Daniel Bevenius [Thu, 21 Aug 2025 10:16:54 +0000 (12:16 +0200)]
examples : add model conversion tool/example (#15455)
* examples : add model conversion tool/example
This commit adds an "example/tool" that is intended to help in the
process of converting models to GGUF. Currently it supports normal
causal models and embedding models. The readme contains instructions and
command to guide through the process.
The motivation for this to have a structured and repeatable process for
model conversions and hopefully with time improve upon it to make the
process easier and more reliable. We have started to use this for new
model conversions internally and will continue doing so and improve it
as we go along. Perhaps with time this should be placed in a different
directory than the examples directory, but for now it seems like a good
place to keep it while we are still developing it.
* squash! examples : add model conversion tool/example
Remove dependency on scikit-learn in model conversion example.
* squash! examples : add model conversion tool/example
Update transformer dep to use non-dev version. And also import
`AutoModelForCausalLM` instead of `AutoModel` to ensure compatibility
with the latest version.
* squash! examples : add model conversion tool/example
Remove the logits requirements file from the all requirements file.
* Update Python environment and tools directory documentation
- Add instructions for using .venv Python environment
- Include flake8 and pyright linting tools from virtual environment
- Add tools/ as core directory in project layout
- Reference existing configuration files (.flake8, pyrightconfig.json)
* add more python dependencies to .venv
* Update copilot instructions: add backend hardware note and server testing
* Apply suggestions from code review
* Apply suggestions from code review
* Replace clang-format with git clang-format to format only changed code
* Minor formatting improvements: remove extra blank line and add trailing newline
* try installing git-clang-format
* try just clang-format
* Remove --binary flag from git clang-format and add git-clang-format installation to CI
Daniel Bevenius [Thu, 21 Aug 2025 04:12:28 +0000 (06:12 +0200)]
examples : remove references to `make` in examples [no ci] (#15457)
This commit removes references to `make` in the examples, as the build
system has been updated to use CMake directly and using `make` will now
generate an error since Commit 37f10f955f70e0158d50343d0b9a3f92d194daae
("make : remove make in favor of CMake (#15449)").
This commit addresses an inconsistency during inference by adding a new
member to the `templates_params` struct to indicate whether the chat is
in inference mode. This allows the gpt-oss specific function
`common_chat_params_init_gpt_oss` to check this flag and the
`add_generation_prompt` flag to determine if it should replace the
`<|return|>` token with the `<|end|>` token in the prompt.
The motivation for this change is to ensure that the formatted prompt of
past messages in `common_chat_format_single` matches the output of the
formatted new message. The issue is that the gpt-oss template returns
different end tags: `<|return|>` when `add_generation_prompt` is false,
and `<|end|>` when `add_generation_prompt` is true. This causes the
substring function to start at an incorrect position, resulting in
tokenization starting with 'tart|>' instead of '<|start|>'.
1. There's no llava directory in the tools directory.
2. Because the command `target_include_directories(mtmd PUBLIC .)` is used in the `mtmd` CMakeLists.txt file, other targets that link against `mtmd` automatically include the `mtmd` directory as a search path for header files. Therefore, you can remove `target_include_directories(${TARGET} PRIVATE ../llava`` or use `target_include_directories(${TARGET} PRIVATE ../mtmd`` to explicitly require the `llama-server` target to use header files from `mtmd`.
Daniel Bevenius [Wed, 20 Aug 2025 10:31:16 +0000 (12:31 +0200)]
make : remove make in favor of CMake (#15449)
This commit removes the content from the Makefile and updates the
current deprecation message to information that `make` has been
replaced by CMake instead.
The message when `make` is invoked will now be the following:
```console
$ make
Makefile:6: *** Build system changed:
The Makefile build has been replaced by CMake.
For build instructions see:
https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md
. Stop.
```
The motivation for this is that many, if not all targets fail to build
now, after changes to the system, and `make` has also been deprected for
some time now.
- Updated pip install commands to include the --break-system-packages
flag, ensuring compatibility when working with system-managed Python
environments (PEP 668).
- Note: The --break-system-packages option was introduced in 2023.
Ensure pip is updated to a recent version before using this flag.
Oleksandr Kuvshynov [Sun, 17 Aug 2025 22:28:58 +0000 (18:28 -0400)]
server : export max observed n_past value (#15361)
Add tracking for high watermark cache usage and make it available in /metrics endpoint.
Use-case: Tracking largest needed cache usage under realistic workload
to better understand memory requirements and be able to adjust
cache size/quantization for model/cache accordingly.
Jeff Bolz [Sun, 17 Aug 2025 16:08:57 +0000 (11:08 -0500)]
vulkan: Use larger workgroups for mul_mat_vec when M is small (#15355)
* vulkan: Use larger workgroups for mul_mat_vec when M is small
Also use subgroup instructions for (part of) the reduction when supported.
Without this, the more expensive reductions would eat into the benefits of
the larger workgroups.
Jeff Bolz [Sun, 17 Aug 2025 08:41:45 +0000 (03:41 -0500)]
vulkan: Optimize argsort (#15354)
- Launch an appropriate number of invocations (next larger power of two).
