Commit Graph

Select branches

Hide Pull Requests

fable5/prefetch-experts

fable5/vram-ondemand

fable5/vram-ondemand-rebased-20260726

gemini

Mono Color

• 8bccdbbff9 chat: fix parallel_tool_calls default setting based on model capabilities, add tests for parallel tool calls and structured outputs (#22217) Piotr Wilkin (ilintar) 2026-04-22 18:10:56 +02:00
• bcb5eeb645 speculative-simple : add checkpoint support (#22227) Georgi Gerganov 2026-04-22 15:44:45 +03:00
• 225088ea76 sycl: Improve mul_mat_id memory efficiency and add BF16 fast path (#22119) Akarshan Biswas 2026-04-22 18:02:56 +05:30
• 82d3f4d3b2 mtmd: also support LLAMA_ROPE_TYPE_NONE (#22242) Xuan-Son Nguyen 2026-04-22 12:16:29 +02:00
• 17f6245168 server: ignore reasoning content from transcription api (#21905) Xuan-Son Nguyen 2026-04-22 12:10:50 +02:00
• 7bfe60fdf9 mtmd, llama : Update HunyuanVL vision-language model support (#22037) manayangandwendadawen 2026-04-22 17:58:43 +08:00
• 750579ff14 common: Refactoring sampler parameters (#20429) (#22233) Ethan Turner 2026-04-22 01:40:19 -07:00
• 134d6e54d4 common/chat, server: refactor, move all conversion functions to common, add tests (#20690) Piotr Wilkin (ilintar) 2026-04-22 10:28:45 +02:00
• ca7f7b7b94 ggml-webgpu(shader): support conv2d kernels. (#21964) Chen YuanandJeremy J. Hartmann 2026-04-21 23:18:57 -04:00
• 0dedb9ef7a hexagon: add support for FILL op (#22198) Aparna M PandMax Krasnyansky 2026-04-22 04:54:20 +05:30
• 2799d933b5 ggml-webgpu: reset CPU/GPU profiling time when freeing context (#22050) Masashi Yoshimura 2026-04-22 08:05:21 +09:00
• 04fe84b69d server: allow cancel loading model (#21814) Xuan-Son Nguyen 2026-04-22 00:26:09 +02:00
• 5a4cd6741f Hexagon: DAIG op (#22195) Shreya Jain 2026-04-21 14:16:04 -07:00
• 2248799a58 hexagon: fix missing v79 entry in libggml-htp.inf (#22194) Mengsheng Wu 2026-04-22 04:53:44 +08:00
• 72d693e4fb spec : reset i_last when low acceptance streak occurs (#22168) Paul Dubs 2026-04-21 20:29:07 +02:00
• 98d2d2884e mtmd: Add support for Reka Edge 2603 (#21616) Kwa Jie HaoandPiotr Wilkin 2026-04-22 02:02:49 +08:00
• 84652b80cf arg : add --spec-default (#22223) Georgi Gerganov 2026-04-21 19:52:02 +03:00
• 52f1096f21 openvino: driver setup, CI split, thread safety, and NPU optimizations (#21944) 4 people 2026-04-21 23:58:34 +08:00
• 606fa42f5d vendor : update cpp-httplib to 0.43.1 (#22143) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-04-21 11:45:48 -03:00
• 7fc1c4ef78 metal : workaround macOS GPU interactivity watchdog (#22216) Georgi Gerganov 2026-04-21 17:24:55 +03:00
• 82209efb7e vulkan: Support F16 OP_FILL (#22177) Jeff Bolz 2026-04-21 11:01:56 +02:00
• 9998d88bc8 mtmd: correct mtmd_decode_use_mrope() (#22188) Xuan-Son Nguyen 2026-04-21 10:53:37 +02:00
• cd03ec7642 llama-ext : fix exports (#22202) Georgi Gerganov 2026-04-21 11:04:46 +03:00
• 4889afba5f sync : ggml Georgi Gerganov 2026-04-21 11:03:42 +03:00
• 041fe83d74 ggml : bump version to 0.10.0 (ggml/1463) Georgi Gerganov 2026-04-21 11:02:56 +03:00
• cfe9838d26 fit-params : refactor + add option to output estimated memory per device (#22171) Georgi Gerganov 2026-04-21 09:54:36 +03:00
• ff6b1062af server : fix hardcoded proxy connection timeout in router mode (#18760) (#22003) xris99andChristian 2026-04-21 06:41:14 +02:00
• 97895129e5 ggml-cuda: flush legacy pool on OOM and retry (#22155) leonardHONG 2026-04-21 05:30:38 +08:00
• 86f8daacfe mtmd: correct get_n_pos / get_decoder_pos (#22175) Xuan-Son Nguyen 2026-04-20 23:29:19 +02:00
• cf8b0dbda9 server : remove /api endpoints (#22165) Georgi Gerganov 2026-04-20 20:41:19 +03:00
• fd6ae4ca1c Tensor-parallel: Fix delayed AllReduce on Gemma-4 MoE (#22129) Gaurav Garg 2026-04-20 21:55:39 +05:30
• fb19f94c71 TP: fix 0-sized tensor slices, AllReduce fallback (#21808) Johannes Gäßler 2026-04-20 18:09:39 +02:00
• 7f251fdbce ggml-cpu: Optimized x86 and generic cpu q1_0 dot (follow up) (#21636) pl752 2026-04-20 21:02:54 +05:00
• a6cc43c286 ggml-webgpu: updated matrix-vector multiplication (#21738) 3 people 2026-04-20 07:37:17 -07:00
• a678916623 mtmd: refactor mtmd_decode_use_mrope (#22161) Xuan-Son Nguyen 2026-04-20 14:45:11 +02:00
• 81df3f7cfa fix: GLM-DSA crash in llama-tokenize when using vocab_only (#22102) SamareshSinghandGeorgi Gerganov 2026-04-20 02:32:46 -05:00
• de71b5f81c server : refactor "use checkpoint" logic (#22114) Georgi Gerganov 2026-04-20 08:42:37 +03:00
• 788fcbc5dd [SYCL] Fix reorder MMVQ assert on unaligned vocab sizes (#22035) Katostrofik 2026-04-20 01:39:45 -04:00
• 9d49acb2a7 server: rename --clear-idle to --cache-idle-slots (#21741) Yes You Can Have Your Own 2026-04-20 08:30:24 +03:00
• e365e658f0 vendor : update cpp-httplib to 0.42.0 (#21781) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-04-19 19:41:43 -03:00
• 4eac5b4509 CUDA: refactor mma data loading for AMD (#22051) Johannes Gäßler 2026-04-19 18:26:59 +02:00
• d5b780a676 common/autoparser : allow space after tool call (#22073) Aldehir Rojas 2026-04-19 06:28:35 -05:00
• 471540ae8a HIP: Remove unesscary NCCL_CHECK (#21914) uvos 2026-04-19 12:59:44 +02:00
• 19124078be mtmd: add pos_0 to mtmd_image_tokens_get_decoder_pos (breaking change) (#22082) Xuan-Son Nguyen 2026-04-19 11:57:21 +02:00
• bcdcc1044f ggml : reduce CPU overhead in meta backend (#22041) Gaurav GargandJohannes Gäßler 2026-04-19 15:18:35 +05:30
• 037bfe38d0 ci : install spirv-headers for vulkan-cross (#22109) Sigbjørn Skjæret 2026-04-19 09:32:08 +02:00
• 8685e7b075 convert : support sentence-transformer 5.4 config files (#22087) DowonandSigbjørn Skjæret 2026-04-19 16:25:39 +09:00
• 09b4efa95f cmake: remove CMP0194 policy to restore MSVC builds (#21934) texasichandtexasich 2026-04-19 02:25:05 -05:00
• 455d8e4be8 server : speculative checkpointing (#19493) Sascha RogmannandGeorgi Gerganov 2026-04-19 09:24:06 +02:00
• 91fef95362 rpc : refactor the RPC transport (#21998) Radoslav Gerganov 2026-04-19 10:21:53 +03:00
• 9e5647affa server: Expose media_tag on /props endpoint. (#22028) Cetarthoriphros 2026-04-18 19:27:17 -03:00
• 4f02d47339 model : refactor bias tensor variable names (#22079) Sigbjørn Skjæret 2026-04-18 20:12:00 +02:00
• 23b8cc4991 android : libcommon -> libllama-common (#22076) Sigbjørn Skjæret 2026-04-18 11:19:40 +02:00
• 59accc8863 ggml-backend-meta: add multi-segment read support in get_tensor (#22063) SamareshSingh 2026-04-18 03:04:51 -05:00
• 83d58e02fc ci : free disk space for rocm release (#22012) Sigbjørn Skjæret 2026-04-18 09:37:30 +02:00
• 89a5474f0e convert : fix (ignore for now) typings errors (#22002) Sigbjørn Skjæret 2026-04-18 09:36:41 +02:00
• fd1c0ec3f0 llama: fit ctx size for CPU only (#21568) Johannes Gäßler 2026-04-18 08:16:04 +02:00
• 45cac7ca70 ggml-webgpu: fix compiler warnings and refactor FlashAttention encoding (#21052) Reese Levine 2026-04-17 09:17:11 -07:00
• b94050e896 CUDA: use LRU based eviction for cuda graphs (#21611) Aman Gupta 2026-04-17 23:24:21 +08:00
• a279d0f0f4 ci : add android arm64 build and release (#21647) Yuri KhrustalevandSigbjørn Skjæret 2026-04-17 05:32:24 -04:00
• 268d61e178 mtmd: add missing struct tag (#22023) 65a 2026-04-17 01:48:33 -07:00
• 6990e2f1f7 libs : rename libcommon -> libllama-common (#21936) Georgi Gerganov 2026-04-17 11:11:46 +03:00
• fcc7508759 model : Gemma4 model type detection (#22027) Eric Zhang 2026-04-17 16:07:11 +08:00
• 5e6c0e18b6 opencl: refactor q8_0 set_tensor and mul_mat host side dispatch for Adreno (#21938) lhez 2026-04-16 22:28:33 -07:00
• 30dce2cf29 cli : use get_media_marker (#22017) Sigbjørn Skjæret 2026-04-17 00:12:31 +02:00
• 089dd41fe3 cmake: use glob to collect src/models sources (#22005) Xuan-Son Nguyen 2026-04-16 23:25:16 +02:00
• 85dde8dc4a hexagon: optimize HMX matmul operations (#21071) nullname 2026-04-17 04:48:34 +08:00
• 4fbdabdc61 model: using single llm_build per arch (#21970) Xuan-Son Nguyen 2026-04-16 21:10:22 +02:00
• e45dbdece8 opencl: add q5_K gemm and gemv kernels for Adreno (#21595) shaofeiqi 2026-04-16 12:08:33 -07:00
• 4adac43f6f server: tests: fetch random media marker via /apply-template (#21962) (#21980) Pascal 2026-04-16 19:46:21 +02:00
• 9db77a020c model : refactor QKV into common build_qkv and create_tensor_qkv helpers (#21245) PikaPikachu 2026-04-16 23:41:34 +08:00
• f772f6e434 model : support NVFP4 tensors for Gemma4 (#21971) Sigbjørn Skjæret 2026-04-16 16:51:47 +02:00
• b572d1ecd6 codeowners: add team member comments (#21714) Ruben Ortlam 2026-04-16 12:13:11 +02:00
• 03b3d07798 Convert: Fix NemotronH Config Parsing (#21664) Anav Prasad 2026-04-16 10:11:45 +00:00
• 3f7c29d318 ggml: add graph_reused (#21764) Aman Gupta 2026-04-16 17:21:28 +08:00
• ae2d34899e metal: Implement ROLL op (#21946) Kusha Gharahi 2026-04-16 03:54:37 -05:00
• 1e796eb41f ggml-cpu: add 128-bit RVV implementation for Quantization Vector Dot (#20633) 3 people 2026-04-16 13:15:15 +05:00
• 5637536517 ggml : implemented simd_gemm kernel for riscv vector extension (#20627) rehan-10xengineerandRehan Qasim 2026-04-16 13:14:26 +05:00
• 90fb96a7b3 devops : added spirv-headers to nix (#21965) Yuannan 2026-04-16 08:12:52 +00:00
• 82677a6ede ggml-webgpu: compute pass batching and removing profiling overhead (#21873) Reese Levine 2026-04-16 01:12:19 -07:00
• 8612ed18b7 ci : Use ggml-org/ccache-action on RISC-V as well (#21632) Ludovic Henry 2026-04-16 10:11:25 +02:00
• b1be68e8ca [SYCL] Fix Q8_0 reorder: garbage on 2nd prompt + crash on full VRAM (#21638) KatostrofikandClaude Opus 4.6 2026-04-16 01:34:05 -04:00
• 408225bb1a server: use random media marker (#21962) Xuan-Son Nguyen 2026-04-15 23:52:22 +02:00
• b3d758750a vulkan: optimize im2col (#21713) Ruben Ortlam 2026-04-15 19:04:51 +02:00
• 7e72b38bc1 cuda: Q1_0 initial backend (#21629) Pasha KhosraviandJohannes Gäßler 2026-04-15 09:38:38 -07:00
• 20d3bc2cc8 ggml-webgpu: Fix dequantization helpers to not pass in pointers (#21872) Reese Levine 2026-04-15 09:14:40 -07:00
• a6206958d2 CUDA: require explicit opt-in for P2P access (#21910) Johannes Gäßler 2026-04-15 16:01:46 +02:00
• 014dca49d6 CUDA: manage NCCL communicators in context (#21891) Johannes Gäßler 2026-04-15 15:58:40 +02:00
• adb541a6ad rpc : add native RDMA transport for RPC backend (RoCEv2) (#20590) Valeriy Dubov 2026-04-15 16:44:02 +03:00
• 80d8770804 docs: more extensive RoPE documentation [no ci] (#21953) Xuan-Son Nguyen 2026-04-15 14:45:16 +02:00
• 8dc530b86d ci: disable test-backend-ops on Vulkan llvmpipe run and resture default timeout (#21901) Ruben Ortlam 2026-04-15 10:55:21 +02:00
• e1a9a6dcbe autoparser: support case of JSON_NATIVE with per-call markers (test case: Reka-Edge) (#21892) Piotr Wilkin (ilintar) 2026-04-15 10:51:50 +02:00
• e39eba26f3 read n_ctx back after making llama_context (#21939) Matt 2026-04-15 00:24:57 -07:00
• 5d14e5d19b hexagon: optimization for HMX mat_mul (#21554) 4 people 2026-04-14 14:09:03 -07:00
• fae3a28070 ggml : remove ggml-ext.h (#21869) Xuan-Son NguyenandGeorgi Gerganov 2026-04-14 16:32:58 +02:00
• c0de6eda72 metal : fix FA support logic (#21898) Georgi Gerganov 2026-04-14 17:32:29 +03:00
• 707c0b7a6e mtmd: add mtmd_image_tokens_get_decoder_pos() API (#21851) Xuan-Son Nguyen 2026-04-14 16:07:41 +02:00
• 1f30ac0cea vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (#21572) Jeff Bolz 2026-04-14 15:17:45 +02:00
• f4b5bf2f32 ci : re-enable mac workflows (#21894) Georgi Gerganov 2026-04-14 15:58:09 +03:00
• aa0f1897b7 metal : add XIELU unary op (#20802) Seyoung Jeong 2026-04-14 21:43:59 +09:00