mirrors/tinygrad
2
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-24 02:14:17 +00:00
Code Issues Projects Releases Packages Wiki Activity
11,106 commits 280 branches 15 tags 325 MiB
Commit graph

11,106 commits

This branch
This branch
All branches
Author SHA1 Message Date
George Hotz
b245f1307e
 		add exp2 (#2192) 2023-10-31 17:48:42 -07:00
qazal
e2428b63a6
 		external (#2191) 2023-10-31 13:57:24 -07:00
Elias Wahl
7e8c5f1a0f
 		Modernize setup.py (#2187) 
* Added pyproject.toml

* Pin onnx
 2023-10-31 13:55:45 -07:00
nimlgen
8c07c73a9b
 		Fix cl map buffer (#2190) 
* fix gpu enqueue_map_buffer out of space

* add test
 2023-10-31 12:02:46 -07:00
George Hotz
c59ea32f90 prevent over unrolling in optimzer 2023-10-31 11:45:18 -07:00
George Hotz
5aaa8a0cc1 fix shape 2023-10-31 11:36:19 -07:00
George Hotz
a27c9f9de5
 		openpilot compile2 (#2189) 
* try compile2

* pass to thneed

* fix tanh onnx
 2023-10-31 11:08:58 -07:00
qazal
be5f185ac0
 		Higher test coverage for dtypes (#2156) 
* refactor unit tests for dtypes

* add missing dtypes in llvmir.py and lib.py

* skip torch tests

* webgpu

* cleaner skips

* fix llvm bool casting issue using compare

* llvm 100% passing

* llvm segfault

* TEMP decrease timeout mins to 11

debug

* add bf16 to setup

* skip half tests in cuda cpu

* check for CUDACPU insetad

* add int16 to triton dtypes

* u16 for triton

* remove debug - diff is still hard to read

* derive from base class TestDType

* enhance test_upcast and downcast by running on every possible version

* dummy commit to rerun the flakey test

* skip the correct tests for CUDA

* bf16 should be skipped in the common TestDType cases

* re-enable bf16

* more consistent structure

* tiny changes to is_dtype_supported 1

* tiny changes 2

add reason

* fuzz

* fuzzer p2

* run fp32 twice

* remove duplicate fp32 run

* clang: use stdbool

* skip triton on bool casts

* merge and resolve conflicts
 2023-10-30 22:38:42 -07:00
forcefieldsovereign
f294bdd681
 		fixed imports (#2185) 2023-10-30 22:07:17 -07:00
Akshay Kashyap
018bd29e37
 		Enable Multi-Output Export (#2179) 
* Enable Multi-Output Export

* Add test

* Update examples and lint

* fix padding

* test ops

* dummy commit to rerun test

* revert cuda lint

* Enforce tuple/list of tensors

* subscripted generics

* put back webgpu test

* Re-enable WebGPU Efficientnet test
 2023-10-30 18:42:26 -07:00
qazal
a7439af786
 		Fix llvm int->bool cast (#2164) 
* add to ir

* add test case

* minimize diff

* todo

* enable fast math

* added both False and True case
 2023-10-30 15:28:23 -07:00
George Hotz
94cf652b6b don't use locals applies to GROUP also 2023-10-30 13:56:43 -07:00
George Hotz
5cc536bcc0 don't use locals applies to LASTLOCAL 2023-10-30 13:53:42 -07:00
chenyu
3c88af5071
 		use unique table name for each disk_cache test (#2184) 2023-10-30 13:49:49 -07:00
George Hotz
608e3ee800
 		fix no locals search and search both (#2171) 
* fix no locals search and search both

* pretty print

* nolocals default no other search
 2023-10-30 10:22:50 -07:00
George Hotz
194e4ad6f8
 		Revert "optimizer: simplify GROUP and LOCAL to have one of each (#2162)" (#2182) 
This reverts commit 8cf0bb9351.
 2023-10-30 10:22:26 -07:00
Ahmed Harmouche
95f7183c3a
 		Reenable global, local limiting (#2095) 2023-10-30 10:17:23 -07:00
chenyu
8548b20b23
 		fix codellama params and repeat_kv (#2181) 2023-10-30 10:16:26 -07:00
George Hotz
c7f4dd6cb0 CACHELEVEL for smaller caches 2023-10-28 07:26:03 -10:00
chenyu
6c58bf3e9c
 		in time_linearizer, allocate a scratch buffer if output buffer is also input (#2152) 
* in time_linearizer, allocate a scratch buffer if output buffer is also input

