Commit Graph

221 Commits

Author SHA1 Message Date
zhouwg
109148ac84 ggml : fix typo in ggml.c (llama/7603) 2024-06-16 18:19:48 +03:00
Georgi Gerganov
812787cbc5 ggml : generalize GGML_OP_CONCAT (llama/7563)
* ggml : generalize GGML_OP_CONCAT (WIP)

ggml-ci

* tests : add dim != 2 tests

* metal : generalize concat kernel

* tests : naming

* cuda : generalize concat kernel

ggml-ci

* sycl : add warning and assert

* ggml : fix op params handling

* metal : bugfix kernel

ggml-ci

* ggml : reimplement CPU and Metal

* cuda : add asserts

ggml-ci

* ggml : fix ptrs

ggml-ci
2024-06-16 18:19:48 +03:00
Georgi Gerganov
5ee048eb67 ggml : restore ggml_rope_xpos_inplace (ggml/0)
ggml-ci
2024-06-16 18:19:48 +03:00
Masaya, Kato
37ed71c964 ggml: aarch64: SVE kernels for q8_0_q8_0, q4_0_q8_0 vector dot (llama/7433)
* Add SVE support for q4_0_q8_0 q8_0_q8_0

* remove ifdef
2024-06-16 18:19:48 +03:00
Georgi Gerganov
04a3279320 ggml : remove ggml_flash_attn and ggml_flash_ff (llama/7463)
ggml-ci
2024-06-16 18:19:48 +03:00
Georgi Gerganov
45ddda8e0c ggml : drop support for QK_K=64 (llama/7473)
* ggml : drop support for QK_K=64

ggml-ci

* opencl : restore QK_K=256 define
2024-06-16 18:19:48 +03:00
Georgi Gerganov
c4d6958b3e cuda : fix rope + add tests (llama/7452)
* cuda : fix rope pos data

ggml-ci

* ggml : drop mode & 1 == 1 support for ggml_rope

ggml-ci

* ggml : support freq_factors for f16 rope (CPU)

ggml-ci

* tests : add rope tests using frequency factors

ggml-ci
2024-06-16 18:19:48 +03:00
liuwei-git
c9dcb75118 llama : add phi3 128K model support (llama/7225)
* add phi3 128k support in convert-hf-to-gguf

* add phi3 128k support in cuda

* address build warnings on llama.cpp

* adjust index value in cuda long rope freq factors

* add long rope support in ggml cpu backend

* make freq factors only depend on ctx size

* remove unused rope scaling type 'su' frin gguf converter

* fix flint warnings on convert-hf-to-gguf.py

* set to the short freq factor when context size is small than trained context size

* add one line of comments

* metal : support rope freq_factors

* ggml : update ggml_rope_ext API to support freq. factors

* backends : add dev messages to support rope freq. factors

* minor : style

* tests : update to use new rope API

* backends : fix pragma semicolons

* minor : cleanup

* llama : move rope factors from KV header to tensors

* llama : remove tmp assert

* cuda : fix compile warning

* convert : read/write n_head_kv

* llama : fix uninitialized tensors

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-06-16 18:19:48 +03:00
junchao-loongson
eb26f55b40 ggml : add loongarch lsx and lasx support (llama/6454)
* add loongarch lsx and lasx optimize code

* Add loongarch compilation support to makefile

* revert stb_image.h

* opt bytes_from_nibbles_32 and sum_i16_pairs_float

* fix undeclared

* format code

* update

* update 2

---------

Co-authored-by: Jinyang He <hejinyang@loongson.cn>
2024-06-16 18:19:48 +03:00
Srihari-mcw
eb2b086584 Add provisions for windows support for BF16 code including CMake provision for enabling AVX512_BF16 (llama/7258) 2024-06-16 18:19:48 +03:00
Johannes Gäßler
2b07dc3186 ggml: implement quantized KV cache for FA (llama/7372) 2024-06-16 18:19:48 +03:00
Georgi Gerganov
705fe30a02 android : use "ci-android" branch for CI (llama/7341)
* android : use "ci-android" branch for CI

