master
model : refactor bias tensor variable names (#22079)
cmake: use glob to collect src/models sources (#22005)
fix: correct misspellings in code comments (#21217)
llama : re-enable manual LoRA adapter free (#19983)
ggml: backend-agnostic tensor parallelism (experimental) (#19378)
kv-cache : fix M-RoPE checkpoints (#20132)
model : add HunyuanOCR support (#21395)
fit-params : refactor + add option to output estimated memory per device (#22171)
cparams : rename LLAMA_MAX_PARALLEL_SEQUENCES to LLAMA_MAX_SEQ (#14188)
llama : enable chunked fused GDN path (#20340)
llama-ext : fix exports (#22202)
common/grammar: fix grammar parsing issues to prevent stack overflow and hangs (#18604)
common/grammar : replace problematic backtracking regex `[\s\S]*` (#18342)
model : refactor QKV into common build_qkv and create_tensor_qkv helpers (#21245)
llama: dynamic head_dim and n_rot for SWA (#20301)
model, mtmd: fix gguf conversion for audio/vision mmproj (#21309)
llama : correct platform-independent loading of BOOL metadata (#21428)
llama : refactor llama_context, llama_kv_cache, llm_build_context (#12181)
(revert) kv-cache : do not quantize SWA KV cache (#21332)
llama: print memory breakdown on exit (#15860)
kv-cache : support attention rotation for heterogeneous iSWA (#21513)
llama: store mrope data in KV cell (#16825)
memory: respect unified KV cache in hybrid memory for eval tasks (#21224)
memory : add llama_memory_hybrid_iswa (#18601)
llama: consistent ctx <-> buf order for KV cache (#16746)
memory : correctly handle failure in apply() (#14438)
llama: fix llama-model-saver (#20503)
ggml: add Q1_0 1-bit quantization support (CPU) (#21273)
llama : refactor `src/llama.cpp` (#10902)
llama : rename llama-sampling to llama-sampler (#19363)
vocab: add gemma4 tokenizer tests, fix edge case (#21534)
vocab: fix Gemma4 tokenizer (#21343)
server : better security control for public deployments (#9776)
llama : reduce compile time and binary size (#9712)
unicode : add custom Qwen2 regex handler to fix segfault on long input (#21257)