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 <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>

1 files changed, 7 insertions, 0 deletions

diff --git a/common/common.cpp b/common/common.cpp
index 099d0356..243b88ab 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -947,6 +947,10 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
         params.cont_batching = true;
         return true;
     }
+    if (arg == "-fa" || arg == "--flash-attn") {
+        params.flash_attn = true;
+        return true;
+    }
     if (arg == "--color") {
         params.use_color = true;
         return true;
@@ -1513,6 +1517,7 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
     printf("  -ns N, --sequences N  number of sequences to decode (default: %d)\n", params.n_sequences);
     printf("  -ps N, --p-split N    speculative decoding split probability (default: %.1f)\n", (double)params.p_split);
     printf("  -cb, --cont-batching  enable continuous batching (a.k.a dynamic batching) (default: disabled)\n");
+    printf("  -fa, --flash-attn     enable Flash Attention (default: %s)\n", params.flash_attn ? "enabled" : "disabled");
     printf("  --mmproj MMPROJ_FILE  path to a multimodal projector file for LLaVA. see examples/llava/README.md\n");
     printf("  --image IMAGE_FILE    path to an image file. use with multimodal models. Specify multiple times for batching\n");
     if (llama_supports_mlock()) {
@@ -1885,6 +1890,7 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
     cparams.cb_eval           = params.cb_eval;
     cparams.cb_eval_user_data = params.cb_eval_user_data;
     cparams.offload_kqv       = !params.no_kv_offload;
+    cparams.flash_attn        = params.flash_attn;
 
     cparams.type_k = kv_cache_type_from_str(params.cache_type_k);
     cparams.type_v = kv_cache_type_from_str(params.cache_type_v);
@@ -2707,6 +2713,7 @@ void dump_non_result_info_yaml(FILE * stream, const gpt_params & params, const l
     fprintf(stream, "seed: %u # default: -1 (random seed)\n", params.seed);
     fprintf(stream, "simple_io: %s # default: false\n", params.simple_io ? "true" : "false");
     fprintf(stream, "cont_batching: %s # default: false\n", params.cont_batching ? "true" : "false");
+    fprintf(stream, "flash_attn: %s # default: false\n", params.flash_attn ? "true" : "false");
     fprintf(stream, "temp: %f # default: 0.8\n", sparams.temp);
 
     const std::vector<float> tensor_split_vector(params.tensor_split, params.tensor_split + llama_max_devices());