]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/log
pkg/ggml/sources/llama.cpp
7 weeks agokv-cache : fix seq_rm with seq_id == -1 (#15226)
Georgi Gerganov [Mon, 11 Aug 2025 10:58:24 +0000 (13:58 +0300)]
kv-cache : fix seq_rm with seq_id == -1 (#15226)

* kv-cache : fix seq_rm with seq_id == -1

ggml-ci

* cont : iterate over streams

ggml-ci

7 weeks agokv-cache : log (debug) all streams in find_slot (#15176)
Daniel Bevenius [Mon, 11 Aug 2025 09:21:19 +0000 (11:21 +0200)]
kv-cache : log (debug) all streams in find_slot (#15176)

This commit updates `llama_kv_cache_unified::find_slot` to log
information for all streams when debug is enabled.

The motivation for this change is that currently if a non-unified
kv-cache is used, then only one stream will be logged because the
code was currently uses `seq_to_stream[1]`.

7 weeks agoconvert : fix merge conflicts (#15229)
Sigbjørn Skjæret [Mon, 11 Aug 2025 09:15:44 +0000 (11:15 +0200)]
convert : fix merge conflicts (#15229)

7 weeks agoperplexity : update comments/error msg to use decode [no ci] (#15227)
Daniel Bevenius [Mon, 11 Aug 2025 08:21:24 +0000 (10:21 +0200)]
perplexity : update comments/error msg to use decode [no ci] (#15227)

This commit updates comments and error messages to use "decode" instead
of "eval" in perplexity.cpp.

The motivation for this is that `llama_eval` was renamed to
`llama_decode` a while ago, but the comments and error messages
still referred to "eval". This change ensures consistency and clarity.

7 weeks agoconvert : improve Mistral models integration (#14737)
Julien Denize [Mon, 11 Aug 2025 08:07:49 +0000 (10:07 +0200)]
convert : improve Mistral models integration (#14737)

* Improve Mistral models integration with llama.cpp

* Revert changes and fix gguf

* Revert change

* refactor convert_mistral_to_gguf.py in convert_hf_to_gguf.py

* Revert collateral

* Rename model name

* refactor

* revert

* remove duplicate

* Remove duplication code

* Fixes

* Fix flake issues

* Apply comments

* Apply comments

* Apply comments

* Fix remote

* add default chat template

* Revert

* nit

7 weeks agokleidiai: fix unsigned overflow bug (#15150)
Charles Xu [Mon, 11 Aug 2025 07:59:26 +0000 (09:59 +0200)]
kleidiai: fix unsigned overflow bug (#15150)

* kleidiai: fix unsigned overflow bug

* address review comments

7 weeks agocuda: refactored ssm_scan and use CUB (#13291)
David Zhao [Sat, 9 Aug 2025 18:29:43 +0000 (13:29 -0500)]
cuda: refactored ssm_scan and use CUB (#13291)

* cuda: refactored ssm_scan to use CUB

* fixed compilation error when when not using CUB

* assign L to constant and use size_t instead of int

* deduplicated functions

* change min blocks per mp to 1

* Use cub load and store warp transpose

* suppress clang warning

7 weeks agoCUDA: add attention sinks for tile and wmma (#15178)
Aman Gupta [Sat, 9 Aug 2025 12:00:24 +0000 (20:00 +0800)]
CUDA: add attention sinks for tile and wmma (#15178)

* CUDA: add attention sinks for tile and wmma

* Review: formatting changes + remove syncthreads from tile + remove warp_reduce_max from wmma

8 weeks agogguf-py : add Numpy MXFP4 de/quantization support (#15111)
compilade [Fri, 8 Aug 2025 21:48:26 +0000 (17:48 -0400)]
gguf-py : add Numpy MXFP4 de/quantization support (#15111)

* gguf-py : add MXFP4 de/quantization support

* ggml-quants : handle zero amax for MXFP4

8 weeks agoserver-bench: external OAI servers, sqlite (#15179)
Johannes Gäßler [Fri, 8 Aug 2025 21:04:36 +0000 (23:04 +0200)]
server-bench: external OAI servers, sqlite (#15179)

* server-bench: external OAI servers, sqlite

* Update scripts/server-bench.py

Co-authored-by: Sigbjørn Skjæret <redacted>
* Update scripts/server-bench.py

Co-authored-by: Sigbjørn Skjæret <redacted>
* Update scripts/server-bench.py

Co-authored-by: Sigbjørn Skjæret <redacted>
* raise_for_status

---------

Co-authored-by: Sigbjørn Skjæret <redacted>
8 weeks agoggml : fix field name when new ggml_backend (#14944)
AN Long [Fri, 8 Aug 2025 12:37:22 +0000 (21:37 +0900)]
ggml : fix field name when new ggml_backend (#14944)

8 weeks agovendor: sync minja (#15161)
Olivier Chafik [Fri, 8 Aug 2025 09:45:18 +0000 (10:45 +0100)]
vendor: sync minja (#15161)

* vendor: sync minja

* Update minja.hpp

* Apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <redacted>
---------

Co-authored-by: Sigbjørn Skjæret <redacted>
8 weeks agoCUDA: attention sinks for mma FlashAttention (#15157)
Johannes Gäßler [Fri, 8 Aug 2025 06:19:58 +0000 (08:19 +0200)]
CUDA: attention sinks for mma FlashAttention (#15157)

