* run the x64 ci on regular machines
* set up the same thing for arm
fix test-quantize-perf just like #12306
* try to disable sve
* add another sve run
* ggml : make gallocr respect the backend's max buffer size
* if the graph requires more memory than can fit into a single allocation, split it into multiple backend buffers
* vulkan: report the actual max allocation size in buffer type interface
* fix missing newline, apple-clang warning
* track size of individual chunks in ggml_dyn_tallocr and raise max chunks.
revert to use suballocation_block_size as max chunk size for vulkan.
* track (chunk, offset) pairs instead of "global" offsets through gallocr.
* simpler, don't need loops to map between local/global offsets
* touches more code
* fix dyn_tallocr_max_size and initialization
* fix memory leak when buffers are reused due to same buffer type appearing multiple times
* make vbuffer allocation follow the same logic as backend_buffer did before
* continue to use leftover unallocated space of previous chunks after a new one has been created
* treat free blocks of each chunk as separate list
* they're still allocated together, but start/end of each chunk is tracked, and allocate/free iterate over sub-ranges
* exhaust freed blocks of all chunks before considering their last blocks with unallocated space
* start with 0 chunks/blocks and create chunks as needed
* allow the last chunk to grow beyond max size
* refactor: move adding new free block and new chunk into separate functions
* allocate chunks individually with a separate free-blocks list for each one
* needs a bit more memory/allocations/indirections, but code is simpler
* fix warnings (missing static) & debug checks
* implement set_rows with i32 index
* template fix
* test quantized path
warnings--
* Apply suggestions from code review
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* forgotten name change
* deduplicate cuda/sycl and test-fix
* indent++
* vulkan: support set_rows with i32 index type (#16162)
* disable i32 index for webgpu for now
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* Vulkan: add conv_transpose_2d operation
* Vulkan: fix typo in conv_transpose_2d shader(s0mp, s0L, s1mp, s1L)
* Vulkan: fix incorrect indentation in conv_transpose_2d shader
* Vulkan: add checking the push constants size limit and reuse conv2d_mm.comp for conv_transpose_2d operation
* Vulkan: revert the order of the index calculation and bound check in conv_2d shader
* Vulkan: explicity check push constants limit in supports_op() for conv_transpose_2d operation.
* Vulkan: remove unnecessary lower bound checks for H/W_idx in the conv_2d shader.
* CUDA: Optimize PAD_REFLECT_1D
feat: add more test cases for PAD_REFLECT_1D
* use fast_div to improve performance
* Apply suggestion from JohannesGaessler
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Apply suggestion from JohannesGaessler
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* optimize
* use a concise expression to further speedup the cuda kernel
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* 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
* some f32 tests passing
* Disable set_rows until it's implemented
* f32 add all tests passing
* Begin work on set_rows
* Work on set rows
* Add error buffers for reporting unsupported SET_ROWS indices
* Remove extra comments
* Add templated addition, clean up code
* Get addition and multiplication working
* Implement rms_norm
* Add get_rows implementation
* Add new get_rows files
* Refactor use of wg size entry
* Fix compilation
* Try manually unrolled q4_0 quant
* Revert "Try manually unrolled q4_0 quant"
This reverts commit 77f8b96515.