* move scratch buffer creation outside search
 2023-10-28 07:17:41 -10:00
Yixiang Gao
902f00b095
 		adding cuda TC headers (#2165) 
* split cuda to renderer and add headers for tc

* fix TritonRenderer

* remove unused import
 2023-10-27 14:25:59 -10:00
David Hou
7f4f925385
 		fix hip del on compile fail (#2163) 
* fix hip del on compile fail

* the test doesn't actually work
 2023-10-27 11:38:07 -10:00
Francis Lam
8cf0bb9351
 		optimizer: simplify GROUP and LOCAL to have one of each (#2162) 
* optimizer: simplify GROUP and LOCAL to have one of each

Now that tensor cores only use LASTLOCAL, we can simplify to use
only that op everywhere.

The only use of GROUP is in matvec hand-coded opts and it doesn't
make a performance difference so switching to use only the top
behavior.

Also adds additional asserts to prevent tensor core dims from
being altered which causes bad kernels to be generated.

* search: remove duplicated actions
 2023-10-27 11:37:44 -10:00
George Hotz
e0201922e3
 		Q network for pruning BEAM / uops deduping / BEAM_ESTIMATE (#2142) 
* stable diffusion < 324ms

* revert swap action

* fix tests due to more sum splitting

* REDUCEOP_SPLIT_THRESHOLD env var

* added from unaligned np test (#2134)

* align cpu buffer before copy into cl buffer (#2135)

* remove shelve from handcode_resnet50_opt.py (#2139)

* Add dictionary keys to reduce db size (#2131)

* work

* ignore beam cache

* dictionary keys are generic

* minor db cleanups

* fix baseline and extract dataset

* fix training

* log likelihood

* more lin to feats

* sts

* training policynet

* net sort of works

* dedup

* refactor, stupid new actions

* fix uops deduping

* BEAM_ESTIMATE

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
Co-authored-by: imaolo <56898718+imaolo@users.noreply.github.com>
 2023-10-27 10:53:06 -10:00
will
bc0829b677
 		Fix llama json loading (#2160) 2023-10-27 10:21:56 -10:00
nimlgen
8d41b3eb3f
 		beam=16 makes gpt2 gpu-time < 5ms on 3090 (#2154) 2023-10-27 10:21:27 -10:00
nimlgen
5204864eca
 		init cudagraph (#2153) 
* init cudagraph

* linter happy

* print warning when cuda graph creation failed
 2023-10-27 16:19:50 -04:00
chenyu
9215bccb41
 		Tensor.uniform set default to standard uniform (#2158) 
* Tensor.uniform set default to standard uniform

* clean up test to reuse function
 2023-10-27 16:15:30 -04:00
Roelof van Dijk
36ab04ae35
 		perf: lazyop as dataclass (#1603) 
* perf: lazyop as dataclass

fix: linter

fix: restore eq

* use builtin methods, buffers to property to allow freezing

* fix: reduce diff

* fix: can't freeze due to KOPT tests, mypy

* fix: explicit hash

* can freeze if tests are fixed

* fix: typo

---------

Co-authored-by: Roelof van Dijk <roelof.van.dijk@vitestro.com>
Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2023-10-25 17:54:30 -04:00
chenyu
0ca0e9ee5e
 		exclude ast with variables from beam search (#2140) 
* exclude ast with variables from beam search

* test that

* add to CI
 2023-10-25 16:35:29 -04:00
Szymon Ożóg
a52b420fb3
 		switch ocelot back to main repo (#2147) 
* return to ocelot main branch

* cd before checkout
 2023-10-25 15:14:26 -04:00
George Hotz
12dd165d38 add WINO/HALF/HIP to AMD benchmark 2023-10-25 13:22:45 -04:00
Francis Lam
bf3490cdf9
 		wmma: refactor tensor cores using existing local dims (#2097) 
* wmma: refactor tensor cores using existing local dims

* optimizer: fix bad rebase and break after one late local

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2023-10-25 13:10:46 -04:00
wozeparrot
c29653605e
 		hip multigpu training (#1878) 
* feat: move to hip

* feat: special path for RawBufferTransfer

* feat: initial rawbuffertransfer

* feat: hip ipc

* feat: working hip ipc

* feat: need to base device without args

* feat: close mem handle

* feat: modified test

* feat: more multihip stuff

* clean: cleanup

* feat: cleaner

* feat: don't crash

* feat: test more

* clean: way cleaner hip wrapper

* feat: barrier

* feat: barrier

* feat: this breaks stuff

* feat: we can use empty here

* feat: maybe fix tests

* feat: maybe fix tests again?

