* iq3_xxs: quantize/dequantize
RMSE seems a bit high-ish at about half-way between q2_K and
q3_K, so need to check more.
* iq3_xxs: CUDA dequantize works
* iq2_xxs: tuning quantization
* iq3_xxs: starting to look better
PPL on wiki.test.raw
LLaMA-v1-7B: 6.4218
LLaMA-v2-7B: 6.3560
Mistral-7B : 6.0717
This is better than Q3_K_XS, with a 5% reduction in quantized model
size.
* iq3_xxs: CUDA dot product
We have
PP-512: 5891 t/s
TG-128: 143.9 t/s
* iq3_xxs: scalar and AVX2 dot products
* iq3_xxs: ARM_NEON and Metal
Metal performance is decent, ARM_NEON is pathetic
* iq3_xxs: slightly better grid points
* Faster iq3_xxs and iq2_xs dot products on CUDA
* iq3_xxs: add some quant mix
* iq3_xxs: fix failing quantization test
Dot product still fails. Is this real?
* iq3_xxs: hopefully fix ROCm
* iq3_xxs: failing tests
This time the dot product accuracy did find an actual bug
in the AVX2 implementation.
* Add IQ3_XXS to test-backend-ops
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* iq2_xs: basics
* iq2_xs: this should have been in the basics
* iq2_xs: CUDA and scalar CPU works
* iq2_xs: WIP Metal
* iq2_xs: Metal now works
* iq2_xs: working, but dog slow, ARM_NEON dot product
* iq2_xs: better ARM_NEON dot product
We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when
running on the CPU.
* iq2_xs: AVX2 dot product - 19.5 t/s
* iq2_xs: faster AVX2 dit product
21.4 t/s for TG-128, 59.2 t/s for PP-512.
The latter is 2x compared to the previous version.
* iq2_xs: had forgotten to delete iq2-data.h
* Add llama enum for IQ2_XS
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* iq2_xxs: basics
* iq2_xxs: scalar and AVX2 dot products
Needed to change Q8_K to have quants in the -127...127 range,
else the IQ2_XXS AVX implementation becomes very awkward.
The alternative would have been to use Q8_0 instead. Perhaps
I'll change later, for now this is what we have.
* iq2_xxs: ARM_NEON dot product
Somehow strangely slow (112 ms/token).
* iq2_xxs: WIP Metal
Dequantize works, something is still wrong with the
dot product.
* iq2_xxs: Metal dot product now works
We have
PP-512 = 475 t/s
TG-128 = 47.3 t/s
Not the greatest performance, but not complete garbage either.
* iq2_xxs: slighty faster dot product
TG-128 is now 48.4 t/s
* iq2_xxs: slighty faster dot product
TG-128 is now 50.9 t/s
* iq2_xxs: even faster Metal dot product
TG-128 is now 54.1 t/s.
Strangely enough, putting the signs lookup table
into shared memory has a bigger impact than the
grid values being in shared memory.
* iq2_xxs: dequantize CUDA kernel - fix conflict with master
* iq2_xxs: quantized CUDA dot product (MMVQ)
We get TG-128 = 153.1 t/s
* iq2_xxs: slightly faster CUDA dot product
TG-128 is now at 155.1 t/s.
* iq2_xxs: add to llama ftype enum
* iq2_xxs: fix MoE on Metal
* Fix missing MMQ ops when on hipBLAS
I had put the ggml_supports_mmq call at the wrong place.
* Fix bug in qequantize_row_iq2_xxs
The 0.25f factor was missing.
Great detective work by @ggerganov!
