Skip to content

Conversation

@ggerganov
Copy link
Member

Fix a regression for computing the imatrix of MoE models introduced in #6083

The wname has to be updated with the expert weight name. Otherwise, master stops early:

Oops: inconsistent size for blk.0.ffn_down.0.weight,  (view) (4096 vs 14336)

Repro:

./imatrix -m ./models/mixtral-8x7b-32k-fast/ggml-model-f16.gguf -f build/wikitext-2-raw/wiki.train.raw

@ggerganov ggerganov merged commit a0e584d into master Mar 24, 2024
@ggerganov ggerganov deleted the gg/fix-imatrix-mm-id-name branch March 24, 2024 14:18
hodlen pushed a commit to hodlen/llama.cpp that referenced this pull request Apr 1, 2024
* imatrix : fix wname for mul_mat_id ops

* also filter tensor names in mul_mat_id ops

---------

Co-authored-by: slaren <[email protected]>
hodlen pushed a commit to hodlen/llama.cpp that referenced this pull request Apr 3, 2024
* imatrix : fix wname for mul_mat_id ops

* also filter tensor names in mul_mat_id ops

---------

Co-authored-by: slaren <[email protected]>
tybalex pushed a commit to rubra-ai/tools.cpp that referenced this pull request Apr 17, 2024
* imatrix : fix wname for mul_mat_id ops

* also filter tensor names in mul_mat_id ops

---------

Co-authored-by: slaren <[email protected]>
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.

3 participants