8 weeks agoopencl: support sink in `soft_max` (attn sinks) (#15152)
lhez [Fri, 8 Aug 2025 04:47:03 +0000 (13:47 +0900)]
opencl: support sink in `soft_max` (attn sinks) (#15152)

8 weeks agoconvert : support non-mxfp4 HF model (#15153)
Xuan-Son Nguyen [Thu, 7 Aug 2025 21:26:03 +0000 (23:26 +0200)]
convert : support non-mxfp4 HF model (#15153)

* convert : support non-mxfp4 HF model

* rm redundant check

* disable debug check

8 weeks agovulkan: support fattn sinks (#15126)
Jeff Bolz [Thu, 7 Aug 2025 20:44:20 +0000 (15:44 -0500)]
vulkan: support fattn sinks (#15126)

8 weeks agovulkan: Add env var to disable host visible vidmem (#15109)
Jeff Bolz [Thu, 7 Aug 2025 20:07:11 +0000 (15:07 -0500)]
vulkan: Add env var to disable host visible vidmem (#15109)

8 weeks agollama : Support intern-s1 (#14875)
RunningLeon [Thu, 7 Aug 2025 16:20:40 +0000 (00:20 +0800)]
llama : Support intern-s1 (#14875)

* support internvl

* support interns1

* resolve comments

* put interns1 in tensor mapping

* resolve comment

* move tokenizer changes to sub class

8 weeks agoHIP: add cmake option to enable compiler output of kernel resource usage metrics...
uvos [Thu, 7 Aug 2025 14:44:14 +0000 (16:44 +0200)]
HIP: add cmake option to enable compiler output of kernel resource usage metrics (#15103)

8 weeks agoggml: Skip backend library linking code when GGML_BACKEND_DL=ON (#15094)
Christian Kastner [Thu, 7 Aug 2025 11:45:41 +0000 (13:45 +0200)]
ggml: Skip backend library linking code when GGML_BACKEND_DL=ON (#15094)

Any available libraries are found and loaded dynamically at runtime.

8 weeks agoCUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16 (#15131)
Johannes Gäßler [Thu, 7 Aug 2025 08:53:21 +0000 (10:53 +0200)]
CUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16 (#15131)

* CUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16

8 weeks agoscripts: fix crash when --tool is not set (#15133)
Johannes Gäßler [Thu, 7 Aug 2025 06:50:30 +0000 (08:50 +0200)]
scripts: fix crash when --tool is not set (#15133)

8 weeks agorequirements : fix PyTorch uint64 compatibility (#15134)
Daniel Bevenius [Thu, 7 Aug 2025 03:31:48 +0000 (05:31 +0200)]
requirements : fix PyTorch uint64 compatibility (#15134)

This commit addresses an issue with the convert_hf_to_gguf script
which is currently failing with:
```console
AttributeError: module 'torch' has no attribute 'uint64'
```

This occurred because safetensors expects torch.uint64 to be available
in the public API, but PyTorch 2.2.x only provides limited support for
unsigned types beyond uint8 it seems. The torch.uint64 dtype exists but
is not exposed in the standard torch namespace
(see pytorch/pytorch#58734).

PyTorch 2.4.0 properly exposes torch.uint64 in the public API, resolving
the compatibility issue with safetensors. This also required torchvision
to updated to =0.19.0 for compatibility.

Refs: https://huggingface.co/spaces/ggml-org/gguf-my-repo/discussions/186#68938de803e47d990aa087fb
Refs: https://github.com/pytorch/pytorch/issues/58734

8 weeks agoggml: Add basic SET_ROWS support in WebGPU (#15137)
Reese Levine [Wed, 6 Aug 2025 22:14:40 +0000 (15:14 -0700)]
ggml: Add basic SET_ROWS support in WebGPU (#15137)

* Begin work on set_rows

* Work on set rows

* Add error buffers for reporting unsupported SET_ROWS indices

* Remove extra comments

8 weeks agofix profiling crash (#15072)
rmatif [Wed, 6 Aug 2025 21:17:51 +0000 (23:17 +0200)]
fix profiling crash (#15072)

8 weeks agoopencl: add `swiglu_oai` and `add_id` (#15121)
lhez [Wed, 6 Aug 2025 19:12:17 +0000 (04:12 +0900)]
opencl: add `swiglu_oai` and  `add_id` (#15121)

* opencl: add `swiglu-oai`

* opencl: add `add_id`

* opencl: add missing `add_id.cl`

8 weeks agochat : support Granite model reasoning and tool call (#14864)
Sachin Desai [Wed, 6 Aug 2025 18:27:30 +0000 (11:27 -0700)]
chat : support Granite model reasoning and tool call (#14864)

8 weeks agoFixed name `-override-tensors` to `-override-tensor` (#15129)
Juk Armstrong [Wed, 6 Aug 2025 16:28:48 +0000 (17:28 +0100)]
Fixed name `-override-tensors` to `-override-tensor` (#15129)

8 weeks agoggml : fix fallback to CPU for ununsupported ops (#15118)
Diego Devesa [Wed, 6 Aug 2025 12:37:35 +0000 (05:37 -0700)]
ggml : fix fallback to CPU for ununsupported ops (#15118)

8 weeks agochat : fix yandex chat template (#15116)
Sigbjørn Skjæret [Wed, 6 Aug 2025 11:26:49 +0000 (13:26 +0200)]
chat : fix yandex chat template (#15116)

8 weeks agochat : fix hunyuan auto-detection (#15114)
stevenkuang [Wed, 6 Aug 2025 09:48:30 +0000 (17:48 +0800)]
chat : fix hunyuan auto-detection (#15114)

Signed-off-by: stevenkuang <redacted>
8 weeks agoCANN: add support for ACL Graph (#15065)
Chenguang Li [Wed, 6 Aug 2025 06:12:42 +0000 (14:12 +0800)]
CANN: add support for ACL Graph (#15065)

* feat(cann): add optional support for ACL Graph execution

This commit adds support for executing ggml computational graphs using
Huawei's ACL graph mode via the USE_CANN_GRAPH flag. The support can be
enabled at compile time using the CMake option:

    -DUSE_CANN_GRAPH=ON

By default, ACL graph execution is **disabled**, and the fallback path
uses node-by-node execution.