* Fixing tests
* PR suggestion
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* ggml : disable fast-math for Metal (cmake build only)
ggml-ci
* metal : fix Metal API debug warnings
* cmake : add -fno-inline for Metal build (llama/4545)
* metal : fix API debug warnings
* metal : fix compile warnings
* metal : use uint64_t for strides
* cmake : rename option to LLAMA_METAL_SHADER_DEBUG
* metal : fix mat-vec Q8_0 kernel for BS > 1
* metal : normalize mat-vec kernel signatures
* cmake : respect LLAMA_QKK_64 option
* metal : fix mat-vec Q4_K kernel for QK_K == 64
* metal : optimizing ggml_mul_mat_id (wip)
* metal : minor fix
* metal : opt mul_mm_id
* ggml : disable fast-math for Metal (cmake build only)
ggml-ci
* metal : fix Metal API debug warnings
* cmake : add -fno-inline for Metal build (llama/4545)
* metal : fix API debug warnings
* metal : fix compile warnings
* metal : use uint64_t for strides
* cmake : rename option to LLAMA_METAL_SHADER_DEBUG
* metal : fix mat-vec Q8_0 kernel for BS > 1
* metal : normalize mat-vec kernel signatures
* cmake : respect LLAMA_QKK_64 option
* metal : fix mat-vec Q4_K kernel for QK_K == 64
ggml-ci
* whisper : migrate to ggml-backend
* whisper : fix logit reading
* whisper : fix tensor allocation during load
* whisper : fix beam-search with CUDA
* whisper : free backends + fix compile warning
* whisper : print when CUDA is enabled
* whisper : fix CoreML
* make : clean-up
* talk : fix compile warning
* whisper : support ggml_conv with CUDA and Metal (#1473)
* ggml : add CUDA support for ggml_conv
* whisper : remove ggml_repeat for conv bias + single backend
* cuda : fix im2col kernel
* metal : add im2col support + mul mat-vec f16 x f16
* bench-all : add q4 models
* whisper : clean-up
* quantize-all : fix
* ggml : im2col opts
* whisper : avoid whisper_model_data wrapper
* whisper : add note that ggml_mul_mat_pad does not work with CUDA
* whisper : factor out graph compute in common function
* whisper : fixes
* whisper : fix UB with measure buffers
* whisper : try to fix the parallel whisper_state functionality (#1479)
* whisper : try to fix the parallel whisper_state functionality
* whisper : fix multi-state Metal
* whisper : free backend instances in whisper_state
* sync : ggml (backend v2, k-quants, CUDA opts, Metal opts, etc.)
* metal : allow env metal variable to override resource path (#1415)
* Allow env variable to override resource path
* Update ggml-metal.m
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* sync : restore common / main from `master`
* sync : restore whisper from `master`
* talk-llama : update to latest llama.cpp
* ruby : fix build
* ggml : fix 32-bit ARM build
* ggml : fix MIN / MAX macro collisions + update ios bindings
* ggml : fix ifdefs and MIN / MAX again
* exampels : fix Obj-C and Swift examples
* ggml : fix 32-bit ARM compatibility
* ggml : one more attempt to fix 32-bit ARM compat
* whisper : fix support for larger graphs
---------
Co-authored-by: Chris Raethke <codesoda@users.noreply.github.com>
* metal : init
* whisper : factor out graph builds
* whisper : allocate encoder and decoder using ggml-alloc
* whisper : ggml-alloc is now supported
* whisper : CoreML support ggml-alloc
* build : fix ggml-alloc
* ios : update submodule
* extra : update sync-ggml.sh script to also sync ggml-alloc
* ci : see if this is causing the crash
* whisper : refactor ggml-alloc init
* whisper.android : try to fix build
* whisper : initial Metal version
* ci : try to debug vmem issue
* metal : decoder works on GPU!
* metal : add multi-decoder support
* ggml : fix ggml_nbytes (probably temp solution)
* metal : run "cross" step on the GPU
* whisper : remove ggml_repeat in the encoder
* whisper : offload the Encoder to Metal
* ggml : use simpler ggml_bytes() implementation
* ggml-alloc : try to make CI happy by reducing vram to 128GB
* whisper : add whisper_allocr to wrap ggml_allocr
* whisper : factor out alloc init in a function
* cmake : update to support Metal build
* whisper : add <functional> header
* objc : fix build (no Metal yet)
* ios : add Metal support
* swiftui : fix build
* metal : speed-up KQ multiplication
* metal : sync latest llama.cpp kernels
* readme : add Metal info
* ios : update submodule
* coreml : add code to toggle Core ML config (CPU, ANE, GPU)
* bench : fix timings by running a pre-heat
* bench : start benching the decoder
* whisper : add ggml_mul_mat_pad
* bench : fix uninitialized vars
* whisper : add comment for disabling mul-mat padding
* whisper : add description of ggml_mul_mat_pad
* whisper : clean-up ggml_mul_mat_pad
* metal : remove the "concurrent" flag
* bench : variable n_past
* ios : update SPM package
