Search: “models”

HIP: broaden MoE ncols_opt tile heuristic on RDNA3.5 architecture (#28935) It's found the MoE ncols_opt tile heuristic needs to be broadened to include the RDNA3.5 architecture. The code change is imp

vulkan: fall back to shared-memory reduction for dmmv on PowerVR (#28341) The Imagination proprietary Vulkan compiler returns VK_ERROR_UNKNOWN from vkCreateComputePipelines for every dequant mul_mat_v

vulkan: add dedicated iq4_xs mat-vec shader (#28426) vulkan: add dedicated iq4_xs mat-vec shader Dedicated mul_mat_vec_iq4_xs for the dmmv path, replacing the generic fallback. ~+6-17% token generatio

vulkan : fuse UNARY(GELU|SIGMOID|SILU|SOFTPLUS) + MUL (#27220) vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL implement fusion in unary.comp behind U

server : apply checkpoint min-step eviction only when the checkpoint list is full (#28302) The spacing eviction in create_checkpoint() keeps the oldest checkpoint and erases every later one within che

CUDA: branchless Q4_K/Q5_K unpack to speed up mmvq, L2 prefetch on DGX Spark (#26705) Update Q4_K and Q5_K to use branchless computation, which stops the scale unpack being re-executed for every colum

CUDA: Allow concurrent streams per split for multi-GPU (#28198) CUDA: Allow CUDA optimization per split for multi-GPU. Previous guard caused multi-GPU to skip the graph optimization. The graph is alre

Ollama released version v0.33.3 on September 2, 2024, based on commit 5ec5804 and signed with GitHub’s verified signature (GPG key B5690EEEBB952194). Key changes include default parameter definitions

sycl: bind the f16 KV cache in place for the oneDNN SDPA path (#27468) Measured at a live KV length of 34816 (32768 depth plus one 2048 ubatch), on Qwen3.8 27B Q4_K_S: per tensor 4 * 34816 * 256 * 2 B

vulkan: fix missing view-alias dependencies in ggml_vk_graph_optimize (#27812) vulkan: fix missing view-alias dependencies in ggml_vk_graph_optimize is_src_of doesn't treat two views of one tensor as

What's Changed Ollama's app now follows the system appearance again, restoring dark mode support Fixed the macOS app to properly hand off to an already-running instance instead of starting a second on

spec : add DFlash2 support (local convolution + candidate selector) (#27342) (#27816) spec : add DFlash2 support (local convolution + candidate selector) (#27342) support DFlash2 Add p_min in DFlash2

What's Changed MLX: Qwen3.8 Flash Next support cmake: make external compat patches idempotent MLX and llama.cpp update mlxrunner: add structured output support mlxrunner: avoid Metal GPU timeouts when

What's Changed mlx: fix mac assumptions on linux/windows by @dhiltgen in #17898 mlx update by @dhiltgen in #17886 lint fixes by @dhiltgen in #17897 app: add claude desktop app by @ParthSareen in #1789

metal : dequantize quantized KV to F16 before flash attention (#27390) metal: dequantize q8_0 KV to f16 before flash attention Add a preprocessing pass for GGML_OP_FLASH_ATTN_EXT on the Metal backend:

server: refactor sleep handling, allow access /metrics during sleep (#27376) add cached responses refactor on_sleeping_state allow accessing metrics during sleep metrics task should not reset timer up

ggml: use dynamic allocation for split graph inputs (#22789) ggml: use dynamic allocation for split graph inputs Replace fixed-size GGML_SCHED_MAX_SPLIT_INPUTS arrays with dynamically allocated buffer

ui: CWD for agent (#26518) server : extend file_glob_search for UI pickers ui : add per-conversation working directory with picker ui : add path navigation and search scope to cwd picker Treat path-li

llama : enforce the same K and V cache types for DeepSeek V4; enable FA if V cache is quantized (#25871) llama : enforce the same K and V cache types for DeepSeek V4; enable FA if V cache is quantized

ggml-cuda : disable MMQ on devices with less than 48 KiB shared memory (#26141) ggml_cuda_should_use_mmq() selects MMQ purely from the quantization type. The current MMQ configurations are designed an

ggml-webgpu: Fix some binding alias issues to support all archs, fix recurrent-state-rollback test (#25931) Add overlap glu variant to support all archs, fix recurrent-state-rollback test format Fix a

mtmd: Add Vision Support for Minimax-M3 (#25113) Add preliminary MiniMax-M3 support Text-only port that re-uses existing components: MiniMax-M2 style GQA with per-head QK-norm and partial rotary, Deep