Key additions:
- CMake option  to toggle graph mode
- Graph capture and execution logic using
- Tensor property matching to determine whether graph update is required
- Safe fallback and logging if the environment variable LLAMA_SET_ROWS
  is unset or invalid

This prepares the backend for performance improvements in repetitive graph
execution scenarios on Ascend devices.

Signed-off-by: noemotiovon <redacted>
* Fix review comments

Signed-off-by: noemotiovon <redacted>
* remane USE_CANN_GRAPH to USE_ACL_GRAPH

Signed-off-by: noemotiovon <redacted>
* fix typo

Signed-off-by: noemotiovon <redacted>
---------

Signed-off-by: noemotiovon <redacted>
8 weeks agoggml: WebGPU disable SET_ROWS for now (#15078)
Reese Levine [Tue, 5 Aug 2025 23:26:38 +0000 (16:26 -0700)]
ggml: WebGPU disable SET_ROWS for now (#15078)

* Add paramater buffer pool, batching of submissions, refactor command building/submission

* Add header for linux builds

* Free staged parameter buffers at once

* Format with clang-format

* Fix thread-safe implementation

* Use device implicit synchronization

* Update workflow to use custom release

* Remove testing branch workflow

* Disable set_rows until it's implemented

* Fix potential issue around empty queue submission

* Try synchronous submission

* Try waiting on all futures explicitly

* Add debug

* Add more debug messages

* Work on getting ssh access for debugging

* Debug on failure

* Disable other tests

* Remove extra if

* Try more locking

* maybe passes?

* test

* Some cleanups

* Restore build file

* Remove extra testing branch ci

8 weeks agollama : add gpt-oss (#15091)
Georgi Gerganov [Tue, 5 Aug 2025 19:10:36 +0000 (22:10 +0300)]
llama : add gpt-oss (#15091)

* oai moe

* compat with new checkpoint

* add attn sink impl

* add rope scaling yarn

* logits match with latest transformers code

* wip chat template

* rm trailing space

* use ggml_scale_bias

* rm redundant is_swa_all

* convert interleaved gate_up

* graph : fix activation function to match reference (#7)

* vocab : handle o200k_harmony special tokens

* ggml : add attention sinks support (#1)

* llama : add attn sinks

* ggml : add attn sinks

* cuda : add attn sinks

* vulkan : add support for sinks in softmax

remove unnecessary return

* ggml : add fused swiglu_oai op (#11)

* ggml : add fused swiglu_oai op

* Update ggml/src/ggml-cpu/ops.cpp

Co-authored-by: Georgi Gerganov <redacted>
* update CUDA impl

* cont : metal impl

* add vulkan impl

* test-backend-ops : more test cases, clean up

* llama : remove unfused impl

* remove extra lines

---------

Co-authored-by: Georgi Gerganov <redacted>
---------

Co-authored-by: slaren <redacted>
* repack mxfp4 upon conversion

* clean up a bit

* enable thinking

* add quick hack to render only some special tokens

* fix bf16 conversion

* remove vocab hack

* webui ok

* support chat parsing for gpt-oss

* fix webui

* direct mapping mxfp4, FINALLY

* force using mxfp4

* properly use lazy tensor

* ggml : add mxfp4

ggml : use e8m0 conversion instead of powf

Co-authored-by: Diego Devesa <redacted>
change kvalues_mxfp4 table to match e2m1 (#6)

metal : remove quantization for now (not used)

cuda : fix disabled CUDA graphs due to ffn moe bias

vulkan : add support for mxfp4

cont : add cm2 dequant

* ggml : add ggml_add_id (#13)

* ggml : add ggml_add_id

* add cuda impl

* llama : add weight support check for add_id

* perf opt

* add vulkan impl

* rename cuda files

* add metal impl

* allow in-place ggml_add_id

* llama : keep biases on CPU with --cpu-moe

* llama : fix compile error

ggml-ci

* cuda : add fallback for __nv_cvt_e8m0_to_bf16raw

ggml-ci

* cleanup

ggml-ci

* sycl : fix supports_op for MXFP4

ggml-ci

* fix Unknown reasoning format

* ggml-cpu : fix AVX build

ggml-ci

* fix hip build

ggml-ci

* cuda : add mxfp4 dequantization support for cuBLAS

ggml-ci

* ggml-cpu : fix mxfp4 fallback definitions for some architectures

ggml-ci

* cuda : fix version required for __nv_cvt_e8m0_to_bf16raw

---------

Co-authored-by: Xuan Son Nguyen <redacted>
Co-authored-by: slaren <redacted>
8 weeks agochat : only remove double bos/eos if added (#15086)
Sigbjørn Skjæret [Tue, 5 Aug 2025 18:43:36 +0000 (20:43 +0200)]
chat : only remove double bos/eos if added (#15086)

