llama.cpp
[Model] Qwen3.5 dense and MoE support (no vision)
#19435

Merged

[Model] Qwen3.5 dense and MoE support (no vision) #19435

pwilkin merged 16 commits into ggml-org:master from pwilkin:qwen35

Unified delta net handling

93654351

Remove old methods.

7118cc22

Refactor and optimize

c58922df

Adapt autoregressive version from @ymcki

e480e243

Change to decay mask approach

50c8c872

Fix bad permute

80d2772a

Qwen 3.5 support

a87d23f8

pwilkin requested a review from

CISC 47 days ago

pwilkin requested a review from

ggerganov 47 days ago

github-actions added model

github-actions added python

CISC commented on 2026-02-08

Apply suggestions from code review

926f6b5f

Further fixes

cbe50dfb

Use inheritance, remove unneeded conts

3c7fddfe

Not like this!

4e425802

Remove ggml.h explicit import

1df85b73

CISC commented on 2026-02-08

Remove transformers, fix the views

b4eea4d1

ACTUALLY fix views, make super calls explicit in conversion.

8b011092

CISC commented on 2026-02-08

CISC approved these changes on 2026-02-08

Fix conversion again

bd078704

Remove extra ggml.h imports

1a8d7215

pwilkin merged 39bf692a into master 47 days ago

ngxson commented on 2026-02-08

Reviewers

CISC

ngxson

ggerganov

Assignees

No one assigned

Labels

model python

Milestone

No milestone

llama.cpp [Model] Qwen3.5 dense and MoE support (no vision) #19435 Merged

[Model] Qwen3.5 dense and MoE support (no vision) #19435

llama.cpp
[Model] Qwen3.5 dense and MoE support (no vision)
#19435

Merged