* ggml : disable SIMD exp and silu for 32-bit ARM

ggml-ci

* android : do not fetch, use add_subdirectory instead

* cmake : provide binary dir
2024-06-16 18:19:48 +03:00
Justine Tunney
574661f2e6 ggml : rewrite silu and softmax for cpu (llama/7154)
This change upstreams llamafile's vectorized expf() functions. This lets
us compute softmax and silu more accurately than the short[65536] lookup
table that GGML previously used to make this operation go faster. We can
support aarch64 and sse2+ with the worst case rounding error of 2ulp. It
makes make -j8 tests && ./tests/test-backend-ops -o SOFT_MAX -b CPU perf
go 1.5x faster for SSE2+FMA, 1.9x faster for AVX2+FMA and 2.1x on AVX512
2024-06-16 18:19:48 +03:00
kunnis
7178cceeaa ggml : use dynamic thread scheduling for matrix multiplication (llama/6915)
* Just reordering some structs.

* Adding in the calls to mm_pause

* Passing around the state

* Renaming and moving a bunch of variables around.

* Extracting the logic to it's own function.

* Moving some variable definitions into the chunk function.

* Moving some variables around

* moving src1_cont inside

* Moving row_size

* adding the current_chunk

* Reorg the code.

* Formatting to match the orig patch

* starting to setup the chunking variables

* Starting the buildup of the loop

* The yield shouldn't be necessary.

* adding the looping structure based on the chunk configuration.

* Add in the re-chunking code.

* Making it much more likely to rechunk.

* disable resizing if numa is enabled.

* Updating comments with what we've learned.

* Fix formatting

* Couple more formatting fixes.

* More style fixes.

* Fix Warnings

* Going with unused because there's conditional logic that needs it.

* Update ggml.c

* Update ggml.c

---------
2024-06-16 18:19:48 +03:00
slaren
37a72cb170 ggml : tag ggml_tensor::backend as deprecated (llama/7290) 2024-06-16 18:19:48 +03:00
John Balis
c4de1e19df ggml : add ggml_upscale_ext (ggml/814)
* initial commit with CPU implementation of upscale to shape and test, cuda implementation next

* experimental commit to see if dst shape is correct

* test version

* test

* removed unnecessary params

* refactor

* fixed tests

* ggml : metal impl + cleanup + sycl dev warnings

* patched ggml_upscale cuda op to handle non-contiguous tensors, added test for non-contiguous behavior

* metal : fix upsacle op to support nb00 + style

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-06-16 18:19:48 +03:00
Georgi Gerganov
1056ad762c
metal : support FA without mask + add asserts (llama/7278)
* ggml : fa without mask + add asserts

ggml-ci

* metal : support non-contiguous KV

ggml-ci
2024-05-14 19:16:29 +03:00
Georgi Gerganov
9506267ce5 ggml : try fix ppc64 (#0) 2024-05-13 11:02:26 +03:00
Georgi Gerganov
91c646c61d ggml : restore sigmoid decl order (ggml/0) 2024-05-13 11:02:26 +03:00
Georgi Gerganov
accada542a ggml : resolve merge (ggml/0)
ggml-ci
2024-05-13 11:02:26 +03:00
Georgi Gerganov
e54329da7b ggml : full ALiBi support (llama/7192)
* ggml : full ALiBi support

* ggml : update ggml_soft_max_ext() CUDA, SYCL

* ggml : ggml_flash_attn_ext() support ALiBi (CPU)

* ggml : ggml_flash_attn_ext() support ALiBi (Metal)

* ggml : fix warning

* ggml : ggml_flash_attn_ext() support ALiBi (CUDA)

ggml-ci

* ggml : fix assert message

* vulkan : add dev notes

* ggml : require mask when using ALiBi

ggml-ci

* convert : fix convert for refact models
2024-05-13 11:02:26 +03:00
Justine Tunney
a2ad810118 ggml : introduce bfloat16 support (llama/6412)
* Introduce bfloat16 support

Many models on Hugging Face (e.g. Mistral, TinyLLaMA) use bfloat16 as
their canonical floating point format.

      ┌sign
      │
      │   ┌exponent
      │   │
      │   │      ┌mantissa
      │   │      │
      │┌──┴───┐┌─┴───┐
    0b0000000000000000 brain16

This encoding has the same number of exponent bits as float32. That
makes conversion relatively straightforward, even in the absence of
hardware support. For example, converting brain16 to binary32 means
simply shifting 16 bits to the left.