* only remove double bos/eos if added

* fix tests

8 weeks agoreadme : update hot topics (#15097)
Georgi Gerganov [Tue, 5 Aug 2025 17:19:33 +0000 (20:19 +0300)]
readme : update hot topics (#15097)

8 weeks agosycl: fix mul_mat selection (#15092)
Romain Biessy [Tue, 5 Aug 2025 16:39:55 +0000 (18:39 +0200)]
sycl: fix mul_mat selection (#15092)

8 weeks agoFix `glm4moe` bug (#15088)
Juk Armstrong [Tue, 5 Aug 2025 12:56:44 +0000 (13:56 +0100)]
Fix `glm4moe` bug (#15088)

8 weeks agowebui: fix markdown table (#15081)
Alex Wu [Tue, 5 Aug 2025 11:56:44 +0000 (19:56 +0800)]
webui: fix markdown table (#15081)

* webui: fix markdown table

* webui: fix table display with themes

8 weeks agocontext : fix index overflow on huge outputs (#15080)
compilade [Tue, 5 Aug 2025 09:27:45 +0000 (05:27 -0400)]
context : fix index overflow on huge outputs (#15080)

* context : fix overflow when re-ordering huge outputs

* context : fix logits size overflow for huge batches

8 weeks agollama : add --n-cpu-moe option (#15077)
Diego Devesa [Mon, 4 Aug 2025 23:05:36 +0000 (16:05 -0700)]
llama : add --n-cpu-moe option (#15077)

* llama : add --n-cpu-moe option

Keeps the MoE weights of the first N layers in the CPU

8 weeks agoimatrix : warn when GGUF imatrix is saved without .gguf suffix (#15076)
compilade [Mon, 4 Aug 2025 21:26:52 +0000 (17:26 -0400)]
imatrix : warn when GGUF imatrix is saved without .gguf suffix (#15076)

* imatrix : add warning when suffix is not .gguf for GGUF imatrix

* imatrix : only warn about suffix when output format is unspecified

8 weeks agocmake: Add GGML_BACKEND_DIR option (#15074)
Christian Kastner [Mon, 4 Aug 2025 19:29:14 +0000 (21:29 +0200)]
cmake: Add GGML_BACKEND_DIR option (#15074)

* cmake: Add GGML_BACKEND_DIR option

This can be used by distributions to specify where to look for backends
when ggml is built with GGML_BACKEND_DL=ON.

* Fix phrasing

8 weeks agogguf-py : add --chat-template-file to gguf_new_metadata (#15075)
Sigbjørn Skjæret [Mon, 4 Aug 2025 19:01:48 +0000 (21:01 +0200)]
gguf-py : add --chat-template-file to gguf_new_metadata (#15075)

8 weeks agomodel: support GLM 4.5 family of models (#14939)
Sam [Mon, 4 Aug 2025 18:29:25 +0000 (04:29 +1000)]
model: support GLM 4.5 family of models (#14939)

* model: Add GLM 4.5 (#14921)

Co-authored-by: Sigbjørn Skjæret <redacted>
* Merge in PR suggestions

Co-authored-by: Sigbjørn Skjæret <redacted>
* model: Add GLM 4.5 family of models (#14921)

1. Updated tensor_mapping.py with NextN tensor mappings

- Added proper tensor mappings for all NextN/MTP tensors in /Users/samm/git/llama.cpp/gguf-py/gguf/tensor_mapping.py
- Added mappings for: eh_proj, embed_tokens, enorm, hnorm, shared_head.head, shared_head.norm

2. Added num_nextn_predict_layers configuration

- Added LLM_KV_NUM_NEXTN_PREDICT_LAYERS constant to llama-arch.h and llama-arch.cpp
- Added num_nextn_predict_layers field to llama_hparams struct
- Updated GLM4_MOE parameter loading in llama-model.cpp to read this parameter
- Modified tensor loading logic to conditionally load NextN tensors based on num_nextn_predict_layers
- Added GGUF writer support in gguf_writer.py with add_num_nextn_predict_layers() method
- Updated conversion script to extract and write this parameter from HuggingFace config

3. Added FIM tokens for GLM4_MOE

- Added GLM-4.5's FIM tokens to llama-vocab.cpp:
  - <|code_prefix|> for FIM_PRE
  - <|code_suffix|> for FIM_SUF
  - <|code_middle|> for FIM_MID

4. Removed manual NextN tensor handling

- Removed the special-case handling in convert_hf_to_gguf.py that manually mapped NextN tensors
- NextN tensors are now handled automatically through the proper tensor mapping system

* glm 4.5 update tensors names

* model: glm 4.5 apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <redacted>
* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <redacted>
* model: glm 4.5 apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <redacted>
* model: glm 4.5 apply suggestions from code review

* Apply suggestions from code review

* patch broken chat template

* typings fix

* add TENSOR_SKIP flag

Co-authored-by: Diego Devesa <redacted>
* Update src/llama-model-loader.h

Co-authored-by: Sigbjørn Skjæret <redacted>
---------

Co-authored-by: Sigbjørn Skjæret <redacted>
Co-authored-by: Diego Devesa <redacted>
8 weeks agoquantize : fix confusing error message if ftype is invalid (#15071)
Sigbjørn Skjæret [Mon, 4 Aug 2025 16:11:02 +0000 (18:11 +0200)]
quantize : fix confusing error message if ftype is invalid (#15071)