32 invocations is common and the barrier is much cheaper there.
- Specialize for "needs bounds checking" vs not.
- Make the code less branchy and [[unroll]] the loops. In the final code,
I see no branches inside the main loop (only predicated stores) when
needs_bounds_check is false.
- Always sort ascending, then apply the ascending vs descending option when
doing the final stores to memory.
- Copy the values into shared memory, makes them slightly cheaper to access.
Jeff Bolz [Sat, 16 Aug 2025 16:48:22 +0000 (11:48 -0500)]
vulkan: fuse adds (#15252)
* vulkan: fuse adds
Fuse adds that have the same shape, which are common in MoE models.
It will currently fuse up to 6 adds, because we assume no more than
8 descriptors per dispatch. But this could be changed.
* check runtimeDescriptorArray feature
* disable multi_add for Intel due to likely driver bug
Jeff Bolz [Sat, 16 Aug 2025 09:18:31 +0000 (04:18 -0500)]
vulkan: Support mul_mat_id with f32 accumulators (#15337)
* vulkan: Add missing bounds checking to scalar/coopmat1 mul_mat_id
* vulkan: Support mul_mat_id with f32 accumulators, but they are not hooked up
- There's no explicit way to request f32 precision for mul_mat_id, but there
probably should be, and this gets the code in place for that.
- A couple fixes to check_results.
- Remove casts to fp16 in coopmat1 FA shader (found by inspection).
Daniel Bevenius [Fri, 15 Aug 2025 17:50:52 +0000 (19:50 +0200)]
common : fix double bos, use common_chat_templates for add_bos and add_eos (#15326)
This commit updates common_chat_templates_apply_jinja to use the
the add_bos and add_eos parameters from the chat template instead of
the inputs.
The motivation for this is that currently if the `add_bos` and `add_eos`
from the input parameters are used it is possible to there will be a
missmatch between the model and the chat template which can lead to the
the removal of duplicate BOS/EOS tokens in chat.cpp `apply` to not
happen leading to two BOS tokens being added to the template.
Daniel Bevenius [Thu, 14 Aug 2025 15:56:26 +0000 (17:56 +0200)]
llama : add 18-layer model type for Gemma 3-270m (#15319)
This commit adds support for the 18-layer model type in the Gemma3
series, which is the size of the Gemma3-270m model.
The motivation for this commit is was the only change required for
Gemma3-270m to be converted to GGUF format and used with llama.cpp.
Once the model has been converted and uploaded to Huggingface it can be
used like this:
```console
$ ./build/bin/llama-cli -hf ggml-org/gemma-3-270m-GGUF:Q8_0
```
uvos [Thu, 14 Aug 2025 14:23:56 +0000 (16:23 +0200)]
HIP: Cleanup hipification header (#15285)
add expicit conversion operator to support older versions of rocm
Switch over to hip_bf16 from legacy hip_bfloat16
Simplify RDNA3 define
Reduce swap over of new hipblas api to rocm 6.5 as this version is used for rocm 7.0 previews
Jeff Bolz [Thu, 14 Aug 2025 13:38:10 +0000 (08:38 -0500)]
vulkan: perf_logger improvements (#15246)
* vulkan: perf_logger improvements
- Account for batch dimension in flops calculation.
- Fix how "_VEC" is detected for mat_mul_id.
- Fix "n" dimension for mat_mul_id (in case of broadcasting).
- Include a->type in name.
kallewoof [Thu, 14 Aug 2025 11:03:30 +0000 (20:03 +0900)]
perplexity : provide a helpful hint for has_cpl case in split_equal error. (#15304)
When attempting to do llama-perplexity on certain tasks which have coupled sequences there is a cryptic error that does not tell you what to do, which is to set the -kvu flag. This adds a hint about that fact.
add unit tested GGML_OPT_OPTIMIZER_SGD to ggml - avoids allocating
m, v tensors.
support finetune.cpp arg -opt SGD (or sgd). (default adamw as before)
llama 3.2-1b-F32 result: observed 11gb gpu ram (41 sec/epoch)
when using SGD instead of 19gb (55 sec/epoch) using adamw.
(wikipedia 100 lines finetune)
(
using the same GPU memory, adamw can only do before OOM 512
batch/context, reaching:
train: [███████▉] data=0000140/0000140 loss=0.02575±0.00099 acc=99.52±0.03% t=00:00:47 ETA=00:00:00
val: [███████▉] data=0000008/0000008 loss=4.76565±0.28810 acc=41.46±0.77% t=00:00:00 ETA=00:00:00
SGD is superior, though it converges slower, with max before OOM 1728
batch/context (esp see the better validation perf):
train: [███████▉] data=0000039/0000039 loss=0.00371±0.00010 acc=99.96±0.01% t=00:00:41 ETA=00:00:00
val: [███████▉] data=0000003/0000003 loss=5.11406±0.76034 acc=48.01±0.69% t=00:00:01 ETA=00:00:00
)
note: when finetuning long enough (or w/ enough -lr),
validation accuracy *eventually* drops ('catastrophic forgetting')
-lr-half (halflife) option useful for SGD to avoid oscillation or
super slow underdamped learning (makes setting -lr more forgiving).
