]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
[Model] Qwen3.5 dense and MoE support (no vision) (#19435)
authorPiotr Wilkin (ilintar) <redacted>
Sun, 8 Feb 2026 23:24:08 +0000 (00:24 +0100)
committerGitHub <redacted>
Sun, 8 Feb 2026 23:24:08 +0000 (00:24 +0100)
commit39bf692af1cba2a1072e4a42425611bf1ec2807d
treecea37ca0bdfd5efc0dd44084f942ca845b2f663c
parente06088da0fa86aa444409f38dff274904931c507
[Model] Qwen3.5 dense and MoE support (no vision) (#19435)

* Unified delta net handling

* Remove old methods.

* Refactor and optimize

* Adapt autoregressive version from @ymcki

* Change to decay mask approach

* Fix bad permute

* Qwen 3.5 support

* Apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <redacted>
* Further fixes

* Use inheritance, remove unneeded conts

* Not like this!

* Remove ggml.h explicit import

* Remove transformers, fix the views

* ACTUALLY fix views, make super calls explicit in conversion.

* Fix conversion again

* Remove extra ggml.h imports

---------

Co-authored-by: Sigbjørn Skjæret <redacted>
14 files changed:
convert_hf_to_gguf.py
gguf-py/gguf/constants.py
gguf-py/gguf/tensor_mapping.py
src/CMakeLists.txt
src/llama-arch.cpp
src/llama-arch.h
src/llama-context.cpp
src/llama-model.cpp
src/models/delta.cpp [new file with mode: 0644]
src/models/kimi-linear.cpp
src/models/models.h
src/models/qwen3-5.cpp [new file with mode: 0644]
src/models/qwen3-5moe.cpp [new file with mode: 0644]
src/models/qwen3next.cpp