Georgi Gerganov [Tue, 30 Apr 2024 09:16:08 +0000 (12:16 +0300)]
ggml : add Flash Attention (#5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (#6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (#6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <redacted> Co-authored-by: Pierrick HYMBERT <redacted>
* Cleaning up integration tests to share code between tests and make it simpler to add new tests.
* Add tests around quantifiers to ensure both matching and non-matching compliance.
* Add slightly more complex grammar with quantifiers to test references with quantifiers.
* Fixing build when C++17 is not present.
* Separating test calls to give more helpful stack traces on failure. Adding verbose messages to give visibility for what is being tested.
* Adding quotes around strings to explicitly show whitespace
* Removing trailing whitespace.
* Implementing suggestions from @ochafik -- grammars and test strings now print and flush before tests to aid in debugging segfaults and whatnot.
* Cleaning up forgotten symbols. Modifying simple test to use test harness. Added comments for more verbose descriptions of what each test is accomplishing.
* Unicode symbol modifications to hopefully make log easier to parse visually.
Daniel Bevenius [Thu, 25 Apr 2024 12:38:14 +0000 (14:38 +0200)]
clip : rename lerp function to avoid conflict (#6894)
This commit renamesthe lerp (linear interpolation) function in clip.cpp
to avoid a conflict with the lerp function in the <cmath> standard C++
library when using c++20.
The motivation for this change is to enable projects that use c++20 to
be able to compile clip.cpp without having to resort to patching it. The
lerp function was added to cmath in version C++20 (202002L) and is why
this is not causing any issue at the moment as C++11/C++17 is currently
used by llama.cpp.
I realize that llama.cpp uses either C++11 (or C++17 in the case for
SYCL) but wanted to ask if this would be an acceptable change just the
same.
common : revert showing control tokens by default for server (#6860)
* fix: revert showing control tokens by default
* feat: revert changes to default behavior of llama_token_to_piece; provide overridden declaration to receive "bool special" param to toggle showing control tokens
* feat: use the overridden declaration of llama_token_to_piece from common/common.cpp to specify "false" so that control tokens are not shown in chat completion responses"
- Re-enable by default
- Fix issue described in #6716
- Make code more abstract, elegant, and maintainable
- Faster handling of weirdly shaped `m` an `n` edge cases
Dave Airlie [Mon, 22 Apr 2024 14:05:06 +0000 (00:05 +1000)]
ggml : fix calloc argument ordering. (#6820)
Latest gcc complains here:
/home/airlied/devel/llama.cpp/ggml-alloc.c: In function ‘ggml_gallocr_new_n’:
/home/airlied/devel/llama.cpp/ggml-alloc.c:374:59: warning: ‘calloc’ sizes specified with ‘sizeof’ in the earlier argument and not in the later argument [-Wcalloc-transposed-args]
374 | ggml_gallocr_t galloc = (ggml_gallocr_t)calloc(sizeof(struct ggml_gallocr), 1);
| ^~~~~~
/home/airlied/devel/llama.cpp/ggml-alloc.c:374:59: note: earlier argument should specify number of elements, later size of each element
and a bunch more.
calloc is specified to take nmemb first then size, so realign the code.
In a couple of places there was a * x, 1 so I fixed those to use calloc properly.
Co-authored-by: Samuel Tallet <redacted>
* Update llama.cpp
Co-authored-by: Samuel Tallet <redacted>
* Update tests/test-chat-template.cpp
Co-authored-by: Samuel Tallet <redacted>
* Added EOS stop sequence according to https://github.com/ggerganov/llama.cpp/pull/6751#issuecomment-2065602862
* Removed adding of BOS token before first message
* Removed bos token from expected output from llama-3
using autoawq to quantize gemma model will include a lm_head.weight tensor in model-00001-of-00002.safetensors. it result in this situation that convert-hf-to-gguf.py can't map lm_head.weight. skip loading this tensor could prevent this error.
* change code to full string match and print necessary message
change code to full string match and print a short message to inform users that lm_head.weight has been skipped.
This change upstreams llamafile's cpu matrix multiplication kernels
which improve image and prompt evaluation speed. For starters, Q4_0
and Q8_0 weights should go ~40% faster on CPU. The biggest benefits
are with data types like f16 / f32, which process prompts 2x faster
thus making them faster than quantized data types for prompt evals.
This change also introduces bona fide AVX512 support since tinyBLAS
is able to exploit the larger register file. For example, on my CPU
llama.cpp llava-cli processes an image prompt at 305 tokens/second,
using the Q4_K and Q4_0 types, which has always been faster than if
we used f16 LLaVA weights, which at HEAD go 188 tokens/second. With
this change, f16 LLaVA performance leap frogs to 464 tokens/second.
On Intel Core i9-14900K this change improves F16 prompt perf by 5x.
For example, using llama.cpp at HEAD with Mistral 7b f16 to process
a 215 token prompt will go 13 tok/sec. This change has fixes making
it go 52 tok/sec. It's mostly thanks to my vectorized outer product
kernels but also because I added support for correctly counting the
number of cores on Alderlake, so the default thread count discounts
Intel's new efficiency cores. Only Linux right now can count cores.
This work was sponsored by Mozilla who's given permission to change
the license of this code from Apache 2.0 to MIT. To read more about
what's improved, and how it works, see: https://justine.lol/matmul/
Daniel Bevenius [Tue, 16 Apr 2024 06:13:13 +0000 (08:13 +0200)]
gguf : add special tokens metadata for FIM/Infill (#6689)
This commit adds special token metadata for Fill-In-the-Middle
(FIM)/Infill to the GGUF model.
The motivation for this is that currently there is support for CodeLlama
but other models exist now like CodeGemma, but the different models use
different token ids for the special tokens and this commit allows for
supporting multiple models.
overview / pkg / ggml / sources / llama.cpp / log