Bump llama.cpp to 5b59b83f4 (b11050+1, +61 from 72b590d65), release v0.8.52 - #96
Merged
Merged
Conversation
…0.8.52 NIF-facing headers changed additively only (LLAMA_VOCAB_TYPE_TEST, llama_adapter_lora_init_from_file_ptr, ggml_dsv4_hc_pre_gated); default params untouched; RPC protocol stays 7.0.0. All three release-guide workarounds still stand. Notable upstream: RPC worker drops cached graphs on free_buffer (#24292 use-after-free), Metal MoE/SSM_CONV fusion and mul_mm_id NaN fix, CUDA graph for MTP draft. Verified on Metal (M1 Max): 430 passed default, 573 passed with smoke/embeddings/slow/mtp, 436 passed / 6 skipped for mtp_sidecar with the Qwen 3.8 pair. Hex source build clones the pinned SHA.
nyo16
force-pushed
the
bump-llama-cpp-5b59b83f4
branch
from
September 19, 2026 12:17
4daf0b9 to
bcaa47b
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bumps
vendor/llama.cppfrom72b590d65(b10989+1, upstream v0.4.1) to5b59b83f4(b11050+1, origin/master), 61 commits.LLAMA_COMMITin the Makefile follows thesubmodule.
API compatibility
NIF source is untouched. Only additive header changes:
llama.h:LLAMA_VOCAB_TYPE_TEST,llama_adapter_lora_init_from_file_ptrggml.h:ggml_dsv4_hc_pre_gatedllama_model_default_params/llama_context_default_paramsunchanged. RPC protocol stays at 7.0.0, so a v0.8.51 worker and v0.8.52 client still interoperate.Notable upstream changes
free_buffernow discards cached compute graphs, closing a use-after-free ingraph_recompute(#24292). Relevant to long-livedLlamaCppEx.RPC.Serverinstances serving multiple model loads.
SSM_CONVfusion (#28948), corrected FA support checks (#29122), NaN fix inmul_mm_idwhen activations exceed f16 range (#26223).\n</think>on reasoning-budget end for qwen3-coder (#28869).ufakzekapre-tokenizer (#29033).Release-guide workarounds
All three still stand (source diff):
ggml-cpu/CMakeLists.txtuntouched;ggml-cuda.cudiff is aDSV4_HC_POSTsupports_optweak nowhere nearggml_backend_cuda_comm_init; RPC bufferset_tensor_2d/get_tensor_2dstillNULL;ggml_backend_rpc_start_serverstill returnsvoid.Verification
macOS, Metal, M1 Max,
GGML_METAL_NO_RESIDENCY=1:mix test(no model)--include smoke --include embeddings --include slow --include mtp(Qwen3.5-0.8B-UD-Q4_K_XL, Qwen3-Embedding-0.6B-f16, Qwen3.6-35B-A3B-MTP-UD-Q4_K_XL)--include mtp_sidecar(Qwen3.8-27B-Q4_K_M + mtp-Qwen3.8-27B-Q4_0)mix hex.buildtarball,LLAMA_BACKEND=cpu5b59b83f4, linksmix format --check-formattedNot re-run:
rpc_live,MTPE4BSidecarTest, dialyzer (no Elixir source touched).