mtmd : remove libllava, remove clip-quantize-cli (⚠️ breaking change) #13460
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ngxson
added
the
breaking change
ggerganov
approved these changes
May 12, 2025
gabe-l-hart
added a commit
to gabe-l-hart/ollama
that referenced
this pull request
Jun 24, 2025
The kv cache hierarchy was squashed so that now all of the llama-kv-cache-* implementations inherit directly from llama_memory_i and there is no intermediary llama_kv_cache base class. ggml-org/llama.cpp#14006 The llava.* tool files were migrated to mtmd.* files ggml-org/llama.cpp#13460 Branch: GraniteFour Signed-off-by: Gabe Goodhart <[email protected]>
gabe-l-hart
added a commit
to gabe-l-hart/ollama
that referenced
this pull request
Jun 25, 2025
The llava.* tool files were migrated to mtmd.* files ggml-org/llama.cpp#13460 Branch: GraniteFour Signed-off-by: Gabe Goodhart <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In this PR:
libllava- it contains too many redundant and unsafe code - thelibmtmdalready covers all use cases with a better APIclip-quantize-clibecause it's already broken a long time ago - it will be replaced soon ; In the meantime, if you need to quantize vision models, useconvert_hf_to_gguf.py --outtype, minimum type supported isq8_0mtmd/legacy-models; new models can be converted usingconvert_hf_to_gguf.py --mmprojNOTE: in the next PR, many APIs will be removed from
clip.h, as we will convertclip.cppto be used internally bylibmtmd