output_s for nvfp4

* Include change for keeping residual for last layer also if requested in future dflash models

* Update conversion/qwen.py

Defensive check, not needed

Co-authored-by: Sigbjørn Skjæret <redacted>
* Fixing bug introduced by merge conflict

---------

Co-authored-by: Sigbjørn Skjæret <redacted>
conversion/base.py
conversion/qwen.py
gguf-py/gguf/constants.py
src/llama-model.h
src/models/dflash.cpp
src/models/nemotron-h.cpp