      ┌sign
      │
      │   ┌exponent
      │   │
      │   │      ┌mantissa
      │   │      │
      │┌──┴───┐┌─┴───────────────────┐
    0b00000000000000000000000000000000 IEEE binary32

The issue is that converting bf16 to fp16 can result in information
loss. Only 13% of bf16 numbers can be precisely represented in fp16
which in practice ends up being 99.71% of Mistral 7b v0.2's weights
however there is currently no way other than fp32 to get the others

      ┌sign
      │
      │  ┌exponent
      │  │
      │  │    ┌mantissa
      │  │    │
      │┌─┴─┐┌─┴──────┐
    0b0000000000000000 IEEE binary16

This change fixes that, by adding a bf16 data type to GGML. Support
for CPU inference has been implemented along with optimizations for
the AVX2, AVX512, and AVX512BF16 ISAs. Perplexity on Mistral 7b 0.2
improves somewhere around -0.0024 to -0.0046 compared to using fp16

* Remove GGML code that's not needed

* Minimize the GGML API surface area for BF16

* Remove bf16 luts

* Make the GGML header look nicer

* Fix documentation

* Apply ggerganov's fixes for test-backend-ops

* Add BF16 code for new ggml_validate_row_data() function
2024-05-13 11:02:26 +03:00
Xuan Son Nguyen
9b84195225 gguf-split: add --no-tensor-first-split (llama/7072) 2024-05-13 11:02:26 +03:00
Georgi Gerganov
156a33a990 ggml : add Flash Attention (llama/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 (llama/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 (llama/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>
2024-05-13 11:02:26 +03:00
Xuan Son Nguyen
388c3462a6 gguf : enforce that tensor names are unique (llama/6905)
* not allow adding duplicated tensor name

* no duplicated tensor while reading gguf

* typo

* throw exception inside llama_model_loader

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
2024-05-13 11:02:26 +03:00
slaren
ecfac1e240 gguf : fix mismatch between alloc and free functions (llama/6929) 2024-05-13 11:02:26 +03:00
Georgi Gerganov
6f7140f568 Merge pull request from GHSA-p5mv-gjc5-mwqv
* always use calloc

clamp n_kv on failure to read a kv

* ggml : alternative ctx->header.n_kv update

---------

Co-authored-by: slaren <slarengh@gmail.com>
2024-05-13 11:02:26 +03:00
Georgi Gerganov
05b17112cf ggml : fix redefinition of vaddvq_f32 for 32-bit ARM (llama/6906) 2024-05-13 11:02:26 +03:00
Justine Tunney
6c3971b29b llamafile : improve sgemm.cpp (llama/6796)
* llamafile : improve sgemm.cpp

- 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

* Address review comments

* Help clang produce fma instructions

* Address review comments
2024-05-13 11:02:26 +03:00
slaren
c96b0a938e ggml : group all experts in a single ggml_mul_mat_id (llama/6505)
* ggml : group all experts in a single ggml_mul_mat_id
cuda : improve mmid row copy

* cuda : fix bin bcast with non-cont src0

* test-backend-ops : only run all mul mat tests for base types

* llama : disable moe offloading with SYCL

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-05-13 11:02:26 +03:00
Georgi Gerganov
c97796aa0f ggml : fix llamafile sgemm wdata offsets (llama/6710)
ggml-ci
2024-05-13 11:02:26 +03:00
Justine Tunney
7a4f7d825e ggml : add llamafile sgemm (llama/6414)
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/
2024-05-13 11:02:26 +03:00
slaren
00a0947c65 metal : unify mul_mv_id kernels (llama/6556) 2024-05-13 11:02:26 +03:00
jiez
60f3713026 llama : add gguf_remove_key + remove split meta during quantize (llama/6591)
* Remove split metadata when quantize model shards

* Find metadata key by enum

* Correct loop range for gguf_remove_key and code format

* Free kv memory

---------

Co-authored-by: z5269887 <z5269887@unsw.edu.au>
2024-05-13 11:02:26 +03:00
Justina Cho
37e6757453 feat: implemented sigmoid function (ggml/806)
* added sigmoid function

* implemented metal kernel for sigmoid

* implemented cuda kernel for sigmoid

* added sigmoid unary op and incremented count
2024-05-13 11:02:26 +03:00
Carolinabanana
526332873b
llama : add Command R Plus support (llama/6491)
* Add Command R Plus GGUF

* Add Command R Plus GGUF

* Loading works up to LayerNorm2D

* Export new tensors in 1D so they are not quantized.

