github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-09 15:08:02 -05:00
Code Issues Packages Projects Releases Wiki Activity
7,104 Commits 203 Branches 13 Tags
e37bff6c1905ec03b21e9521cda459d1ad2a2b52
Commit Graph

7104 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
George Hotz
e37bff6c19 fix bug in jit prune with copy [pr] (#8073) 2024-12-06 18:38:23 +08:00
George Hotz
aae8557ada test copy inside jit [pr] (#8072) 2024-12-06 17:51:50 +08:00
George Hotz
e2fe7f0d2f hotfix: actually fix pylint, it's a python 3.10 issue 2024-12-06 13:53:46 +08:00
George Hotz
b28d660172 update self_tokenize, fix pylint maybe 2024-12-06 13:49:41 +08:00
George Hotz
344fd4845c example: self_tokenize. someday tinygrad will be recursively self improving 2024-12-06 13:35:02 +08:00
JaSpa99
3c5d5f9414 mypy==1.13.0 (#7990) 
* explicit instantiation and narrowing asserts

* explicit cast

* bump

* one line assert

* handle case for no copy_queue_t

* Revert "handle case for no copy_queue_t"

This reverts commit 38347806ca.

* more readable control flow

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2024-12-06 12:09:14 +08:00
leopf
65b6696f3b refactor safe_load (#8035) 
* refactor safe_load

* cleanup
 2024-12-06 12:08:21 +08:00
chenyu
e7d5fe4a32 improve idiv _min_max (#8066) 
for the cases that the we don't know the exact bounds, we might still know the sign. with this, can remove some resolve for symbolic shapetracker
 2024-12-05 23:02:16 -05:00
chenyu
13b954f22c unify expand conditions [pr] (#8065) 
same condition (check if old == new or old == 1) in tensor and view. also renamed _pad_left to _align_left because it's not really a pad
 2024-12-05 21:40:14 -05:00
chenyu
aefdff4ef5 reshape mask cleanups [pr] (#8064) 
don't need canonicalize_st because we always merge 1 in `_merge_dims`
 2024-12-05 20:20:43 -05:00
chenyu
05dba6e4ee minor to_indexed_uops cleanup [pr] (#8063) 2024-12-05 17:15:03 -05:00
chenyu
b2dd703592 fix typing of UOp.range [pr] (#8062) 
start/end should not be float or bool
 2024-12-05 14:56:34 -05:00
Sieds Lykles
49c6dab74b Add pattern for div mod recombine with gcd (#8061) 
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-05 13:16:58 -05:00
geohotstan
707e9a9c8e add _one_hot_along_dim helper for Tensor.arange masking (#8039) 
* feelsbadman

* feelsextrabadman

* make sure indices is on same device as self Tensor

* renamed to _one_hot_along_dim

* revert onnx change will do them in onnx only PRs

* address feedback

* add onnx changes here too

* make pad arg better

* revert pad arg

* maybe still keep dim

* simplify onehot onnx ops more

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-05 12:43:00 -05:00
chenyu
3c5983473a combine parentless reduce rule [pr] (#8059) 2024-12-05 11:28:35 -05:00
chenyu
87594a8153 simpler dtypes.max for int [pr] (#8058) 2024-12-05 10:31:41 -05:00
geohotstan
66b8242375 Simple onnx.py clean ups (#8054) 
* start

* simplify ops

* why did this not work before

* will split buffer parse to separate pr

* flip the error order

* only this much for now

* to_python_const clean up

* minimize diff

* move tensor_methods into onnx.py

* improve some type signatures

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-05 10:31:26 -05:00
chenyu
5c6ed5dba6 lower test_conv_3x3_256_32_32_256_256 expectation (#8060) 
failed https://github.com/tinygrad/tinygrad/actions/runs/12182799887/job/33982676812#step:9:210
 2024-12-05 10:30:56 -05:00
Ahmed Harmouche
c6f5bb03fa YoloV8 WebGPU fixes (#8057) 
* Bump up input size to 416, show if webgpu is not supported

* Minor fix in export_model
 2024-12-05 16:23:45 +01:00
nimlgen
78c01a5c2b amd general _gpu_alloc (#8056) 
* amd general _gpu_alloc

* hmm

* ops
 2024-12-05 15:50:23 +03:00
nimlgen
8071600897 nv one _gpu_alloc (#8055) 2024-12-05 15:22:03 +03:00
Ahmed Harmouche
ff9a89f714 Proper dtypes for input/output of exported WebGPU model (#8053) 
* Respect input/output dtypes in exported WebGPU model

* Add some comments about skipped dtypes
 2024-12-05 10:38:05 +01:00
qazal
435a51e10c reduce folding simple tests [pr] (#8040) 
* reduce folding simple tests [pr]

* test for view and realized src pattern

* realize / buffer behavior
 2024-12-05 12:22:45 +08:00
George Hotz
20878be2af lower test_gemv_4096_16384 expectations 2024-12-05 12:08:26 +08:00
George Hotz
83aecbdc70 do gpuocelot copy manually [pr] (#8050) 2024-12-05 11:51:20 +08:00
George Hotz
4a208bfb28 bump download cache version 2024-12-05 11:42:34 +08:00
George Hotz
df18e7cc37 accept filename decorator [pr] (#8049) 
* accept filename decorator [pr]