* Move to constant max wg size
* Check for tensor size in supports_op
* Vectorize f32 and change default workgroup size
* Move f32 get_rows from < 4 to % 4 != 0
* fix linter errors
* Add in-place tests
---------
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
* metal : improve naming
* metal : refactor device
ggml-ci
* cont : props
ggml-ci
* metal : apply ggml_mem_ranges_t
ggml-ci
* metal : remove GGML_METAL_USE_BF16
ggml-ci
* metal : refactor device buffer
ggml-ci
* cont : fix naming
* metal : sync before destroying the backend
ggml-ci
* metal : refactor context
ggml-ci
* metal : migrate ggml-metal.m to ggml-metal.cpp
ggml-ci
* metal : adjust ops API
ggml-ci
* metal : use C++ to store piplienes
ggml-ci
* metal : migrate ops to separate functions
ggml-ci
* metal : add ggml_metal_library_t
ggml-ci
* metal : improve naming
ggml-ci
* metal : cleanp
ggml-ci
* metal : add support for GGML_OP_LOG
ggml-ci
* metal : fix error handling
ggml-ci
* Add fastdiv and fastmodulo to k_bin_bcast kernel
* Address review comments
* `prod_` instead of `prod` suffix
* Add test case for `k_bin_bcast_unravel` in CUDA backend
* CUDA: Add mul_mat_id support the mmf
Add support for mul_mat_id for bs < 16
* Review: use warp_size, fix should_use_mmf condition
* Launch one block per expert, stride along n_expert_used
* templatize mul_mat_id
* Pad shmem to 16 bytes, add helper function mul_mat_f_switch_ids
* Reduce compile times by dividing mmf into f16, bf16 and f32 variants
* Divide mmf by ncols_dst
* Add missing files
* Fix MUSA/HIP builds
* requirements : update transformers/torch for Embedding Gemma
This commit updates the requirements to support converting
Embedding Gemma 300m models.
The motivation for this change is that during development I had a local
copy of the transformers package which is what I used for converting
the models. This was a mistake on my part and I should have also updated
my transformers version to the official release.
I had checked the requirements/requirements-convert_legacy_llama.txt
file and noted that the version was >=4.45.1,<5.0.0 and came to the
conculusion that no updated would be needed, this assumed that
Embedding Gemma would be in a transformers release at the time
Commit fb15d649ed ("llama : add support
for EmbeddingGemma 300m (#15798)) was merged. So anyone wanting to
convert themselves would be able to do so. However, Embedding Gemma is
a preview release and this commit updates the requirements to use this
preview release.
* resolve additional python dependencies
* fix pyright errors in tokenizer test and remove unused import
* Add DeepSeek V3.1 thinking mode support
- Added COMMON_CHAT_FORMAT_DEEPSEEK_V3_1 enum value
- Created common_chat_params_init_deepseek_v3_1() function (currently uses R1 implementation)
- Created common_chat_parse_deepseek_v3_1() function that handles V3.1 thinking format:
- Extracts reasoning content before '</think>' tag into reasoning_content
- Extracts regular content after '</think>' tag into content
- No opening '<think>' tag in V3.1 format
- Added detection logic for V3.1 templates based on pattern: 'message['prefix'] is defined and message['prefix'] and thinking'
- Added V3.1 case to parsing switch statement
This addresses the issue where V3.1 outputs reasoning content followed by '</think>' and then regular content without the opening '<think>' tag.
* Another attempt by V3.1 non-thinking
* Fix test, but it's not asserting anything.
* Ignore vim swap files in tests dir
* Update the test
* Try using try_find_literal instead of regex
* passing test
* Revert "Try using try_find_literal instead of regex"
This reverts commit c50d887ec2.
* Remove unnecessary change
* Remove comment
* Add code to handle non-thinking mode.
* Try to set message['prefix'] when thinking is enabled.
* This fixes reasoning, but breaks normal content. We need state in the
chat parser.
* DeepSeek V3.1 thinking is now the default. Disable with `--reasoning-budget 0`.
* Simplify (DeepSeek V3.1 reasoning)
* Fix sign inversion bug
* Add some tool calling code (not working).
* Tool calls working in non-reasoning mode.
* Attempt a unit test for tool call parsing.
* Passing test
* Add tests for both happy path and broken fenced DeepSeek V3.1 tool call variants.
* Passing DeepSeek V3.1 tool call tests, but model is not working.
* Revert assistance response prefill change. Not my monkeys.
* Add fenced_thinking unit test variant. Passes, but thinking tool calling
still isn't working for some reason.
* Tests pass in reasoning mode. Also e2e tool test passes.
* Make a copy of the parse_json_tool_calls function for deepseek-v3.1 so
as to not accidentally introduce regressions.
* Fix thinking_forced_open logic. tool calling broken. Need to add another
test case.
* That's what I get for cargo culting a newline.
* Add multi tool call test for deepseek v3.1 non-reasoning
* Move test, remove .gitignore change
* Place deepseek-v3.1 reasoning test directly into existing reasoning
function per CISC's request.