* fix: probably fix tests

* feat: no waiting here

* feat: wait here

* feat: much larger test

* feat: need to sync here

* feat: make this async

* feat: no waiting!

* feat: cut here

* feat: sync copy

* feat: random imports

* feat: much cleaner world

* feat: restore this

* feat: restore this

* clean: cleanup

* feat: set this
 2023-10-24 17:35:53 -04:00
nimlgen
2e89fd264f
 		Refactor hipgraph (#2141) 
* refactor hip graph

* linter happy

* happy liner
 2023-10-24 15:45:56 -04:00
nimlgen
e21bf776c8
 		fix debug=1 llama/gpt2 timings (#2143) 2023-10-24 15:45:00 -04:00
chenyu
4444e6d4b3
 		stable diffusion < 324ms (#2129) 
* stable diffusion < 324ms

* revert swap action

* fix tests due to more sum splitting

* REDUCEOP_SPLIT_THRESHOLD env var
 2023-10-24 14:56:12 -04:00
George Hotz
cea2bc7964
 		Add dictionary keys to reduce db size (#2131) 
* work

* ignore beam cache

* dictionary keys are generic

* minor db cleanups

* fix baseline and extract dataset

* fix training

* log likelihood
 2023-10-24 10:49:22 -04:00
chenyu
d5e2fdea22
 		remove shelve from handcode_resnet50_opt.py (#2139) 2023-10-24 10:37:30 -04:00
imaolo
228b310478
 		align cpu buffer before copy into cl buffer (#2135) 2023-10-23 21:04:35 -04:00
imaolo
6ee0435263
 		added from unaligned np test (#2134) 2023-10-23 11:38:57 -04:00
George Hotz
3c56c181f6 string formatting 25 -> 30 to fit 2023-10-22 10:57:34 -07:00
George Hotz
6dc8eb5bfd
 		universal disk cache (#2130) 
* caching infra for tinygrad

* nons tr key

* fix linter

* no shelve in beam search

* beam search caching

* check tensor cores with beam too

* pretty print

* LATEBEAM in stable diffusion
 2023-10-22 10:56:57 -07:00
Francis Lam
ace6b2a151
 		optimizer: add test for correctness of opts (#2124) 
* optimizer: add test for correctness of opts

Also added OptOps.UPCASTMID to constrain valid axes for opts with
group_for_reduce.

* llvm: fix LinearizerOptions to correctly not has_shared

* optimizer: remove premature test scaffold for TC opts

* search: fix the action space
 2023-10-22 08:02:22 -07:00
George Hotz
abeba8f1fc
 		optimization: get actions in CI (#2125) 
* get actions in CI

* actually run the test

* pythonpath
 2023-10-20 12:22:01 -07:00
qazal
14625721e9
 		minor triton casting refactor (#2118) 
* minor refactor

* render_cast taking an x like cstyle

* fix fmt strings

* tl.where

* fix alu render

* use dtype

* newline eof

* better diff
 2023-10-20 12:11:55 -07:00
George Hotz
cb508e6923
 		uops graphing + phi (#2120) 
* uops graphing

* add_phi_node

* less phi nodes

* where graph uops should live

* naming

* move it to external

* fix triton yolo

* fix clang and preserve behavior
 2023-10-19 22:26:28 -07:00
20kdc
bedd028061
 		waifu2x vgg7: testcase, auto-RGBA->RGB, function to grab pretrained models, training "fix" (#2117) 2023-10-19 22:07:15 -07:00
Szymon Ożóg
e0b2bf46b4
 		Improve triton generated code quality (#2119) 2023-10-19 22:06:19 -07:00
qazal
36d4001b4f
 		add test coverage for search (#2104) 
* add test coverage for search

* only in compiled backends

* dont use device.default in decorator

* time_til is the other way around xd
 2023-10-19 17:06:47 -07:00