Upgrade llama.cpp from b9495 to b9543#211
Merged
Merged
Conversation
Audit: zero project sources reference any renamed or removed symbols (hparams.n_layer, nextn_predict_layers, LLAMA_STATE_SEQ_FLAGS_ON_DEVICE, clip_image_u8/f32 field access, removed clip helpers, new mtmd_helper_bitmap placeholder parameter). All breaking changes are absorbed inside upstream-compiled translation units. https://claude.ai/code/session_0162Zk8UWHYrRYjbaU4QPtt7
bernardladenthin
had a problem deploying
to
startgate
GitHub Actions
Error
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Details
The b9495→b9543 range includes significant upstream refactoring across multiple subsystems:
hparams refactor:
n_layerfield split inton_layer_all(raw count) andn_layer()member function (effective count). Sibling renamenextn_predict_layers→n_layer_nextn. Newdeepstack_mapping_arrfor per-layer deepstack injection. Project does not reference any of these symbols directly (verified via grep).State-seq API simplification: Removed
LLAMA_STATE_SEQ_FLAGS_ON_DEVICEflag; on-device is now default for partial saves/loads. Project does not callllama_state_seq_*functions directly.MTMD subsystem refactor:
clip_image_u8/clip_image_f32switched from public POD fields to private members with getter/setter API. Several public helpers removed (clip_embd_nbytes,clip_build_img_from_pixels, etc.). New Granite4 Vision projector type with QFormer-window deepstack. Project does not link MTMD TUs into JNI build; vision API surfaces throughmtmd-helper.hwith zero breaking references.New imatrix loader library: Extracted shared imatrix-loading logic into
common/imatrix-loader.{h,cpp}. Automatically picked up by JNI build via existingtarget_link_libraries(jllama PRIVATE common). Project does not use imatrix loading from Java.Server endpoints & chat parser: New token-counting endpoints.
process_mtmd_prompt()gained optionalbool is_placeholderparameter (defaulted, so existing call sites compile unchanged). LFM2/2.5 chat parser unified with newtool_list_tokensflag.Backend numerical work: Per-backend optimizations (CUDA, RISC-V, WebAssembly, Metal, OpenCL, SYCL, Vulkan, WebGPU, KleidiAI). All internal; no API surface change.
Python conversion tools: New
Granite4VisionMmprojModel, Gemma4 fix-ups, new GGUF metadata keys for deepstack/vision projectors.Test plan
grepmatches insrc/main/cpp/for any renamed/removed symbolsprocess_mtmd_promptparameter) is defaultedRelated issues / PRs
Follows the llama.cpp upgrade process documented in CLAUDE.md.
Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_0162Zk8UWHYrRYjbaU4QPtt7