* Address whitespace CI failure.
* Merge two assert_equals per CISC's request.
* Add DeepSeek-V3.1 tests to tests/test-chat.cpp per CISC's request.
* Merge deepseek V3.1 and regular parse_json_tool_calls() function
behaviors by adding optional update_cursor argument.
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* DeepSeek V3.1 fix reasoning_format none
* Strip grammar down to strictly what we expect based on model card. Throw
out parts we cargo culted from R1 that don't make sense.
* Update tests/test-chat-parser.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* DeepSeek V3.1 - Add edge case where thinking is forced open, there is
tool calling in the reasoning content, but then the model just stops the
output without closing the </think> tag, so it's not a partial. In this
case, use the tool call in the reasoning content.
* DeepSeek V3.1 - simplify update_cursor
* Update common/chat.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update common/chat.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update common/chat.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Fix indent
---------
Co-authored-by: openhands <openhands@all-hands.dev>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* ggml: allow casting between f32 and i32
* fix cuda
* add vulkan
* fix CPU non-cont
* add non-cont test case
* add note
* extend test number range
* correct note
* add cont version for vulkan
This commit adds two new command-line options to the
test-backend-ops.cpp that allow users to list all available GGML
operations and to show test coverage of these operations.
The motivation for this is that it can be useful to quickly see which
operations are currently covered by tests and which are not. Also it
migth be useful when using the `support` mode.
* sampling : optimize sorting using bucket sort in more places
ggml-ci
* sampling : do not sort in dist sampler
ggml-ci
* sampling : avoid heap allocations for sort buffers
ggml-ci
* common : add option to sort sampling candidates by probability
ggml-ci
* sampling : revert the change for preserving sort buffers
* sampling : use std::copy instead of memcpy
* sampling : clarify purpose of partial sort helpers
ggml-ci
* cont : remove wrong comment [no ci]
* common : update comment
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* metal : mul_mm_id remove hdst
* metal : remove mul_mm_id hsrc1
* metal : mul_mm_id simplify + add test
* metal : opt mul_mm_id map0
* metal : optimize mul_mm_id id gathering
* metal : mul/div opt
* metal : optimize mul_mm_id_map0
ggml-ci
The scalar FA shader already handled multiples of 8. The coopmat1 FA
shader assumed 16x16x16 and the shared memory allocations need the HSK
dimensions padded to a multiple of 16. NVIDIA's coopmat2 implementation
requires multiples of 16 for N and K, and needs the matrix dimensions
padded and loads clamped.
Store the FA pipelines in a map, indexed by the pipeline state.
* vulkan: optimize rms_norm, and allow the work to spread across multiple SMs
There are really two parts to this change:
(1) Some optimizations similar to what we have in soft_max, to unroll with
different numbers of iterations.
(2) A fusion optimization where we detect add followed by rms_norm, and make
the add shader atomically accumulate the values^2 into memory. Then the
rms_norm shader can just load that sum. This allows the rms_norm to be
parallelized across multiple workgroups, it just becomes a simple per-element
multiply.
The fusion optimization is currently only applied when the rms_norm is on a
single vector. This previously always ran on a single SM. It could apply more
broadly, but when there are other dimensions the work can already spread across
SMs, and there would be some complexity to tracking multiple atomic sums.
* Change add+rms_norm optimization to write out an array of partial sums
rather than using atomic add, to make it deterministic. The rms_norm
shader fetches a subgroup's worth in parallel and uses subgroupAdd to
add them up.
* complete rebase against fused adds - multi_add shader can also compute partial sums
* fix validation errors
* disable add_rms_fusion for Intel due to possible driver bug
* resolve against #15489, sync after clearing partial sums
* vulkan : support ggml_mean
* vulkan : support sum, sum_rows and mean with non-contiguous tensors
* vulkan : fix subbuffer size not accounting for misalign offset
* tests : add backend-op tests for non-contiguous sum_rows
* cuda : require contiguous src for SUM_ROWS, MEAN support
* sycl : require contiguous src for SUM, SUM_ROWS, ARGSORT support
* require ggml_contiguous_rows in supports_op and expect nb00=1 in the shader
* vulkan: Reuse conversion results in prealloc_y
Cache the pipeline and tensor that were most recently used to fill prealloc_y,
and skip the conversion if the current pipeline/tensor match.