* Fix embedding layer based on Noeda's example

* Whitespace

* Add line

* Fix unexpected tokens on MPS. Re-add F16 fix. ((Noeda)

* dranger003: Fix block index overflow in CUDA dequantizing.

* Reverted blocked multiplication code as it still has issues and could affect other Llama arches

* export norms as f32

* fix overflow issues during quant and other cleanup

* Type convention

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* dranger003: Fix more int overflow during quant.

---------

Co-authored-by: S <seast@Ss-Mac-Studio.local>
Co-authored-by: S <s@example.com>
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-09 20:26:18 +03:00
slaren
1dce94cf26
ggml : mul_mat_id use the same tensor for all the experts (llama/6387)
* ggml : update mul_mat_id to use the same tensor for all the experts

* update cuda

* minor

* update metal

* update test-backend-ops

* fix cuda

* Update ggml-metal.m

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* update convert.py

* update convert-hf-to-gguf.py

* update convert.py for mixtral hf models

* Update convert-hf-to-gguf.py

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* cuda : support non-pow-2 number of experts

* allow quantize to work for split and merged experts models in the same way

* cleanup + disable mmap automatically with split tensors models

* update imatrix

* test-backend-ops : test qwen argsort

* update grok model loading

* llama : add merged experts tensors to the grok tensor map

* minor

* gguf : bump version

* fix quantizing of merged experts

* convert-hf-to-gguf.py : update grok (untested)

* make linter happy

* cuda/argsort : use shared memory instead of pool memory

* convert : fix grok tensor names

* metal : add support for non-pow-2 argsort

* llama : more loader cleanup, better error checking

* cuda : fix warning

* llama : still use mmap for loading old models, but copy the data to a host buffer

* add review note

* llama : remove ffn tensor counting + add sanity check

ggml-ci

* convert : fix handling of n_experts == None

ggml-ci

* imatrix : fix ncall counters

* llama : produce error if imatrix size does not match

* quantize : terminate on errors + trace logs

ggml-ci

* metal : pad shared memory to 16 bytes

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-04-07 16:15:57 +03:00
0cc4m
fa966b9b40
Vulkan k-quant mmq and ggml-backend offload functionality (llama/6155)
* Fix Vulkan no kv offload incoherence

* Add k-quant mul mat mat shaders

* Rework working buffer allocation, reduces vram use noticeably

Clean up cpu assist code, replaced with ggml-backend offload function

* Default to all dedicated GPUs

* Add fallback for integrated GPUs if no dedicated GPUs are found

* Add debug info which device is allocating memory

* Fix Intel dequant issue

Fix validation issue

* Fix Vulkan GGML_OP_GET_ROWS implementation

* Clean up merge artifacts

* Remove Vulkan warning
2024-04-07 16:15:57 +03:00
slaren
3adbf2fb03
ggml : fix bounds checking of zero size views (llama/6347) 2024-04-07 16:15:56 +03:00
Georgi Gerganov
2948c740a2
sync : ggml (#2001)
* sync : update scripts

* sync : ggml

* talk-llama : sync llama.cpp

* make : WHISPER_CUBLAS -> WHISPER_CUDA

* ci : try to fix sycl build

* talk-llama : fix make build
2024-03-27 18:55:10 +02:00
slaren
8932c2d6ce
llama : add pipeline parallelism support (llama/6017)
* llama : add pipeline parallelism support for batch processing with multiple CUDA GPUs

ggml-ci

* server : add -ub, --ubatch-size parameter

* fix server embedding test

* llama : fix Mamba inference for pipeline parallelism

Tested to work correctly with both `main` and `parallel` examples.

