llama.cpp Releases
609 releases of ggml-org/llama.cpp
- b10267
Refactors the initialization code for speculative decoding to reduce code duplication.
Aug 4, 2026
- b10265
Synchronizes with the ggml library repository.
Aug 4, 2026
- b10262
Implements the GATED_LINEAR_ATTN operation for the Vulkan backend.
Aug 4, 2026
- b10261
Adds validation for byte tokens in the Plamo2 vocabulary.
Aug 4, 2026
- b10259
Enables the reshaping of tensors during model loading to improve flexibility.
Aug 4, 2026
- b10258
Refactors sampler data structure by moving vocabulary size (n_vocab) into the penalty_sampler struct.
Aug 4, 2026
- b10256
Optimizes SYCL performance by parallelizing the non-contiguous concatenation kernel.
Aug 4, 2026
- b10255
Extends SYCL oneDNN Self-Attention (SDPA) support to non-FP16 KV caches (Q4_0–Q8_0 and FP32).
Aug 4, 2026
- b10254
Introduces a new chat template for DeepSeek V4 Flash 0731 with support for structured output and reasoning effort mapping.
Aug 4, 2026
- b10253
Updates the embedded cpp-httplib library to version 0.52.0.
Aug 4, 2026
- b10252
Updates the embedded BoringSSL library to version 0.20260803.0.
Aug 4, 2026
- b10251
Adds support for Multimodal (MTP) processing in the GLM-4.7-Flash model.
Aug 4, 2026
- b10250
Enhances testing infrastructure for model resolution by implementing synthetic repo listings and a loopback HTTP server.
Aug 4, 2026
- b10249
Adds a diagnostic 'get_info' tool to the server and improves OS probing robustness by reporting unknown OS on probe failure.
Aug 4, 2026
- b10248
Validates default special token IDs within the vocabulary system to ensure consistency.
Aug 4, 2026
- b10247
Replaced fixed-size arrays with dynamic allocation for split graph inputs to fix crashes when loading wide MoE models (Gemma 4, Qwen MoE, Mixtral, DeepSeek).
Aug 4, 2026
- b10246
Optimized OpenCL kernel routing for large Q6_K language model heads to use flat GEMV operations, fixing performance issues with large models like Gemma-4 E2B.
Aug 3, 2026
- b10245
Fixed a bug in the Minimax M3 graph where certain input tensors were not being utilized.
Aug 3, 2026
- b10244
Refactored the MSA (Multi-Head Attention) logic for the M3 model into a dedicated memory implementation module.
Aug 3, 2026
- b10243
Optimized memory allocation by restricting indexer cache creation to specific 'full' indexer layers to reduce memory usage.
Aug 3, 2026