* don't use shared pointer for prealloc_y_last_pipeline_used
- 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.
* 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
* model : add harmony parser for gpt-oss
* gpt-oss : fix grammar trigger from causing empty stack
* gpt-oss: tweak the grammar trigger again
* gpt-oss : add support for recipient in role header
* gpt-oss : fix ungrouped tool calls in grammar
* gpt-oss : loosen function name matching during parse
* gpt-oss : clean up workarounds
* gpt-oss : add template tests
* gpt-oss : simulate thinking and tool call tags
* gpt-oss : undo think tags when reasoning_format is none
* gpt-oss : set special tokens back to user defined
* gpt-oss : update openai-gpt-oss template
* server : filter out harmony thought messages
* gpt-oss : simplify parsing
* examples/finetune -opt SGD (stochastic gradient descent) memory opt
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 <picard12@live.de>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* 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
* Extend test case filtering
1. Allow passing multiple (comma-separated?) ops to test-backend-ops. This can be convenient when working on a set of ops, when you'd want to test them together (but without having to run every single op). For example:
`test-backend-ops.exe test -o "ADD,RMS_NORM,ROPE,SILU,SOFT_MAX"`
2. Support full test-case variation string in addition to basic op names. This would make it easy to select a single variation, either for testing or for benchmarking. It can be particularly useful for profiling a particular variation (ex. a CUDA kernel), for example:
`test-backend-ops.exe perf -b CUDA0 -o "MUL_MAT(type_a=f16,type_b=f32,m=4096,n=512,k=14336,bs=[1,1],nr=[1,1],per=[0,1,2,3],v=2)"`
These two can be combined. As the current `-o`, this change doesn't try to detect/report an error if an filter doesn't name existing ops (ex. misspelled)
* Updating the usage help text
* Update tests/test-backend-ops.cpp
* ggml/ggml-vulkan/test-backend-ops: adds CONV_2D for Vulkan
* ggml-vulkan: adds f32 scalar shader to compute 2D convolution directly
with gemm (no need for im2col),
* test-backend-ops: adds test_case_ref to check the validity/performance of ops
against reference implementations having different graphs, adds tests
* * Performance fixes: minimized branch divergence, uses collectives to
eliminate redundant calculation, macros removed.
* Kernel shared memory size check
* Updates test-backend-ops to support graphs for performance
measurement.
* * Apple/Win32 compile errors fixed
* Subgroup size used to determine tile size -> fixes llvmpipe errors.
* Collectives disabled by default.
* Intel support is disabled as the performance is poor.
* Conv2d enabled for Intel with disabled collectives, disabled for Apple
* test-backend-ops modifications are reverted
* Trailing spaces and missing override fixed.
* Triggering pipeline relaunch.
* Code formatted with .clang-format.
* ggml : add ggml_scale_bias
* ggml_vec_mad1_f32
* add more simd
* add CUDA
* sycl
* vulkan
* cann (placeholder)
* opencl
* will this fix cpu?
* fix cuda
* suggestions from coderabbit
* fix cann compile error
* vDSP_vsmsa
* rm __ARM_FEATURE_SVE
* use memcpy for op params
* make code looks more consistent
* use scalar for __ARM_FEATURE_SVE
* add x param to ggml_vec_mad1_f32
The fused operation was grabbing the epsilon value from the wrong place.
Add an env var to disable fusion.
Add some missing checks for supported shapes/types.
Handle fused rms_norm+mul in check_results.
* llama : initial Mamba-2 support
* ggml : SIMD ggml_ssm_scan for Mamba-2
* ggml : improve ggml_mul speed when masking recurrent states
* llama : support running Mamba-Codestral-7B-v0.1
* llama : fix Mamba-2 conv state saving
* ggml : make the ggml_mul fast broadcast path more consistently formatted
* llama : remove unused variable
* llama : add missing break
* convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present
The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.