* llama : limit max batch size to n_batch

* add LLAMA_SCHED_MAX_COPIES to configure the number of input copies for pipeline parallelism
default increase to 4 (from 2)

changing this value may improve performance for some systems, but increases memory usage

* fix hip build

* fix sycl build (disable cpy_tensor_async)

* fix hip build

* llama : limit n_batch and n_ubatch to n_ctx during context creation

* llama : fix norm backend

* batched-bench : sync after decode

* swiftui : sync after decode

* ggml : allow ggml_get_rows to use multiple threads if they are available

* check n_ubatch >= n_tokens with non-casual attention

* llama : do not limit n_batch to n_ctx with non-casual attn

* server : construct batch with size of llama_n_batch

* ggml_backend_cpu_graph_compute : fix return value when alloc fails

* llama : better n_batch and n_ubatch comment

* fix merge

* small fix

* reduce default n_batch to 2048

---------

Co-authored-by: Francis Couture-Harpin <git@compilade.net>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-15 14:01:13 +02:00
Michael Podvitskiy
d73a63629e
ggml, ci : Windows ARM runner and build fixes (llama/5979)
* windows arm ci

* fix `error C2078: too many initializers` with ggml_vld1q_u32 macro for MSVC ARM64

* fix `warning C4146: unary minus operator applied to unsigned type, result still unsigned`

* fix `error C2065: '__fp16': undeclared identifier`
2024-03-15 14:01:13 +02:00
Georgi Gerganov
2fef660d0a
ggml : remove old quantization functions (llama/5942)
* ggml : remove old quantization functions

ggml-ci

* ggml : simplify ggml_quantize_chunk

ggml-ci

* ggml : restrict correctness

ggml-ci

* ggml : remove hist data from the quantization API

ggml-ci

* tests : remove hist usage in test-backend-ops

ggml-ci

* vulkan : remove hist and fix typo
2024-03-15 14:01:12 +02:00
compilade
6e9d3aa32d
llama : support Mamba Selective State Space Models (llama/5328)
* mamba : begin working on support for Mamba SSM

* mamba : begin figuring out how to (ab)use the kv cache for Mamba

* mamba : recurrent inference almost works, but incoherent

* mamba : recurrent inference WORKS!!!

* convert : optionally use d_conv and d_state from config.json for Mamba

* mamba : refactor recurrent conv, resulting in 20% perf increase

It's still slower than I'd like, but I did not really optimize `ggml_exp` yet.

I also refactored `ggml_exp` to work with tensors with more than 2 dimensions.

* ggml : parallelize ggml_exp

This results in 8% faster token generation for Mamba-130M.

* mamba : simplify the conv step with a self-overlapping view

Turns out the conv_state can be made smaller by one column.
Note that this breaks existing GGUFs of Mamba,
because the key_value_length field is tied to the conv_state size.

Convolution with a self-overlapping view is cool!
And it's much simpler than what I initially thought would be necessary
to make the convolution step work with more than 1 token at a time.

Next step is to make the SSM step work on batches of tokens too,
and thus I need to figure out a way to make a parallel selective scan
which will keep the ssm_state small and won't make it bigger
by a factor of (n_layer * batch_size).

* llama : fix Mamba KV self size wrongly displaying as f16 instead of f32

Relatedly, I also tried to see if other types than f32 worked for the states,
but they don't, because of the operators used.
It's probably better anyway to keep lots of precision there,
since the states are small anyway.

* mamba : fix self-overlapping view depth stride

* mamba : handle batches of more than 1 token

This means running Mamba no longer crashes when using the default settings!
And probably also slightly faster prompt processing.
Both batched and non-batched processing yield the same output.

Previously, the state was not cleared when starting a sequence.
Next step is to make the KV cache API work as expected for Mamba models.

* ggml: add ggml_ssm_scan to help with parallel selective scan

If the selective scan was implemented without a custom operator,
there would be waaay too many nodes in the graph. For example,
for Mamba-130M, with a batch size of 512 (the default),
a naive selective scan could add at least 24*512=12288 nodes,
which is more than LLAMA_MAX_NODES (8192),
and that's only for the smallest Mamba model.
So it's much cleaner with a custom operator.
Not sure about the name, though.

