Upgrade vendored llama.cpp to b9585

[mcharytoniuk]

Upgrades the vendored llama.cpp submodule to the official b9585 tag, adds a per-model chat-parser cache, and removes all clippy allow/expect suppressions.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates the vendored llama.cpp integration and adapts the Rust bindings/tests to match the new upstream APIs. It introduces a per-LlamaModel cached chat parser handle (to avoid repeated template analysis), updates chat-template application to support an enable_thinking flag, and replaces a multi-argument multimodal evaluation call with a parameter struct. It also centralizes several Clippy suppressions into per-crate Cargo.toml lint config.

Changes:

• Upgrade the vendored llama.cpp interface and adjust C++ wrappers (chat parsing, chat template application, MTMD bitmap init, and build defines).
• Add a per-model cached chat parser in LlamaModel and update parsing/apply APIs (including an enable_thinking switch).
• Refactor multimodal chunk evaluation parameters into EvalMultimodalChunksParams and update tests accordingly; move Clippy lint exceptions to crate-level config.

Reviewed changes

Copilot reviewed 35 out of 35 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
llama-cpp-test-harness/tests/harness_self_test.rs	Removes per-file Clippy expect now handled via crate lint config.
llama-cpp-test-harness/Cargo.toml	Allows unnecessary_wraps with rationale for harness trial function signatures.
llama-cpp-bindings/src/tool_call_format/paired_quote_args.rs	Removes test-module Clippy expect; relies on crate lint config.
llama-cpp-bindings/src/send_logs_to_log.rs	Makes “deny panic/unwrap/…” conditional on not(test); removes test-level suppression.
llama-cpp-bindings/src/sampled_token_classifier.rs	Changes multimodal eval API to accept EvalMultimodalChunksParams.
llama-cpp-bindings/src/mtmd/mtmd_input_chunk.rs	Updates MTMD image batch-size mismatch construction to avoid lossy casts.
llama-cpp-bindings/src/mtmd/mtmd_bitmap.rs	Adapts to upstream MTMD bitmap wrapper return type and new init flag; updates audio test fixture generation.
llama-cpp-bindings/src/mtmd/image_chunk_batch_size_mismatch.rs	Changes public mismatch field types (u32→usize/i32).
llama-cpp-bindings/src/model.rs	Adds cached chat parser handle + FFI status mapping; adds enable_thinking arg to apply_chat_template.
llama-cpp-bindings/src/llama_batch.rs	Removes Clippy suppression by using pointer casting helpers for llama_batch_get_one.
llama-cpp-bindings/src/lib.rs	Exposes new eval_multimodal_chunks_params module/type.
llama-cpp-bindings/src/eval_multimodal_chunks_params.rs	New params struct for multimodal evaluation calls.
llama-cpp-bindings/src/context/params.rs	Makes LlamaContextParams Copy and removes a Clippy suppression.
llama-cpp-bindings/src/context.rs	Removes per-fn Clippy suppression now that params are Copy and used by value.
llama-cpp-bindings/Cargo.toml	Allows literal_string_with_formatting_args crate-wide for tool-call fixtures.
llama-cpp-bindings-tests/tests/vocabulary_and_metadata.rs	Removes file-level Clippy expect; renames variables for clarity.
llama-cpp-bindings-tests/tests/sampling_and_constrained_decoding.rs	Updates throughput computation and apply_chat_template call signature.
llama-cpp-bindings-tests/tests/reasoning_markers_and_tool_calls.rs	Refactors a large test into helpers; updates apply_chat_template calls.
llama-cpp-bindings-tests/tests/multimodal_vision.rs	Updates to EvalMultimodalChunksParams-based multimodal eval API.
llama-cpp-bindings-tests/tests/multimodal_image_and_audio.rs	Adds helper to load fixture bitmaps; updates multimodal eval and expected output assertion.
llama-cpp-bindings-tests/tests/multimodal_audio.rs	Updates apply_chat_template and multimodal eval calls to new signatures.
llama-cpp-bindings-tests/tests/model_loading_errors.rs	Removes file-level Clippy expect.
llama-cpp-bindings-tests/tests/kv_cache_and_session.rs	Removes file-level Clippy expect.
llama-cpp-bindings-tests/tests/embedding_and_encoder.rs	Updates throughput calculation to avoid precision-loss lint.
llama-cpp-bindings-tests/tests/chat_template_and_message_parsing.rs	Updates apply_chat_template calls to include enable_thinking.
llama-cpp-bindings-tests/tests/backend_initialization.rs	Removes file-level Clippy expect.
llama-cpp-bindings-tests/src/build_user_prompt_with_media_marker.rs	Updates apply_chat_template call to include enable_thinking.
llama-cpp-bindings-tests/Cargo.toml	Reorders lint entries and allows unnecessary_wraps for harness-style trial fns.
llama-cpp-bindings-sys/wrapper_mtmd.cpp	Adapts to upstream mtmd_helper_bitmap_init_from_file(..., false) wrapper return type.
llama-cpp-bindings-sys/wrapper_chat_parse.h	Introduces chat parser handle API (create/free) and updates parse entrypoint to accept the parser handle.
llama-cpp-bindings-sys/wrapper_chat_parse.cpp	Implements chat parser caching primitives and PEG-native parsing flow.
llama-cpp-bindings-sys/wrapper_chat_apply.h	Extends chat-template application FFI with enable_thinking.
llama-cpp-bindings-sys/wrapper_chat_apply.cpp	Sets inputs.enable_thinking when applying chat templates.
llama-cpp-bindings-build/src/cmake_config.rs	Disables upstream app build; replaces runtime feature assertion with compile_error!.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 61 out of 61 changed files in this pull request and generated no new comments.