nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

StepFun 3.5 MTP (#23274)

Browse Source 
* StepFun 3.5 MTP

* Simplify to single layer

* Rollback core changes

* fix flake8 errors

* Remove scripts

* modify to convention

* Apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* dos2unix

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Piotr Wilkin (ilintar)
2026-06-02 17:44:35 +02:00
committed by GitHub
parent 0b7154066e
commit 2187e00337
5 changed files with 418 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
gguf-py/gguf/constants.py
View File

@@ -3994,6 +3994,13 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE_SHEXP,
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_EXP_PROBS_B,
# NextN/MTP tensors (Step3p5 draft head)
MODEL_TENSOR.NEXTN_EH_PROJ,
MODEL_TENSOR.NEXTN_EMBED_TOKENS,
MODEL_TENSOR.NEXTN_ENORM,
MODEL_TENSOR.NEXTN_HNORM,
MODEL_TENSOR.NEXTN_SHARED_HEAD_HEAD,
MODEL_TENSOR.NEXTN_SHARED_HEAD_NORM,
],
MODEL_ARCH.LLAMA_EMBED: [
MODEL_TENSOR.TOKEN_EMBD,