8 weeks agoggml: WebGPU backend host improvements and style fixing (#14978)
Reese Levine [Mon, 4 Aug 2025 15:52:43 +0000 (08:52 -0700)]
ggml: WebGPU backend host improvements and style fixing (#14978)

* Add parameter buffer pool, batching of submissions, refactor command building/submission

* Add header for linux builds

* Free staged parameter buffers at once

* Format with clang-format

* Fix thread-safe implementation

* Use device implicit synchronization

* Update workflow to use custom release

* Remove testing branch workflow

8 weeks agovulkan: fix build when using glslang that does not support coopmat2 (#15062)
Jeff Bolz [Mon, 4 Aug 2025 05:09:19 +0000 (00:09 -0500)]
vulkan: fix build when using glslang that does not support coopmat2 (#15062)

2 months agoimatrix : use GGUF by default (#14842)
compilade [Sun, 3 Aug 2025 20:00:05 +0000 (16:00 -0400)]
imatrix : use GGUF by default (#14842)

* imatrix : use GGUF by default

* imatrix : use GGUF regardless of the output filename

The legacy format can only be produced with --output-format dat

2 months agoimatrix : fix 3d activation handling for hybrid and recurrent models (#14994)
compilade [Sun, 3 Aug 2025 19:49:13 +0000 (15:49 -0400)]
imatrix : fix 3d activation handling for hybrid and recurrent models (#14994)

* imatrix : use a single count for dense 3d tensors

* imatrix : fix 3d activations when model tensor is 2d

* imatrix : fix 3d tensor counts

2 months agomemory : handle kv_unified for hybrid models (#15050)
compilade [Sun, 3 Aug 2025 19:43:07 +0000 (15:43 -0400)]
memory : handle kv_unified for hybrid models (#15050)

2 months agovocab : JetBrains Mellum pre-tokenizer (#15045)
Csaba Kecskemeti [Sun, 3 Aug 2025 19:38:18 +0000 (12:38 -0700)]
vocab : JetBrains Mellum pre-tokenizer (#15045)

2 months agomodel : add text-only support for Kimi-VL (and find special tokens in text_config...
Gabriel Larson [Sun, 3 Aug 2025 14:56:25 +0000 (09:56 -0500)]
model : add text-only support for Kimi-VL (and find special tokens in text_config)  (#15051)

* basic kimi-vl textmodel conversion

* check config["text_config"] for special tokens

2 months agovulkan: Use coopmat2 for conv2d (#14982)
Jeff Bolz [Sun, 3 Aug 2025 12:23:57 +0000 (07:23 -0500)]
vulkan: Use coopmat2 for conv2d (#14982)

2 months agoopencl: fix adreno compiler detection logic (#15029)
lhez [Sat, 2 Aug 2025 17:51:18 +0000 (10:51 -0700)]
opencl: fix adreno compiler detection logic (#15029)

2 months agoCUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)
Johannes Gäßler [Sat, 2 Aug 2025 14:37:08 +0000 (16:37 +0200)]
CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)

2 months agocuda: make im2col a little faster (#15025) upstream/0.0.6073
leejet [Sat, 2 Aug 2025 14:15:36 +0000 (22:15 +0800)]
cuda: make im2col a little faster (#15025)

2 months agokv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)
Daniel Bevenius [Sat, 2 Aug 2025 14:14:57 +0000 (16:14 +0200)]
kv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)

This commit removes the right alignment the `n_stream` value in the
log message in the `llama_kv_cache_unified` constructor.

The motivation for this change is to enhance the readability of log
message. Currently the output looks like this:
```console
llama_kv_cache_unified: size = 2048.00 MiB (  4096 cells,  32 layers,  1/ 1 seqs), K (f16): 1024.00 MiB, V (f16): 1024.00 MiB
```
Notice that the `n_stream` value is right aligned, which makes it a
little harder to read.

With the change in this commit the output will look like
```console
llama_kv_cache_unified: size = 2048.00 MiB (  4096 cells,  32 layers, 1/1 seqs), K (f16): 1024.00 MiB, V (f16): 1024.00 MiB
```

2 months agollama : enable LLAMA_SET_ROWS=1 by default (#14959)
Georgi Gerganov [Sat, 2 Aug 2025 14:14:21 +0000 (17:14 +0300)]
llama : enable LLAMA_SET_ROWS=1 by default (#14959)

ggml-ci

2 months agocuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (#15038)
Georgi Gerganov [Sat, 2 Aug 2025 14:13:05 +0000 (17:13 +0300)]
cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (#15038)

* cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1

ggml-ci

* cont : fix cont types

ggml-ci

* cont : adopt variable names and comment from the other branch

2 months agoci : check that pre-tokenizer hashes are up-to-date (#15032)
Sigbjørn Skjæret [Sat, 2 Aug 2025 12:39:01 +0000 (14:39 +0200)]
ci : check that pre-tokenizer hashes are up-to-date (#15032)

* torch is not required for convert_hf_to_gguf_update

* add --check-missing parameter

* check that pre-tokenizer hashes are up-to-date

2 months agoconvert : fix Qwen3-Embedding pre-tokenizer hash (#15030)
Douglas Hanley [Sat, 2 Aug 2025 10:51:02 +0000 (05:51 -0500)]
convert : fix Qwen3-Embedding pre-tokenizer hash (#15030)

2 months agochat : fix multiple tool_calls on hermes-2-pro (#14962)
Jhen-Jie Hong [Sat, 2 Aug 2025 10:04:48 +0000 (18:04 +0800)]
chat : fix multiple tool_calls on hermes-2-pro (#14962)

2 months agovulkan: coopmat2 mul_mat optimizations (#14934)
Jeff Bolz [Sat, 2 Aug 2025 09:21:37 +0000 (04:21 -0500)]
vulkan: coopmat2 mul_mat optimizations (#14934)

- Increase tile size for k-quants, to match non-k-quants
- Choose more carefully between large and medium tiles, considering how it
  interacts with split_k
- Allow larger/non-power of two split_k, and make the splits a multiple of 256
- Use split_k==3 to when >1/2 and <=2/3 of the SMs would hae been used

2 months agollama-bench: rename DB table name from test to llama_bench (#15003)
R0CKSTAR [Sat, 2 Aug 2025 09:20:40 +0000 (17:20 +0800)]
llama-bench: rename DB table name from test to llama_bench (#15003)

Signed-off-by: Xiaodong Ye <redacted>
2 months agovulkan: Support ne[3]>1 in noncontig matrix-vector multiply (#15015)
Jeff Bolz [Sat, 2 Aug 2025 08:48:30 +0000 (03:48 -0500)]
vulkan: Support ne[3]>1 in noncontig matrix-vector multiply (#15015)

2 months agomodel : support Qwen3-Embedding (#15023)
Douglas Hanley [Sat, 2 Aug 2025 08:44:50 +0000 (03:44 -0500)]
model : support Qwen3-Embedding (#15023)

2 months agoserver: enable token array inputs for OAI API (#15001)
Johannes Gäßler [Sat, 2 Aug 2025 08:12:41 +0000 (10:12 +0200)]
server: enable token array inputs for OAI API (#15001)

2 months agovulkan: optimizations for direct convolution (#14933)
Jeff Bolz [Sat, 2 Aug 2025 07:57:04 +0000 (02:57 -0500)]
vulkan: optimizations for direct convolution (#14933)

* vulkan: optimizations for direct convolution

- Empirically choose a better tile size. Reducing BS_K/BS_NPQ helps fill
  the GPU. The new size should be amenable to using coopmat, too.
- Fix shmem bank conflicts. 16B padding should work with coopmat.
- Some explicit loop unrolling.
- Skip math/stores work for parts of the tile that are OOB.
- Apply fastdiv opt.
- Disable shuffles for NV.

* Three tiles sizes for CONV_2D, and a heuristic to choose

* reallow collectives for pre-Turing

* make SHMEM_PAD a spec constant

* fixes for intel perf - no shmem padding, placeholder shader core count

* shader variants with/without unrolling

* 0cc4m's fixes for AMD perf

Co-authored-by: 0cc4m <redacted>
---------

Co-authored-by: 0cc4m <redacted>
2 months agoCUDA: fix MMQ nwarps for AMD with warp_size==32 (#15014)
Johannes Gäßler [Fri, 1 Aug 2025 18:47:32 +0000 (20:47 +0200)]
CUDA: fix MMQ nwarps for AMD with warp_size==32 (#15014)

2 months agovendor : update vendored copy of google/minja (#15011)
l-austenfeld [Fri, 1 Aug 2025 14:59:06 +0000 (16:59 +0200)]
vendor : update vendored copy of google/minja (#15011)

* vendor : update vendored copy of google/minja

Signed-off-by: Lennart Austenfeld <redacted>
* Re-remove trailing whitespace

Signed-off-by: Lennart Austenfeld <redacted>
* Remove another trailing whitespace

Signed-off-by: Lennart Austenfeld <redacted>
---------

Signed-off-by: Lennart Austenfeld <redacted>
2 months agomodel : add hunyuan dense (#14878)
stevenkuang [Fri, 1 Aug 2025 13:31:12 +0000 (21:31 +0800)]
model : add hunyuan dense (#14878)

* support hunyuan_v1_dense

Signed-off-by: stevenkuang <redacted>
* update hunyuan_moe to hunyuan_v1_moe

Signed-off-by: stevenkuang <redacted>
* fix rope alpha assert and bos token

Signed-off-by: stevenkuang <redacted>
* add blank line

Signed-off-by: stevenkuang <redacted>
* Revert "update hunyuan_moe to hunyuan_v1_moe"

This reverts commit aa973ca21913aba77f6e81a935270ef7be222e75.

* use hunyuan_dense instead of hunyuan_v1_dense

Signed-off-by: stevenkuang <redacted>
* fix hunyuan_moe chat template

Signed-off-by: stevenkuang <redacted>
* remove leftover code

Signed-off-by: stevenkuang <redacted>
* update hunyuan dense chat template

Signed-off-by: stevenkuang <redacted>
* fix hunyuan dense vocab and chat template

Signed-off-by: stevenkuang <redacted>
---------

Signed-off-by: stevenkuang <redacted>
2 months agoopencl: add f16 for `add`, `sub`, `mul`, `div` (#14984)
lhez [Fri, 1 Aug 2025 11:15:44 +0000 (04:15 -0700)]
opencl: add f16 for `add`, `sub`, `mul`, `div` (#14984)

2 months agoggml : Q2k interleaving implementation - x86/x64 SIMD (#14373)
Srihari-mcw [Fri, 1 Aug 2025 06:20:33 +0000 (11:50 +0530)]
ggml : Q2k interleaving implementation - x86/x64 SIMD (#14373)

* Initial Q2_K Block Interleaving Implementation

* Addressed review comments and clean up of the code

* Post rebase fixes

* Initial CI/CD fixes

* Update declarations in arch-fallback.h

* Changes for GEMV Q2_K in arch-fallback.h

* Enable repacking only on AVX-512 machines

* Update comments in repack.cpp

* Address q2k comments

---------

Co-authored-by: Manogna-Sree <redacted>
2 months agograph : fix equal_seq() check (#14986)
Georgi Gerganov [Fri, 1 Aug 2025 03:38:12 +0000 (06:38 +0300)]
graph : fix equal_seq() check (#14986)

ggml-ci

2 months agodocker : add cann build pipline (#14591)
diannao [Fri, 1 Aug 2025 02:02:34 +0000 (10:02 +0800)]
docker : add cann build pipline (#14591)

* docker: add cann build pipline

* docker: add cann build pipline

* docker: fix cann devops

* cann : fix multi card hccl

* Update ggml/src/ggml-cann/ggml-cann.cpp

Co-authored-by: Xuan-Son Nguyen <redacted>
* Update ggml-cann.cpp

---------

Co-authored-by: Georgi Gerganov <redacted>
Co-authored-by: Xuan-Son Nguyen <redacted>
2 months agocompare-commits.sh: support both llama-bench and test-backend-ops (#14392)
R0CKSTAR [Fri, 1 Aug 2025 00:47:27 +0000 (08:47 +0800)]
compare-commits.sh: support both llama-bench and test-backend-ops (#14392)

* compare-commits.sh: support both llama-bench and test-backend-ops

Signed-off-by: Xiaodong Ye <redacted>
* Speed up the build by specifying -j 12

Signed-off-by: Xiaodong Ye <redacted>
* Remove build_number from test-backend-ops db

Signed-off-by: Xiaodong Ye <redacted>
* Apply suggestion from @JohannesGaessler

Co-authored-by: Johannes Gäßler <redacted>
* Refine tool selection logic

Signed-off-by: Xiaodong Ye <redacted>
* Address review comments

Signed-off-by: Xiaodong Ye <redacted>
---------

Signed-off-by: Xiaodong Ye <redacted>
Signed-off-by: Xiaodong Ye <redacted>
Co-authored-by: Johannes Gäßler <redacted>
2 months agoquantize : skip tensor override when in fallback mode (#14995)
Ed Addario [Thu, 31 Jul 2025 19:32:18 +0000 (20:32 +0100)]
quantize : skip tensor override when in fallback mode (#14995)

2 months agollama : add simple option to enable CPU for MoE weights (--cpu-moe) (#14992)
Diego Devesa [Thu, 31 Jul 2025 18:15:41 +0000 (11:15 -0700)]
llama : add simple option to enable CPU for MoE weights (--cpu-moe) (#14992)

2 months agoFix params bug in diffusion example (#14993)
Aman Gupta [Thu, 31 Jul 2025 17:22:58 +0000 (01:22 +0800)]
Fix params bug in diffusion example (#14993)

2 months agollama : allow other bufts when overriding to CPU, add --no-repack option (#14990)
Diego Devesa [Thu, 31 Jul 2025 16:11:34 +0000 (09:11 -0700)]
llama : allow other bufts when overriding to CPU, add --no-repack option (#14990)

2 months agoVulkan: Fix minor debug mode issues (#14899)
Ruben Ortlam [Thu, 31 Jul 2025 15:46:54 +0000 (17:46 +0200)]
Vulkan: Fix minor debug mode issues (#14899)

* vulkan: fix debug mode issues

* vulkan: remove broken check_results GGML_OP_SET_ROWS support

2 months agomtmd : support MiniCPM-V 4.0 (#14983)
tc-mb [Thu, 31 Jul 2025 15:22:17 +0000 (23:22 +0800)]
mtmd : support MiniCPM-V 4.0 (#14983)

* support minicpm-v 4

* add md

* support MiniCPM-o 4.0

* add default location

* temp rm MiniCPM-o 4.0

* fix code

* fix "minicpmv_projector" default path

2 months agoMODEL_TENSOR.SSM_DT_NORM has defined twice (#14991)
Csaba Kecskemeti [Thu, 31 Jul 2025 14:59:49 +0000 (07:59 -0700)]
MODEL_TENSOR.SSM_DT_NORM has defined twice (#14991)

* MODEL_TENSOR.SSM_DT_NORM has defined twice, and second overwritten the jamba model's layername

* correct order

2 months agoserver : implement universal assisted decoding (#12635)
g2mt [Thu, 31 Jul 2025 12:25:23 +0000 (05:25 -0700)]
server : implement universal assisted decoding (#12635)

* llama-server : implement universal assisted decoding

* Erase prompt tail for kv-cache

* set vocab_dft_compatible in common_speculative

* rename ctx_main to ctx_tgt

* move vocab_dft_compatible to spec struct

* clear mem_dft, remove mem

* detokenize id_last for incompatible models

* update comment

* add --spec-replace flag

* accept special tokens when translating between draft/main models

* Escape spec-replace

* clamp draft result to size to params.n_draft

* fix comment

* clean up code

* restore old example

* log common_speculative_are_compatible in speculative example

* fix

* Update common/speculative.cpp

Co-authored-by: Georgi Gerganov <redacted>
* Update common/speculative.cpp

Co-authored-by: Georgi Gerganov <redacted>
* Update common/speculative.cpp

Co-authored-by: Georgi Gerganov <redacted>
---------

Co-authored-by: Georgi Gerganov <redacted>
2 months agollama : merge build_moe_ffn_from_probs function into build_moe_ffn (#14968)
Dongliang Wei [Thu, 31 Jul 2025 12:12:20 +0000 (20:12 +0800)]
llama : merge build_moe_ffn_from_probs function into build_moe_ffn (#14968)

2 months agoserver : add openai-style logit_bias support (#14946)
Lukas Straub [Thu, 31 Jul 2025 12:08:23 +0000 (14:08 +0200)]
server : add openai-style logit_bias support (#14946)

Signed-off-by: Lukas Straub <redacted>
2 months agoAdd LLaDA 8b Diffusion model (#14771)
Aman Gupta [Thu, 31 Jul 2025 11:49:09 +0000 (19:49 +0800)]
Add LLaDA 8b Diffusion model (#14771)

* Add support for Llada-8b: diffusion model

* Add README

* Fix README and convert_hf_to_gguf

* convert_hf_to_gguf.py: address review comments

* Make everything in a single example

* Remove model-specific sampling

* Remove unused argmax

* Remove braced initializers, improve README.md a bit

* Add diffusion specific gguf params in set_vocab, remove setting rope_theta and rms_norm_eps

* Remove adding the mask token

* Move add_add_bos_token to set_vocab

* use add_bool in gguf_writer.py

2 months agoCANN: Improve loading efficiency after converting weights to NZ format. (#14985)
hipudding [Thu, 31 Jul 2025 11:47:20 +0000 (19:47 +0800)]
CANN: Improve loading efficiency after converting weights to NZ format. (#14985)

* CANN: Improve loading efficiency after converting weights to NZ format.

* CANN: fix typo

2 months agograph : reduce splits for recurrent and hybrid models (#14825)
compilade [Thu, 31 Jul 2025 05:02:46 +0000 (01:02 -0400)]
graph : reduce splits for recurrent and hybrid models (#14825)

* graph : avoid creating redundant s_copy views

* graph : comment the s_copy views

2 months agoopencl: add `mul_mat_f32_f32_l4_lm` and `mul_mat_f16_f32_l4_lm` (#14809)
lhez [Wed, 30 Jul 2025 21:56:55 +0000 (14:56 -0700)]
opencl: add `mul_mat_f32_f32_l4_lm` and `mul_mat_f16_f32_l4_lm` (#14809)

2 months agoquantize : fix using combined imatrix GGUFs (multiple datasets) (#14973)
Ed Addario [Wed, 30 Jul 2025 19:11:56 +0000 (20:11 +0100)]
quantize : fix using combined imatrix GGUFs (multiple datasets) (#14973)

2 months agoserver : add support for `embd_normalize` parameter (#14964)
Daniel Bevenius [Wed, 30 Jul 2025 16:07:11 +0000 (18:07 +0200)]
server : add support for `embd_normalize` parameter (#14964)

This commit adds support for the `embd_normalize` parameter in the
server code.

The motivation for this is that currently if the server is started with
a pooling type that is not `none`, then Euclidean/L2 normalization will
be the normalization method used for embeddings. However, this is not
always the desired behavior, and users may want to use other
normalization (or none) and this commit allows that.

Example usage:
```console
curl --request POST \
    --url http://localhost:8080/embedding \
    --header "Content-Type: application/json" \
    --data '{"input": "Hello world today", "embd_normalize": -1}
```

2 months agoHIP: enable mfma mmq on gfx908 and gfx90a for select datatypes and shapes (#14949)
uvos [Wed, 30 Jul 2025 15:38:06 +0000 (17:38 +0200)]
HIP: enable mfma mmq on gfx908 and gfx90a for select datatypes and shapes (#14949)

2 months agosync : ggml
Georgi Gerganov [Wed, 30 Jul 2025 13:03:13 +0000 (16:03 +0300)]
sync : ggml

ggml-ci

2 months agocmake : Fix BLAS link interface (ggml/1316)
Kai Pastor [Wed, 30 Jul 2025 12:53:16 +0000 (14:53 +0200)]
cmake : Fix BLAS link interface (ggml/1316)

2 months agovulkan : fix 32-bit builds (ggml/1313)
Kai Pastor [Wed, 30 Jul 2025 12:52:26 +0000 (14:52 +0200)]
vulkan : fix 32-bit builds (ggml/1313)

The pipeline member can be cast to VkPipeline.
This is a VkPipeline_T* on 64 bit but a uint64_t on 32 bit.
Cf. VK_DEFINE_NON_DISPATCHABLE_HANDLE documentation.

2 months agoCUDA: skip masked KV slices for all FA kernels (#14924)
Johannes Gäßler [Wed, 30 Jul 2025 13:46:13 +0000 (15:46 +0200)]
CUDA: skip masked KV slices for all FA kernels (#14924)

2 months agotests : update for LLAMA_SET_ROWS=1 (#14961)
Georgi Gerganov [Wed, 30 Jul 2025 12:12:02 +0000 (15:12 +0300)]
tests : update for LLAMA_SET_ROWS=1 (#14961)

* test-thread-safety : each context uses a single sequence

* embedding : handle --parallel argument

ggml-ci

* save-load : handle -np 1

ggml-ci

* thread-safety : avoid overriding threads, reduce test case arg

ggml-ci

2 months agograph : fix stack-use-after-return (#14960)
Georgi Gerganov [Wed, 30 Jul 2025 10:52:11 +0000 (13:52 +0300)]
graph : fix stack-use-after-return (#14960)

ggml-ci