* ggml : in ggml_ssm_scan, merge multiple rows in the same vec operation

This will help with performance on CPU if ggml_vec_mul_f32
and ggml_vec_add_f32 are ever optimized with SIMD.

* mamba : very basic quantization support

Mostly works, but there is currently no difference
between the variants of a k-quant (e.g. Q4_K_S and Q4_K_M are the same).
Most of the SSM-specific weights can be kept in f32 without affecting
the size that much, since they are relatively small.
(the linear projection weights are responsible for most of Mamba's size)

Too much quantization seems to make the state degrade quite fast, and
the model begins to output gibberish.
It seems to affect bigger models to a lesser extent than small models,
but I'm not sure by how much.

Experimentation will be needed to figure out which weights are more important
for the _M (and _L?) variants of k-quants for Mamba.

* convert : fix wrong name for layer norm weight of offical Mamba models

I was using Q-bert/Mamba-* models before, which have a slighlty different
naming scheme for the weights.
(they start with "model.layers" instead of "backbone.layers")

* mamba : fuse more steps of the SSM scan in the ggml_ssm_scan operator

This increases performance on CPU by around 30% for prompt processing,
and by around 20% for text generation.

However, it also makes the ggml_exp and ggml_soft_plus operators unused.
Whether or not they should be kept will be decided later.

* convert : for Mamba, also consider the "MambaLMHeadModel" arch name

It's the name of the class of the official implementation,
though they don't use it (yet) in the "architectures" field of config.json

* mamba : fix vocab size problems with official models

The perplexity was waaaay to high for models with a non-round vocab size.
Not sure why, but it needed to be fixed in the metadata.

Note that this breaks existing GGUF-converted Mamba models,
but **only if** the vocab size was not already rounded.

* ggml : remove ggml_exp and ggml_soft_plus

They did not exist anyway outside of this branch,
and since ggml_ssm_scan fused operations together, they are unused.
It's always possible to bring them back if needed.

* mamba : remove some useless comments

No code change.

* convert : fix flake8 linter errors

* mamba : apply suggestions from code review

* mamba : remove unecessary branch for row-wise ssm_state and C multiplication

It was previously done to avoid permuting when only one token is processed
at a time (like when generating text), but permuting is cheap,
and dynamically changing the compute graph is not future-proof.

* ggml : in ggml_ssm_scan, use more appropriate asserts

* ggml : rename the destination pointer in ggml_compute_forward_ssm_scan_f32

* mamba : multiple sequences, but one at a time

This is a step towards making this Mamba implementation usable
with the server example (the way the system prompt is kept when clearing
the client slots will need to be changed before this can work, though).

The KV cache size for this kind of model is tied to the maximum number
of sequences kept at any single time.
For now, this number is obtained from n_parallel (plus one,
to have an extra sequence to dedicate to the system prompt),
but there might be a better way to do this which won't also
make the main example use 2 cells even if only 1 is really used.
(for this specific case, --parallel 0 helps)

Simultaneous sequence processing will probably require changes to
ggml_ssm_scan, and possibly a new operator for the conv step.

* mamba : support llama_kv_cache_seq_cp

This (mis)uses the logic around K shifts, because tokens in a state
can't be shifted anyway, and because inp_K_shift has the right shape and type.
Using ggml_get_rows is a nice way to do copies, but copy chains can't work.
Fortunately, copy chains don't really seem to be used in the examples.

Each KV cell is dedicated to the sequence ID corresponding to its own index.

* mamba : use a state mask

It's cleaner than the previous heuristic of
checking for the pos of the first token in the batch.

inp_KQ_mask could not be re-used for this, because it has the wrong shape
and because it seems more suited to the next step of
simultaneous sequence processing (helping with the problem of
remembering which token belongs to which sequence(s)/state(s)).

* llama : replace the usage of n_ctx with kv_self.size in many places

* mamba : use n_tokens directly instead of n_tok

* mamba : in comments, properly refer to KV cells instead of slots

* mamba : reduce memory usage of ggml_ssm_scan

From 290.37 MiB to 140.68 MiB of CPU compute buffer size
with Mamba 3B with a batch size of 512.