* llama : avoid redundant state copy for Mamba 1 and 2
* metal : attempt to adapt SSM_SCAN for Mamba-2
* metal : fix SSM_SCAN pipeline scope
* metal : use log and exp instead of log1pf and expf in SSM_SCAN
* metal : remove unused arguments for SSM_SCAN
The max index is 31, so trimming the arguments is necessary.
* metal : add back n_seqs to SSM_SCAN args
Whoops, this is needed for the offset in the concatenated output.
* metal : fix SSM_SCAN state head offset
* metal : fix wrong number of tokens per sequence in SSM_SCAN
* ggml : remove unused fast broadcast path in GGML_MUL
This was initially added because states were masked with ggml_mul,
but this is no longer done and so this "optimisation" is no longer
necessary, or at least not worth the additional code complexity.
* ggml : avoid multiply by D in GGML_OP_SSM_SCAN
This makes the weight buft detection in src/llama.cpp simpler.
* convert : transpose Mamba-2 A, D and reshape SSM_NORM
This breaks existing conversions of Mamba-2 models
to avoid some reshapes.
Not sure if it's a good idea,
but it makes the graph slightly cleaner.
* llama : more appropriate SSM_SCAN and SSM_CONV buft support checks
* convert : fix flake8 lint
* metal : fix confusion between ; and ,
* metal : add missing args for nb references in ssm_scan_f32_group
* metal : single-user mamba2 inference works
* kv-cache : remove const_cast when setting inputs for s_copy
And also fix multi-user inference for recurrent models
by using cell_id instead of i as the kv cell index
when populating s_copy.
* convert : avoid AutoConfig for Mamba and Mamba2 hparams
* kv-cache : allow context shift for recurrent models
* graph : fix recurrent state copies when avoiding copies
Works, but using lambda functions might not be that clean.
* ggml : fix mamba2 ssm scan when compiled with SVE
* ggml-cpu : reorder SVE FMA for consistency with other SIMD arches
* cuda : implement ssm scan for Mamba2
There is still room for improvement, but it works!
* cuda : adapt Mamba1 ssm scan to shape changes from Mamba2
* mamba : fix mismatched new and delete size for llm_build_mamba
Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON
* cuda : graceful fallback for Mamba-1 models with weird embd size
* add "align corners" mode for bilinear upscale, and allow downscaling
* add ggml_interpolate, deprecate ggml_upscale_ext, pass in align-corners as bit-flag
* test-backend-ops: replace ggml_upscale_ext with ggml_interpolate, add test cases for downscale and align-corners
* implement unary REGLU/GEGLU/SWIGLU cpu ops
* relax constraints
* duplicate shape of source
* fix ggml_vec_geglu_f16
* special case gated ops
* implement unary REGLU/GEGLU/SWIGLU cuda ops
* tighten constraints again
* refactor into GGML_GLU_OP
* metal : add glu kernels
ggml-ci
* add CUDA_GLU_BLOCK_SIZE [no ci]
* more constraints and use 64bit ints
ggml-ci
* 64bit multiplication [no ci]
* implement swapped variants (cpu/cuda)
* update comment [no ci]
ggml-ci
* Vulkan: Add GLU ops and shaders
* SYCL: Implement fused kernel GEGLU, SWIGLU and REGLU for single up+gate
* ggml : implement GLU for split up/gate (#14181)
* implement GLU for split up/gate
* add tests for ggml_glu_split
* Vulkan: Implement glu_split logic and shader support
* add split to logging [no ci]
* SYCL: refactor element_size ops and add split up and gate support to gated kernels
* SYCL: switch GEGLU to use tanh approximation
---------
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
* GGML: increase OP count in assertion
* Refactor: Optimize SYCL element-wise operations with unary function inlining
This commit refactors the SYCL element-wise operations to improve performance by:
- Inlining unary operations (sgn, abs, elu, gelu, silu, etc.) to reduce kernel launch overhead.
- Introducing helper functions `op_xxx` for each unary operation to encapsulate the logic.
