Community GGUF dialect compat for ds4f-mxfp4: metadata derivation and tensor-name aliases - #662
Open
nexus-cw wants to merge 4 commits into
Open
Community GGUF dialect compat for ds4f-mxfp4: metadata derivation and tensor-name aliases#662nexus-cw wants to merge 4 commits into
nexus-cw wants to merge 4 commits into
Conversation
Community llama.cpp-dialect GGUFs of DeepSeek V4 Flash omit eight deepseek4.* keys the loader hard-requires. Derive the hc, lora and hash-layer values from keys the community dialect does carry, keeping native GGUFs on the exact required path.
Community GGUF exports use llama.cpp tensor naming that differs from ds4 canonical names in about twenty places, including missing suffixes and renamed attention and expert tensors. Add an alias table consulted only when the canonical lookup misses, so native GGUFs are unaffected.
Some community GGUFs omit deepseek4.vocab_size. When the key is missing, derive vocab size from the length of tokenizer.ggml.tokens instead of failing the load.
Author
|
Part 2 of the series is #664, stacked on this branch. |
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Author
|
Added a short README subsection (Community GGUF dialects, under Model Weights) documenting what this PR does: direct loading of llama.cpp-dialect community GGUFs, the metadata keys derived when missing, tensor-name aliasing, and the vocab_size fallback. Docs only, no code change. 🤖 Generated with Claude Code |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Addresses #661 (part 1 of 2). Lets community llama.cpp-dialect GGUFs of
DeepSeek V4 Flash, e.g.
bullerwins/DeepSeek-V4-Flash-0731-MXFP4_MOE-Q8_0,get past metadata loading and tensor lookup on
ds4f-mxfp4. Part 2, thedense tensor type conversion, is a separate PR stacked on this one.
Changes
Three commits,
ds4.conly, about 340 net lines:deepseek4.*metadata keys. The eight hc/lora/hash-layer keys the loader hard-requires are computed from keys the
community dialect does carry. Derivation runs only when a required key
is absent; native GGUFs stay on the exact
required_u32path.ds4 canonical names. The table is consulted only when the canonical
lookup misses.
deepseek4.vocab_sizeis absent, deriveit from the length of
tokenizer.ggml.tokens.Testing
make cpuon this branch: clean build, no new warnings (the only warningsare the two pre-existing
rax.cuse-after-realloc ones, untouched here).Together with part 2, this stack has served the bullerwins GGUF in
production on a GB10 CUDA node with
--ssd-streamingfor several weeks,12/12 on our eval battery.