Bump llama.cpp to b6b003d2c (b10944, +64 from 4b98ab805) - #93
Merged
Merged
Conversation
Submodule vendor/llama.cpp 4b98ab805 -> b6b003d2c and Makefile LLAMA_COMMIT to match. One binding edit: #28715 renamed common_speculative_draft_params.n_past to pos0 (pure rename), so the MTP draft loop sets dp.pos0. include/llama.h, ggml-backend.h, ggml-rpc.h and the default param values are unchanged; #28736 rewrote json-schema-to-grammar.h but kept the json_schema_to_grammar overload the NIF uses, and changed the rejected-schema error text. The two grammar tests that pinned that wording now assert the error names the offending type/$ref. Metal source build on M4 Max: default suite 428 passed / 149 excluded; --include smoke --include embeddings --include slow 558 passed / 19 excluded (Qwen3.5-0.8B-UD-Q8_K_XL, Qwen3-Embedding-0.6B-f16). MTP tags not run: no MTP model available. The three tracked upstream defects still stand (source diff, see docs/release-guide.md).
CHANGELOG: `## Unreleased` becomes `## v0.8.49`, with a note that the section also covers v0.8.44-v0.8.48, which were tagged without a heading, and a re-verification paragraph for b10944. mix.exs @Version 0.8.43 -> 0.8.49; it had not moved since v0.8.43 (the tag drives CI, which rewrites it), so this brings the checked-in value back in line with the tag.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bump llama.cpp to b10944 (
b6b003d2c), release v0.8.49Submodule
vendor/llama.cpp4b98ab805→b6b003d2c(tagb10944, +64 commits; +114 since b10830),LLAMA_COMMITmoved with it.@version0.8.43→0.8.49— it had not been bumped since v0.8.43 whilev0.8.44–v0.8.48 shipped off tags, so this brings the checked-in value back in line.
Binding changes
common/speculative.h(#28715) renamedcommon_speculative_draft_params.n_past→pos0. Pure rename (samellama_pos, the positionid_lastis decoded at); the MTP draft loop now setsdp.pos0.common/json-schema-to-grammar.h(#28736) rewritten aroundcommon_chat_schema. Thejson_schema_to_grammar(const common_json &, bool)overload the NIF calls is intact;common_schema_info(unusedhere) is gone. Rejected-schema error text changed (
Unrecognized schema: {…}→JSON schema error at #: unrecognized type <name>,Error resolving ref→cannot resolve $ref <path>). Two grammar testspinned the old wording; they now assert the error names the offending type /
$ref.include/llama.h,ggml-backend.h,ggml-rpc.h: unchanged.llama_model_default_params()/llama_context_default_params(): value-identical.common/chat.h: additive only(
common_chat_tool_parameters).Upstream defects we work around
All three still stand, re-checked as a source diff over b10830 → b10944:
ggml-rpc.cppuntouched (2-D tensor hooks stillNULL),ggml_backend_cuda_comm_inituntouched,-mcpu=nativeprobe untouched(
ggml-cpu/CMakeLists.txtdiff is #28091 PCH/unity build and #28667 s390x repack). Noted indocs/release-guide.md.Verification (Metal source build, Apple M4 Max)
mix test(no model)--include smoke --include embeddings --include slow— Qwen3.5-0.8B-UD-Q8_K_XL, Qwen3-Embedding-0.6B-f16mix format --check-formattedNot run:
:mtp/:mtp_sidecar(no MTP GGUF on hand),:rpc_live(needs a worker).Changelog
## Unreleased→## v0.8.49, with a note that the section also covers v0.8.44–v0.8.48 (tagged without headings) and a re-verification paragraph for b10944.Release
Merge, then
git tag v0.8.49 && git push origin v0.8.49; the precompile workflow builds artifacts, commitschecksum.exs, and publishes to Hex.