The result tensor of ggml_ssm_scan was previously a big part
of the CPU compute buffer size. To make it smaller,
it does not contain the intermediate ssm states anymore.
Both y and the last ssm state are combined in the result tensor,
because it seems only a single tensor can be returned by an operator
with the way the graph is built.

* mamba : simultaneous sequence processing

A batch can now contain tokens from multiple sequences.

This is necessary for at least the parallel example, the server example,
and the HellaSwag test in the perplexity example.

However, for this to be useful, uses of llama_kv_cache_seq_rm/cp
will need to be changed to work on whole sequences.

* ggml : add ggml_ssm_conv as a new operator for the conv step of Mamba

This operator makes it possible to use and update the correct states
for each token of the batch in the same way as ggml_ssm_scan.
Other solutions which use existing operators would need loops which would
add too many nodes to the graph (at least the ones I thought of).

Using this operator further reduces the size of the CPU compute buffer
from 140.68 MiB to 103.20 MiB with Mamba 3B with a batch size of 512.
And (at least on CPU), it's a bit faster than before.

Note that "ggml_ssm_conv" is probably not the most appropriate name,
and it could be changed if a better one is found.

* llama : add inp_s_seq as a new input tensor

The most convenient implementation to select the correct state (for Mamba)
for each token is to directly get the correct index from a tensor.
This is why inp_s_seq is storing int32_t and not floats.

The other, less convenient way to select the correct state would be
to have inp_KQ_mask contain 1.0f for each state used by a token
and 0.0f otherwise. This complicates quickly fetching the first used
state of a token, and is also less efficient because a whole row
of the mask would always need to be read for each token.

Using indexes makes it easy to stop searching when there are
no more sequences for a token, and the first sequence assigned
is always very quickly available (it's the first element of each row).

* mamba : support llama_kv_cache_seq_cp copy chains

* mamba : support shifting and dividing the kv cache pos

* mamba : make the server and parallel examples work with whole sequences

A seq_id is dedicated to the system prompt in both cases.

* llama : make llama_kv_cache_seq_rm return whether it succeeded or not

* mamba : dedicate an input tensor for state copy indices

This is cleaner and makes it easier to adapt when/if token positions
(and by extension, inp_K_shift) are no longer integers.

* mamba : adapt perplexity, batched, and batched-bench examples

* perplexity : limit the max number of sequences

This adapts to what the loaded model can provide.

* llama : add llama_n_max_seq to get the upper limit for seq_ids

Used by the perplexity example.

* batched : pass n_parallel to the model's context params

This should have been there already, but it wasn't.

* batched-bench : reserve sequences to support Mamba

* batched-bench : fix tokens being put in wrong sequences

Generation quality isn't what's measured in there anyway,
but at least using the correct sequences avoids using non-consecutive
token positions.

* mamba : stop abusing attention metadata

This breaks existing converted-to-GGUF Mamba models,
but will allow supporting mixed architectures like MambaFormer
without needing to break Mamba models.

This will also allow changing the size of Mamba's states
without having to reconvert models in the future.
(e.g. using something else than d_conv - 1 columns for the conv_states
 will not require breaking existing converted Mamba models again)

* gguf-py : add new KV metadata key-value pairs for Mamba

* llama : add new metadata key-value pairs for Mamba

* llama : guard against divisions by zero when n_head is 0

* mamba : rename "unlimited" KV cache property to "recurrent"

* mamba : more correctly update the "used" field of the KV cache

* ggml : in ggml_ssm_scan, use a threshold for soft_plus

This is how the official Mamba implementation does it,
and it's also what torch.nn.Softplus does.

* convert : for Mamba, fallback to internal NeoX tokenizer

The resulting models are exactly the same
as if the tokenizer.json and tokenizer_config.json of GPT-NeoX were there.

* mamba : support state saving and restoring

* ggml : implicitly pass src tensors through dst for Mamba-related ops

* mamba : clarify some comments

* server : fix cache_tokens not getting correctly resized

Otherwise, when the "we have to evaluate at least 1 token" special case
was triggered, an extra token was kept in cache_tokens even if it was
removed from the KV cache.

For Mamba, this caused useless prompt reprocessing when the previous
request triggered the above case.