* add test for safe_load

* bring old tar tests back
 2024-12-05 11:40:59 +08:00
Francis Lata
c3187087f7 QwQ-32B-Preview support (#7962) 
* load weights with some debugging

* start running a prompt

* cleanup

* optionally permute layers and cleanup

* add validation for simple prompt

* small cleanup

* minor cleanup with formatting download links

* add a longer prompt

* add timing option

* some typings

* remove unused arg

* reset GlobalCounters

* minor cleanups
 2024-12-04 21:46:37 -05:00
chenyu
b3220ca7b1 test cases of always True/False lt (#8048) 
* test cases of always True/False lt

* one more
 2024-12-04 20:38:40 -05:00
chenyu
8bb806888b hook_overflow -> safe_exp2 [pr] (#8047) 
that's the only use case, so no need for indirection
 2024-12-04 19:05:38 -05:00
chenyu
99abdc6d39 minor push_swizzle_down_through_elementwise cleanup [pr] (#8046) 
walrus, and if x are the same, prod(x) must be the same
 2024-12-04 17:22:37 -05:00
chenyu
5933ec8dc3 use argfix in smax/smin and remove if [pr] (#8045) 2024-12-04 17:06:13 -05:00
chenyu
4e518334b8 minor get_grouped_dims cleanup [pr] (#8044) 2024-12-04 16:22:51 -05:00
geohotstan
5ce8090d42 simple onnx_ops cleanups (#8003) 
* simple clean ups first

* more work

* kinda have adam

* ooo momentum worked nicely

* almost there

* wow.. is the onnx test wrong

* nicer optim stuff

* just skip that test

* small comment changes

* use naming convention from other parts of codebase

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-04 15:33:03 -05:00
Sieds Lykles
70db1bab5c Fold nested div with const (#8010) 
* Rebase nested div and with const

* Update the ordering

* return None on vectors

Fixes cpu test

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-04 14:59:09 -05:00
chenyu
0693158d28 lower v_theoretical gemv on red (#8042) 
tiny7 is still slower https://github.com/tinygrad/tinygrad/actions/runs/12166149038/job/33931736130#step:8:209
 2024-12-04 13:59:40 -05:00
chenyu
5c2b1089b2 vectorized input in div_and_mod_folding returns None [pr] (#8041) 2024-12-04 13:36:41 -05:00
qazal
ff6def9ffb simple contiguous_while_contiguous prereqs [pr] (#8038) 
* simple contiguous_while_contiguous prereqs [pr]

* early realize

* fine if it's folding a non-contig buffer
 2024-12-04 23:00:28 +08:00
Ahmed Harmouche
c9e7701417 Fast YoloV8 on WebGPU (#8036) 
* Fast yolov8 with downscaled input

* Faster + FPS meter

* Add loader while model is downloading/compiling

* Title touchup
 2024-12-04 15:23:09 +01:00
qazal
b116e1511d make device on uop optional [pr] (#8034) 2024-12-04 20:18:00 +08:00
Ahmed Harmouche
13eedd373b Run WebGPU tests on ubuntu (#8033) 2024-12-04 12:42:04 +01:00
leopf
fb89971e73 use BufferedReader (#8032) 2024-12-04 19:08:54 +08:00
George Hotz
08657cb7b0 hotfix: bump expectations in speed_v_theoretical 2024-12-04 19:00:33 +08:00
George Hotz
ea65c79ba2 hotfix: don't spam BEAM debug in speed_v_theoretical 2024-12-04 18:47:16 +08:00
George Hotz
09b00b1b04 hotfix: use kernel timings instead of python timings in speed_v_theoretical 2024-12-04 18:36:17 +08:00
George Hotz
8f65c1fafb simpler block reorder function [pr] (#8031) 
* simpler block reorder function [pr]

* simpler

* block_reorder in substitute, so wasteful otherwise

* extend and count

* leave push logic for same order

* sort new ctx

* less loop

* Revert "less loop"

This reverts commit 30249d097a.
 2024-12-04 17:57:35 +08:00
leopf
f0401e14e8 tar_extract with Tensors (#7853) 
* initial

* USTAR, PAX and GNU support + testing

* from_bytes byteorder

* use TarInfo.frombuf

* tensor only usage

* remove contextlib.suppress

* shorter ow,pax

* more tests

* testing length + move tests

* cleanup

* new approach: RawTensorIO

* fix fetch

* enable read test

* cleanup and ignore fix

* fix for python < 3.12

* make it RawIO

* functions

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-12-04 17:03:19 +08:00
George Hotz
1e06aefde7 bunch up ops for lines [pr] (#8030) 2024-12-04 17:03:01 +08:00
uuuvn
e9c5b23ba1 Use MTLCompiler directly (v2) (#7920) 
* Use MTLCompiler directly (v2)

* to_block_literal and REQUEST_TYPE_COMPILE

* Rewrite command encoding

* Revert to_block_literal

* Maybe that's more readable to some people?

* Typo and comment about stdlib caching

* Update ops_metal.py

* Update ops_metal.py

* Update ops_metal.py

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2024-12-04 16:36:48 +08:00
George Hotz
bb98bae751 local reordering in block (#8029) 
* local reordering in block

* load (and parents) is highest priority

* minor loads in order

* comments

* explicit depth

* simpler

* matters less, but store early too
 2024-12-04 15:11:29 +08:00