terminal -lr for now is set by lr-halvings i.e. if you want at most
1/8 the inital -lr you set -lr-halvings 3.
note: objective loss not directly comparable between adamw, sgd? -
check perplexity or accuracy or consider relative improvements
for convergence
new finetune args -wd 1e-9 to enable weight decay in sgd or adamw,
and max -epochs N (default 2 as before)
cache (1 - wd*alpha) in 'adamw' opt struct -
no noticeable perf benefit, disabled (still done
for new SGD though)
since opt. memory is pre-allocated, the ggml_opt_get_optimizer_params
would probably be able to change between SGD and AdamW with each epoch
but would need to use adamw for the first (unconfirmed - no cmdline arg
to set such a policy yet)
test-opt checks adamw as before and now sgd (except for a few disabled
tests for sgd only; probably just needs logging values and adding
alternate reference values); tolerance on the 'regression'
test is broader for sgd (so we don't need many more epochs)
* Vulkan: Implement GGML_OP_OPT_STEP_SGD
* tests: Fix OPT_STEP_SGD test-backend-ops
* SGD op param store weight-decay and not 1-alpha*wd
* minor + cosmetic changes
* fix vulkan sgd
* try CI fix
---------
Co-authored-by: 0cc4m <redacted> Co-authored-by: Johannes Gäßler <redacted>
Bas Nijholt [Wed, 13 Aug 2025 18:21:31 +0000 (11:21 -0700)]
fix(nix): remove non-functional llama-cpp cachix cache from flake.nix (#15295)
The flake.nix included references to llama-cpp.cachix.org cache with a comment
claiming it's 'Populated by the CI in ggml-org/llama.cpp', but:
1. No visible CI workflow populates this cache
2. The cache is empty for recent builds (tested b6150, etc.)
3. This misleads users into expecting pre-built binaries that don't exist
This change removes the non-functional cache references entirely, leaving only
the working cuda-maintainers cache that actually provides CUDA dependencies.
Users can still manually add the llama-cpp cache if it becomes functional in the future.
Removed the legacy .devops/cloud-v-pipeline Jenkins CI configuration and introduced .github/workflows/build-riscv-native.yml for native RISC-V builds using GitHub Actions.
Oliver Simons [Wed, 13 Aug 2025 08:04:46 +0000 (10:04 +0200)]
CUDA: Optimize `reduce_rows_f32` kernel, leading up to 25x perf improvement on kernel-level and 10% perf increase for Gemma3n (#15132)
* Factor out `reduce_rows_f32` from common.cuh
This increases iteration cycle speed by not having to recompile
every kernel all the time
* Hide memory-latency by loop unrolling in reduce_rows_f32
* Further optimizations to `reduce_rows_f32`
1. Increase threadblock size to better hide latency of memory requests.
As a consequence of bigger threadblocks, do 2-step summation, using
shared memory to communicate results between invocations
2. Use sum_temp array to reduce waits on sum
3. Adjust num_unroll to reflext bigger threadblock
4. Improve default block_dims, increase support for more block_dims
* Add perf tests for `reduce_rows_f32` kernel
* Add heuristic to toggle 128/512 threads based on sm count
Break even point was the minimum of the following multiples.
| GPU Model | Nrow SM Count Multiple |
| ----------- | ----------- |
| RTX 4000 SFF ADA | 2.0x |
| RTX 6000 ADA | 2.5x |
| RTX PRO 6000 Blackwell Max-Q | 3.04x |
| RTX PRO 4500 Blackwell | 3.15x |
* Ensure perf gains also for small ncols and large nrows
Alternative to this, one could have also made the number of unrollings
template-able, but that would require compiling the kernel multiple
times, increasing binary size unnecessarily
* Modify perf and unit-tests
* Apply auto-formatting by clang
* Fix CI build failure
See https://github.com/ggml-org/llama.cpp/actions/runs/16798370266/job/47573716079?pr=15132#step:7:486
Building with VS generator worked though.
* Remove sm_count property from `ggml_backend_cuda_context`
Requested by @JohannesGaessler, and should fix remaining CI issues as a
side-effect
* Add CUB-based implementation for GGML_OP_MEAN
Currently this branch is only executed for nrows==1
* Add heuristics to execute CUB branch only when it brings perf
Heuristics were determined on the following HW:
* RTX 4000 SFF ADA
* RTX 6000 ADA
* RTX PRO 6000 Blackwell Max-Q
* RTX PRO 4500 Blackwell
* Add unit-test for CUB-based mean
Tests should run with CUDA Graphs enabled per default on NVGPUs
* Rename `USE_CUB` to `GGML_CUDA_USE_CUB`
Suggested by @JohannesGaessler
* Unindent Preprocessor directives
See
https://github.com/ggml-org/llama.cpp/pull/15132#discussion_r2269213506
overview / pkg / ggml / sources / llama.cpp / log