* convert-hf : support new metadata keys for Mamba

For the models available at
https://huggingface.co/collections/state-spaces/transformers-compatible-mamba-65e7b40ab87e5297e45ae406

* mamba : rename metadata to be more similar to transformers library

This breaks existing converted-to-GGUF models,
but the metadata names are more "standard".

* mamba : support mamba-*-hf models

These models share their token_embd.weight with their output.weight

* mamba : add missing spaces

This is purely a formatting change.

* convert-hf : omit output.weight when identical with token_embd.weight

Only for Mamba for now, but it might be relevant for other models eventually.
Most Mamba models actually share these two tensors, albeit implicitly.

* readme : add Mamba to supported models, and add recent API changes

* mamba : move state_seq and state_mask views outside layer loop

A few tensors were also missing `struct` in front of `ggml_tensor`.
2024-03-15 14:01:12 +02:00
Jared Van Bortel
1daeffca54
ggml : use SYS_get_cpu if SYS_getcpu is not defined (llama/5906)
Fixes #5694
Fixes ggerganov/whisper.cpp#1894
2024-03-08 11:38:33 +02:00
Georgi Gerganov
edd8b38a75
ggml : fix unknown status (llama/0) 2024-03-08 11:38:32 +02:00
Michael Podvitskiy
9a0b59d990
ggml : introduce ggml_status (ggml/750)
* using enum as an exit code instead of macros

* update return type from enum to unsigned int

* indentation fix

* compound update
ggml_compute_exit_code -> ggml_status
changed ggml_status from a bit-field type to simple codes
ggml_status to string cast

* ggml_status to string cast

* GGML_CALL was removed

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-08 11:38:32 +02:00
leejet
8cee7c08b6
add some new ops, fix some operators and add batch operations to certain operators. (ggml/747)
* cuda: fix group_norm

* cuda: add batch inference support for ggml_pad/ggml_upscale

* add ggml_arrange

* add ggml_timestep_embedding

* update ggml_arange/ggml_timestep_embedding tests

* cuda: fix im2col

* add ggml_arange/ggml_timestep_embbeding support for metal backend

* fix some bugs

* fix some bugs

* Update ggml.h

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Update ggml-cuda.cu

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Update ggml-metal.m

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Update ggml-metal.m

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Update ggml-metal.metal

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* modify according to the review comments

* ggml : fix compile warnings + code style

* ggml : normalize compute_forward calls + fix seg fault in debug

* minor

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: slaren <slarengh@gmail.com>
2024-03-08 11:38:31 +02:00
Kawrakow
2533305596
ggml : make i-quants work with super-blocks of 64 (CPU,Metal) (llama/5760)
* WIP: make i-quants work for QK_K = 64

* iq2_xs: attempt to fix AVX dot product for QK_K = 64

Tests pass, but I get gibberish.

* QK_K = 64 tests pass on ARM_NEON and Metal

Sadly, that does not mean it actually works.

* Make CUDA compile with QK_K = 64

Tests don't pass, plus we get misaligned access

* Q2_K: fixed bug in imatrix quantization for QK_K = 64

* iq1_s: turn off SIMD implementation for QK_K = 64 (it does not work)

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-28 13:00:30 +02:00
Kawrakow
013e394a4b
IQ4_XS: a 4.25 bpw quantization (llama/5747)
* Try IQ4_NL with blocks of 64 - does not look good

* iq4_xs: go to super-blocks of 256 and 6-bit scales for blocks of 32

* iq4_xs: CUDA works - 133.2 t/s

* iq4_xs: AVX2 dot product

* iq4_xs: ARM_NEON dot product

* iq4_nl: Metal implementation

As usual, Metal / Apple Silicon don't like my quants.

* iq3_xs: minor fix

* iq4_xs: shrink by using IQ3_S for attn_k and attn_q

* iq4_xs: revert using IQ3_S for attn_k and attn_v

PPL vs size is good, but CPU performance suffers: on M2 Max
TG-128 drops to 21.7 t/s from 28.8, and on a Ryzen-7950X
to 14.5 t/s from 15.8 t/s. On CUDA we have 135 t/s when
using IQ3_S vs 133 t/s with pure IQ4_XS.

* Fix CI

* iq4_xs: Added forgotten check for 256 divisibility

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-02-28 13:00:29 +02:00