- Replacing direct kernel calls with calls to these inlined functions.
- Using `__dpct_inline__` to encourage compiler inlining.
- Minor code cleanup and consistency improvements.
The changes aim to reduce kernel launch overhead and improve the overall efficiency of element-wise operations on SYCL devices.
* vulkan: Increase workgroup size for GLU, for performance (#14345)
* vulkan: Increase workgroup size for GLU, for performance
* vulkan: change GLU shaders to do one element per invocation rather than one row per workgroup
* merge fix
* metal : add support for split and swap
ggml-ci
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* vulkan: Add fusion support for RMS_NORM+MUL
- Add a use_count to ggml_tensor, so we can detect if an output is used more than once.
- Change the ggml-vulkan rms_norm shader to optionally multiply by another tensor.
- Add detection logic and basic fusion logic in ggml-vulkan.
- Add some testing support for fusion. Rather than computing one node at a time, allow
for computing the whole graph and just testing one node's results. Add rms_norm_mul tests
and enable a llama test.
* extract some common fusion logic
* fix -Winconsistent-missing-override
* move ggml_can_fuse to a common function
* build fix
* C and C++ versions of can_fuse
* move use count to the graph to avoid data races and double increments when used in multiple threads
* use hash table lookup to find node index
* change use_counts to be indexed by hash table slot
* minimize hash lookups
style fixes
* last node doesn't need single use.
fix type.
handle mul operands being swapped.
* remove redundant parameter
---------
Co-authored-by: slaren <slarengh@gmail.com>
* CUDA: add bf16 and f32 support to cublas_mul_mat_batched
* Review: add type traits and make function more generic
* Review: make check more explicit, add back comments, and fix formatting
* Review: fix formatting, remove useless type conversion, fix naming for bools
* ggml : add ggml_set_rows
Add ggml_set_rows(a, b, c) which copies rows from 'b' into 'a' using
indices from 'c'.
ref: #8366
* use I64 for indices
* ggml : add repeat impl for i64
* ggml : add ggml_is_contiguous_rows
* ggml : ggml_set_rows support broadcast
* ggml : ggml_set_rows support quantized dst
ggml-ci
* ggml : support GGML_TYPE_F32 ".from_float" trait
* ggml : ggml_set_rows update comment + better index name
* tests : add ggml_set_rows
* metal : add ggml_set_rows implementation
ggml-ci
* ggml : simplify forward_dup_f32
* ggml : fix supports_op
* tests : add comment to set_rows
* ggml : leave the repeat_i64 for a separate PR
ggml-ci
* ggml : set_rows use std::min instead of MIN
* ggml : better error message for set_rows unsupported type
* metal : perform op->type check only once
* tests : more consistent implementation + more tests
ggml-ci
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama : add thread safety test
* llamafile : remove global state
* llama : better LLAMA_SPLIT_MODE_NONE logic
when main_gpu < 0 GPU devices are not used
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* llama : allow building all tests on windows when not using shared libraries
* add static windows build to ci
* tests : enable debug logs for test-chat
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* * ggml-vulkan: adds op CONV_TRANSPOSE_1D
* test-backend-ops: adds more spohisticated tests for CONV_TRANSPOSE_1D
* Missing barrier added to shader.
Number of additional tests reduced to 108.
* * Fixes typo in variable name.
* Removes extra whitespaces.
* Adds int64->int32 casts to prevent possible warnings.
* Problem size reduced in tests to pass tests with llvmpipe.
* supports_op condition moved from unintended position
* sampling : min-p should always return at least one token
ggml-ci
* sampling : same for typical sampling
* tests : sampling tests use min_keep == 0
ggml-ci
* add preludes to content on partial regex match
* allow all parsers to parse non-tool-call content.
