Skip to content

Gemma4 MOE Fix / Qwen MOECompressed - #297

Merged
cavusmustafa merged 4 commits into
ravi9:dev_backend_openvinofrom
cavusmustafa:moe_gathermatmul_v4
Sep 4, 2026
Merged

Gemma4 MOE Fix / Qwen MOECompressed#297
cavusmustafa merged 4 commits into
ravi9:dev_backend_openvinofrom
cavusmustafa:moe_gathermatmul_v4

Conversation

@cavusmustafa

@cavusmustafa cavusmustafa commented Aug 25, 2026

Copy link
Copy Markdown
Collaborator

Gemma4 MOE Fix: force requantize to q4 for expert weights as a workaround to accuracy bug. Added TODO, and I will create a separate ticket with reproducer to gpu plugin.
MOECompressed works for qwen only for now.

@wine99
wine99 force-pushed the dev_backend_openvino branch from 01d5a57 to 40d3fdd Compare August 28, 2026 01:36
@cavusmustafa
cavusmustafa marked this pull request as ready for review September 3, 2026 00:01
@cavusmustafa
cavusmustafa requested a review from wine99 as a code owner September 3, 2026 00:01
@ravi9
ravi9 force-pushed the dev_backend_openvino branch from ec98377 to fd9bc04 Compare September 3, 2026 00:04
@cavusmustafa cavusmustafa changed the title Moe gathermatmul v4 Moe Gemma4 Fix / Qwen MOECompressed Sep 3, 2026
@cavusmustafa cavusmustafa changed the title Moe Gemma4 Fix / Qwen MOECompressed Gemma4 MOE Fix / Qwen MOECompressed Sep 3, 2026
@ravi9
ravi9 force-pushed the dev_backend_openvino branch from fd9bc04 to 33237ab Compare September 3, 2026 21:10
@wine99 wine99 mentioned this pull request Sep 4, 2026
@cavusmustafa
cavusmustafa merged commit fdb0241 into ravi9:dev_backend_openvino Sep 4, 2026
4 of 16 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants