Commit Graph
10284 Commits
Author SHA1 Message Date
Xuan Son Nguyen d827cdedcf rename CODE2WAV --> GEN_WAV, update docs 2026-08-03 00:13:25 +02:00
Xuan Son Nguyen 1df59d388e use ISO 639-1 language code 2026-08-03 00:01:56 +02:00
Xuan Son Nguyen 0b6e340205 Merge branch 'master' into xsn/qwen3-tts 2026-08-02 23:55:38 +02:00
Talha Adnan 221f0f6356 metal : add SILU_BACK (#25982)
* feat(silu_back): implemented silu_back op for f32

* fix(silu_back): removed redundant asserts in ggml-metal-ops.cpp function ggml_metal_op_silu_back.
b10235
2026-08-02 22:39:28 +03:00
Georgi Gerganov 9d21b57f2e metal : add F16 support for bin ops (#26465) b10234 2026-08-02 22:28:17 +03:00
mgroeber9110 0ab9d6fed7 opencl: limit local workgroup size for GLU operation (#26383) b10233 2026-08-02 11:44:00 -07:00
Georgi GerganovandThiago Padilha fffbcbdb9d metal: implement DeepSeek V4 hyper-connections (#26459)
- Implement GGML_OP_DSV4_HC_COMB, GGML_OP_DSV4_HC_PRE, and
  GGML_OP_DSV4_HC_POST with SIMDgroup register and shuffle optimized kernels.
- Add Metal dispatch and support plumbing and test the production Sinkhorn
  iteration count and embedding width.

Assisted-by: Codex

Co-authored-by: Thiago Padilha <[email protected]>
b10232
2026-08-02 21:06:02 +03:00
Pascal bb4e0e1b3f common: support the DSpark sidecar resolution (#26458)
The dspark- files resolve like the other speculative sidecars: the
-hfd tag applies to them, a requested sidecar resolves without a full
model at the tag, and an explicit -md selection disables the discovery.
When no type is requested, dspark outranks dflash in the auto-selection
since its sidecar carries the extra Markov head.
b10231
2026-08-02 19:25:27 +02:00
Aman Gupta 3581ba0cf5 convert: add option to create separate dspark GGUF (#26452)
* convert: add option to create separate dspark GGUF

* add --no-nextn

* fix convert bug
2026-08-02 23:16:31 +08:00
akleine c745be2a2c opencl: bugfix increment ref_count in ggml_backend_opencl_init() (#26162)
Incrementing `ref_count` at the beginning is important later
in the `free()` method of the `ggml_backend_opencl_context` at program end.
If we do not increment the `ref_count`, the result would be -1 here,
and consequently, the profiling data would not be flushed and written.
( #ifdef GGML_OPENCL_PROFILING )
b10229
2026-08-02 06:43:00 -07:00
Aman Gupta 596a5795bd DeepseekV4 MTP + DSpark (#25784) b10228 2026-08-02 20:55:34 +08:00
Aldehir RojasandBart de Boer f5919bf458 chat : add qwen3 specialized parser (#26252)
* Add tagged thinking tool parser

* chat : refactor and add permute helper

* cont : add support for <tool_call> omission

* cont : update tool delimiters

* cont : add comment for qwen3-coder

* cont : fix trigger pattern for <function

---------

Co-authored-by: Bart de Boer <[email protected]>
b10227
2026-08-02 04:13:20 -05:00
KyleHagy 272700b360 sycl: fix classification of iGPUs (#26105) b10226 2026-08-02 15:10:32 +08:00
Sigbjørn Skjæret 75587a05b3 model : load MiMo V2 MTP tensors only if used (#26412) b10225 2026-08-02 09:03:05 +02:00
Masashi Yoshimura 7a2db1a0cf ggml-webgpu: add support for f16 repeat (#26307) b10224 2026-08-02 08:28:31 +02:00
Xuan-Son Nguyen 11924d4c17 test: fix some CI errors (#26415) b10223 2026-08-02 00:16:29 +02:00
Jeff Bolz a7a6d0d269 vulkan: extend topk_moe fusion to support sqrt(softplus) (#26124) 2026-08-01 14:18:07 -05:00
Alessandro de Oliveira Faria (A.K.A.CABELO) 815a2a5915 vendor : update BoringSSL to 0.20260730.0 (#26353) b10221 2026-08-01 20:53:00 +02:00
Xuan Son Nguyen 6e4da4a745 Merge branch 'master' into xsn/qwen3-tts 2026-08-01 19:02:46 +02:00
Xuan Son Nguyen dc5dd58bea fix ci 2026-08-01 19:02:42 +02:00
Xuan Son Nguyen 2307bf5a4d use ggml_build_forward_select, clean up comments 2026-08-01 18:50:12 +02:00
Xuan-Son Nguyen 89482bd665 agents: clarify comment style and jinja knowledge (#26405)
* agents: clarify comment style and jinja knowledge

* improve Security review a bit
2026-08-01 18:45:46 +02:00
Xuan Son Nguyen 026d35cfc8 update docs 2026-08-01 18:26:59 +02:00
Xuan Son Nguyen 4a9e349f63 security fixes 2026-08-01 18:10:40 +02:00
Nico c629da565c cli : persist reasoning_content in chat history (#26362)
* cli : persist reasoning_content in chat history

llama-cli collected reasoning from the stream for display but only
stored assistant content in messages, so --reasoning-preserve could
not re-inject prior thoughts on later turns.
b10219
2026-08-01 18:03:32 +02:00
Xuan Son Nguyen f695c2226c Merge branch 'master' into xsn/qwen3-tts 2026-08-01 17:03:21 +02:00
Xuan Son Nguyen e298a2e463 clean up code comments 2026-08-01 17:02:28 +02:00
Xuan Son Nguyen 8a40209704 more timings info 2026-08-01 16:43:02 +02:00
Xuan Son Nguyen 4523db999e add guide + docs 2026-08-01 16:35:17 +02:00
Xuan Son Nguyen bfc0714ad3 nits 2026-08-01 16:13:06 +02:00
Xuan Son Nguyen 511cc2fa0d fix lint 2026-08-01 15:41:58 +02:00
Xuan Son Nguyen 82dc811a6e fix compile 2026-08-01 15:38:27 +02:00
Xuan Son Nguyen bd52f3606f update docs 2026-08-01 15:38:16 +02:00
Xuan Son Nguyen 8f94fc5267 revamp llama-tts binary 2026-08-01 15:33:14 +02:00
Xuan Son Nguyen 22bc9df347 polymorphism helper 2026-08-01 15:09:29 +02:00
Xuan Son Nguyen 86020c467a update dev docs 2026-08-01 15:07:28 +02:00
Xuan Son Nguyen eae9c2c3f5 demo: use proper sampling 2026-08-01 14:42:10 +02:00
tc-mb de699957b9 mtmd: add minicpmv46 downsample (#25993)
* add minicpmv46 downsample

Signed-off-by: tc-mb <[email protected]>

* put downsample mode inside gguf.

Signed-off-by: tc-mb <[email protected]>

* build mtmd_image_preprocessor_llava_uhd

Signed-off-by: tc-mb <[email protected]>

* fix code

Signed-off-by: tc-mb <[email protected]>

* add convert

Signed-off-by: tc-mb <[email protected]>

* add 4x ignore vit merger

Signed-off-by: tc-mb <[email protected]>

---------

Signed-off-by: tc-mb <[email protected]>
b10218
2026-08-01 13:38:36 +02:00
Xuan Son NguyenandPascal 06aae58899 fuse snake op
Co-authored-by: Pascal <[email protected]>
2026-08-01 13:16:29 +02:00
Xuan Son NguyenandPascal 914fd950a3 fix clamp cold prefix
Co-authored-by: Pascal <[email protected]>
2026-08-01 12:33:52 +02:00
Piotr Wilkin (ilintar) ddd4ec1428 chat : enable tool call in thinking for DS4 (#26269) b10217 2026-08-01 00:13:07 -05:00
Xuan Son Nguyen ffe530beb8 mtmd_helper_gen_audio API 2026-08-01 00:37:00 +02:00
Xuan Son Nguyen 1f096f4ade llama: add llama_model_get_tok_embd 2026-07-31 22:49:52 +02:00
Xuan Son Nguyen 3e9e484919 demo voice clone 2026-07-31 21:56:22 +02:00
Xuan Son Nguyen a9df03d08e code2wav preserve kv between calls 2026-07-31 19:59:41 +02:00
Xuan Son Nguyen 2348de100f demo (to be removed) 2026-07-31 18:39:44 +02:00
Xuan Son Nguyen 51027785e9 wip, correct code2wav 2026-07-31 17:00:17 +02:00
Anand Patil 876a432116 vulkan: add POOL_1D op (#25431)
* vulkan : add pool1d push constants and pipeline field

Declared data structures needed for POOL1D OP, which are the vk_op_pool1d_push_constants struct and pipeline_pool1d_f32 field.

* vulkan : add pool1d compute shader

Added pool1d.comp for Vulkan backend mirroring the existing pool2d shader.

* vulkan : add full GGML_OP_POOL_1D support

Added pipeline creation and op dispatch for 1D pooling in the Vulkan backend.

* vulkan : fix pool1d shader logic

Registered pool1d_f32 in vulkan-shaders-gen.cpp and fixed tensor dimension indices and avg pool scale.

* vulkan : fix pool1d end boundary crash and expand test coverage

Fixed an issue where the shader crashed when the end boundary was negative when k0 < p0. Also, added more test cases related to this fix.
b10216
2026-07-31 16:48:58 +02:00
Xuan Son Nguyen 13bfdc667b wire up 2026-07-31 16:45:20 +02:00
Masato Nakasaka eb41d503ba vulkan: Introduce driver version check for Windows Intel GPU to mitigate crashing (#25192)
* Removed crash guard for Intel

Crash fixed from driver 32.0.101.8860

* Added driver version check for windows

* Change to convert from driverVersion rather than string

* No need to use signed

* Refactor

* allow GPU other than Xe2+

* adjusted function body position
b10215
2026-07-31 16:26:37 +02:00