What's Changed launch: keep Claude Code channels available by @hoyyeva in #17210 cmd: remove dead agent prompt wrappers by @ParthSareen in #17227 agent: reorder working directory instruction by @Parth

[SYCL] Flash Attention with XMX engine via oneDNN (#25222) [SYCL] F16 (default) Flash Attention with XMX engine via oneDNN graph API; Qwen3.6-27b-Q8_0 prefill speed up x1.21 at p=512 and x4.26 at p=80

sycl: Increase minimum buffer size for USM system allocations (#25525) Raise the threshold for minimum buffer size from 1 GiB to 4 GiB, based on real-world experiments of overcommitting device memory

DeepseekV4: reduce graph splits (#25702) macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu

b9949 is a release build of llama.cpp, the open-source C/C++ inference engine for running large language models locally. llama.cpp releases are published frequently with build versions in the b-series

v0.30.7-rc1 is a release candidate for Ollama, an open-source platform for running and managing large language models locally. The v0.30 series represents improved compatibility and performance using

b9523 is a release build of llama.cpp, an open-source tool for LLM inference in C/C++ that enables language model execution with minimal setup on a wide range of hardware locally and in the cloud. The

B9493 is a release of llama.cpp, an LLM inference framework in C/C++ . This release includes updates to model support, such as centralized hidden activation mappings and additions for granite embeddin

This release candidate merges updates from the upstream main branch into the llama-runner-phase-0 branch, likely incorporating recent improvements and bug fixes into the development version. Version 0

**b8771** is a build release of [llama.cpp](https://github.com/ggml-org/llama.cpp), the open-source C/C++ library for running large language model (LLM) inference locally. Like other incremental build

Ollama v0.20.7-rc0 is a release candidate that introduces renderer tests for the `nothink` mode specific to the Gemma 4 model (PR #15554). The `nothink` feature allows Gemma 4 to bypass its default ch

Ollama v0.20.8-rc0 is a release candidate that introduces MLX support for Google's Gemma 4 model family on Apple Silicon, addressing a prior limitation where Ollama would throw a `Gemma4ForConditional

**llama.cpp release b8750** is a tagged build of [llama.cpp](https://github.com/ggml-org/llama.cpp), the open-source C/C++ inference engine for large language models maintained under the ggml-org G...

Ollama v0.20.6-rc0 is a pre-release update to the Ollama local model runner, published on April 10, 2026. Key changes include adding a Hermes agent integration guide to the docs, fixing missing par...

chat: replace output string with parts (#30389) chat: replace output string with parts reserve first Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/5563975

server : escape the special tokens without std::regex (#30243) The d1-omni and kev prompts escape '<|name|>' in the request text with std::regex_replace. libstdc++ recurses for each char of the name,

webgpu: (Fix CI) fix mul_mat_id with duplicate expert ids (#30451) Fix mm-id of webgpu in the case of ids duplication formatting fix comments to avoid breaking up sentences in comments Website: https:

unicode : add opt-in NFC normalization for Unigram tokenizers (#30390) unicode : add opt-in NFC normalization for Unigram tokenizers Note: used help from Codex, Sol 6.1 Medium Fixing initial bot comme

vulkan: fix compile error (#30495) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/55839699 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm6

opencl: fix q8_0 dp4a mul_mat_id and mark PREC_F32 mul_mat_id unsupported for adreno (#30365) opencl: fix min handling for moe q8_0 dp4a GEMM Co-authored-by: Hongqiang Wang wangh@qti.qualcomm.com open

opencl: use buf+offset instead of subbuffer for adreno moe expert sort (#30418) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/55971785 macOS/iOS: macOS Ap

jinja : write floats in tojson as Python does (#30362) tojson printed floats through a C++ stream (6 significant digits, no '.0'), so 48.858844 became 48.8588 and 1.0 became 1 in tool definitions, too

cuda : use SIMT implicit GEMM conv2d on AMD without WMMA (#30325) cuda : use SIMT implicit GEMM conv2d on AMD without WMMA Assisted-by: Claude Opus cuda : update conv2d SIMT comment cuda : skip SIMT c

ggml : improve RPC tensor transfer logs (#30481) add rpc_log_format_size() and print byte counts in KiB/MiB/GiB across the RPC client, server and transport traces add rpc_log_format_tensor() and inclu