* tweak order of <|python_tag|> vs <function= parsing for functionary v3.1 format. still not ideal but hopefully less prone to crash
* fix deltas of tool_call.function.name
* fix tool_call.id (was in tool_call.function.id!) + add function type
* add tool_call.type
* populate empty tool_call.function.arguments on first delta
* llama/ggml: add LLM training support
more compact progress bar
llama_save_model_to_file
llama_opt_param_filter
ggml_graph_dup force_grads
refactor ggml_opt, fix test-opt
* remove logits_all
* refactor CUDA implementation for ACC
* reset graph at beginning of opt period
* cmake : do not include ./src as public for libllama
ggml-ci
* cmake : rework tests
ggml-ci
* llguidance : remove unicode include
ggml-ci
* cmake : make c++17 private
ggml-ci
* graph : make mla compatible with FA
* metal : add exp FA kernels for DeepSeek models
ggml-ci
* llama : minor naming updates
ggml-ci
* ggml : disable FA for DS head sizes
* tests : add FA tests for MLA shapes
ggml-ci
The grouped query attention optmization doesn't require a power of two ratio,
the only thing relying on it was the modulo operation written as bitwise &.
split_k need not depend on gqa_ratio - enable it any time there's only one
workgroup in the X dimension. The shader gets the split index from the x coord,
and multiple workgroups in the X dimension (pre-split) indicates a larger
FA operation that wouldn't need splitting.
When using group query attention, we have one workgroup per KV batch and this
can be very few workgroups (e.g. just 8 in some models). Enable split_k to
spread the work across SMs. This helps a lot when the KV cache is large.
* (wip) refactor downloading system [no ci]
* fix all examples
* fix mmproj with -hf
* gemma3: update readme
* only handle mmproj in llava example
* fix multi-shard download
* windows: fix problem with std::min and std::max
* fix 2
* ggml : FA with different K, V head sizes (CPU)
ggml-ci
* metal : add FA with HS=192
* metal : extend FA to support different K and V head sizes
ggml-ci
* metal : add FA vector kernels for heads K 192 and V 128
ggml-ci
* ggml : restrict op on other backends to equal head sizes
ggml-ci
* metal : optimize FA-vec kernel
ggml-ci
* metal : FA remove mq registers
* metal : improve MoE mul_mat_id condition
ggml-ci
* metal : fix comments + remove unnecessary addition
ggml-ci
* metal : avoid too much shared memory usage with mul_mat_id
ggml-ci
The OOB calculation could be wrong if the last iteration was during one of
the unrolled loops. Adjust the unrolling counts to avoid this. Add a couple
new backend tests that hit this failure on NVIDIA GPUs.
* tests: add mul_mat perf/functional tests for p021/nc vulkan shaders
* vulkan: Optimize mul_mat_vec p021 and nc shaders.
These shaders are used in attention calculations, and when the KV cache grows
large they start to dominate the run time. For the nc shader (which is called
with large 'k' dimension), use unrolling and vector loads. For the p021 shader
(which is called with large 'm' and small 'k' dimensions), take advantage of
grouped query attention to reuse loads from the A matrix for the whole group,
and reduce the number of workgroups (too much overhead from tiny dispatches).
Using subgroupAdd in the p021 shader also helps, use that conditionally.
- Find out active blocks per SM using cudaOccupancyMaxActiveBlocksPerMultiprocessor API. Use this value to determine the optimal parallel_blocks value.
- Prefer vector flash attention kernels over MMA kernel for BS=1
Fixes Issue: #12182
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Eve
Acly
Sigbjørn Skjæret
Shin-myoung-serp
Ruben Ortlam
shun095
Xuan-Son Nguyen
Bowen Han
Reese Levine
Georgi Gerganov
Oliver Simons
Daniel Bevenius
Jesse
Jeff Bolz
Aman Gupta
Aldehir Rojas
Piotr Wilkin (ilintar)
leejet
rmatif
Diego Devesa
Johannes Gäßler
Jonathan Graehl
Sachin Desai
Jhen-Jie Hong
Leonard Mosescu
Erik Scholz
Ervin Áron Tasnádi
Tarek Dakhran
compilade
R0CKSTAR
Vedran Miletić
Radoslav Gerganov
Olivier Chafik
Olivier Chafik
David Huang
DocShotgun
piDack
matteo
frob
Plamen Minev
R0CKSTAR
Sergei Vorobyov
Michał Moskal
Gaurav Garg
Molly Sophia