github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-10 07:28:15 -05:00
Code Issues Packages Projects Releases Wiki Activity
7,347 Commits 203 Branches 13 Tags
a647f3dd2c46c4959688058f977f933a6ec358e3
Commit Graph

7347 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
nimlgen
a647f3dd2c move mockgpu to tests [pr] (#8396) 
* move mockgpu to tests

* linter

* i'm so sorry

* sorry, python

* path
 2024-12-24 23:48:02 +03:00
chenyu
2c93f27652 remove explicit np.array and np.int32 in test_div_int [pr] (#8395) 
vals default loads as int32 now in test_ops
 2024-12-24 13:09:30 -05:00
qazal
5c2fe04bb6 a few more UPat.var -> UPat.cvar in the scheduler [pr] (#8391) 
* a few more UPat.var -> UPat.cvar in the scheduler [pr]

* keep it assert

* minimal diff
 2024-12-24 20:36:24 +08:00
qazal
3273972f44 delete is_unrealized_const, it's just CONST [pr] (#8390) 2024-12-24 16:46:12 +08:00
qazal
3a556a7e8b fully local tensor const representation: CONST(VIEW(DEVICE)) [pr] (#8389) 2024-12-24 16:15:56 +08:00
George Hotz
b589dec06e remove some VIEWs we don't need [pr] (#8353) 
* remove some VIEWs we don't need [pr]

* unmasked view and movement op on BUFFER are a part of the spec

---------

Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com>
Co-authored-by: qazal <qazal.software@gmail.com>
 2024-12-24 14:57:42 +08:00
chenyu
0d6fe6200c test case view from an empty view (#8388) 
currently it behaves differently depends on first view somehow
 2024-12-23 17:40:49 -05:00
chenyu
c587b3b08c test case view the padded area of a view (#8386) 
these cases view the padded area of first view
 2024-12-23 16:47:31 -05:00
Francis Lata
239d2a7214 explicitly check value for not None (#8382) 2024-12-23 11:12:39 -05:00
geohotstan
78cb47dfc5 docs and tests clean ups (#8383) 2024-12-23 11:12:13 -05:00
chenyu
a556adf028 add test for Tensor silu and swish (#8381) 
was only tested in onnx, added to test_ops for completeness
 2024-12-22 21:08:59 -05:00
chenyu
572ebd9f27 unneeded TYPE_CHECKING in uopgraph.py [pr] (#8379) 
can just import. 3 TYPE_CHECKING left. 2 are about python 3.12 and format str and those are fine. the last one is ops <-> device and shapetracker
 2024-12-22 12:58:08 -05:00
qazal
8d4439282b swizzle assign [pr] (#8378) 2024-12-22 18:17:50 +02:00
qazal
e6f4c24619 try 2 on VIEW(BUFFER, <op>) scheduling + spec [pr] (#8377) 
* second iteration on VIEW(BUFFER, <op>) scheduling + spec [pr]

* image

* notes
 2024-12-22 16:30:35 +02:00
chenyu
b7397c1322 more typing cleanups [pr] (#8376) 
List, Tuple, DefaultDict
 2024-12-22 05:21:03 -05:00
chenyu
afcd70af97 remove untriggered/untested code [pr] (#8375) 
ran with `coverage` on test. not sure about if we still want max_var_const so just commented it out
 2024-12-22 04:07:53 -05:00
chenyu
8fdcb60461 remove unused View.t and lt [pr] (#8374) 2024-12-22 02:26:54 -05:00
chenyu
7ea633f94f remove from __future__ import annotations from runtimes [pr] (#8373) 
it's not needed if we move the Device before Program and Allocator, which need Device.

not updating hcq because it has a lot more stuff, and CLDevice requires CLDevice
 2024-12-21 23:46:07 -05:00
chenyu
e934f987c6 minor cstyle cleanup [pr] (#8371) 
* minor cstyle cleanup [pr]

* *
 2024-12-21 22:17:45 -05:00
qazal
514a6740e4 Revert "CONST(VIEW(DEVICE)) (#8365)" (#8372) 
This reverts commit 83284985f0.
 2024-12-22 04:44:34 +02:00
qazal
83284985f0 CONST(VIEW(DEVICE)) (#8365) 2024-12-22 04:18:35 +02:00
qazal
88bc51385c scheduler: don't trade complexity for speed (#8370) 
* scheduler: don't trade complexity for speed

* don't need is_scheduled

* make those tests real world

* graph_rewrite dedup
 2024-12-22 03:30:51 +02:00
qazal
991b91d4d6 fix string repr of arg in viz and print [pr] (#8369) 2024-12-21 23:44:10 +02:00
ignaciosica
ba0c844a83 special tol when f16 and bf16 are tc input dtypes (#8183) 2024-12-21 11:32:26 -05:00
geohotstan
3f83748661 update onnx and onnx_ops to 3.10+ typing (#8360) 
* fixed mypy and updated to modern typing

* selective ruff check changes (all except E501)

* some more clean ups

* fix comment

* small nit
 2024-12-21 11:17:47 -05:00
qazal
72aa38aa3b BIND in tensor_uop_spec + cleanups [pr] (#8363) 
* Ops.BIND pattern in tensor_uop_spec + cleanups [pr]

* use metaops there
 2024-12-21 21:26:47 +08:00
qazal
4e8812db37 asserts to prepare for Tensor BIND proposal [pr] (#8362) 2024-12-21 20:35:08 +08:00
chenyu
1ce9851ba6 import and type cleanups [pr] (#8359) 
Dict and DefaultDict and some imports
 2024-12-20 21:52:02 -05:00
chenyu
18dca3c3d7 isolate train_gpt2 slow kernels [pr] (#8358) 
also fixed run_linearizer with var_vals=None
 2024-12-20 17:59:01 -05:00
George Hotz
9f62c80f68 hotfix: this is a loan 2024-12-20 14:47:04 -08:00
qazal
2649e87546 delete the fake buffer from const (#8355) 
* delete the fake buffer from const

* fix test_sink_childless_const_alt

* it should be CONST(VIEW(DEVICE))
 2024-12-21 04:20:28 +08:00
George Hotz
b7499764f5 horfix: have viz hide the stupid -1 BUFFERs 2024-12-20 10:47:44 -08:00
chenyu
cd79a904c5 add back explicit dict[DType, str] in ptx [pr] (#8352) 2024-12-20 13:19:48 -05:00
George Hotz
074315ec08 hotfix: simpler test_mnist_model 2024-12-20 10:18:17 -08:00
chenyu
20eebbc61a minor PTX cleanups [pr] (#8351) 2024-12-20 12:52:53 -05:00
qazal
59f4b8da95 Tensor uop spec (#8311) 
* Tensor uop spec

* minor

* feedback

* restrict ShapeTracker of VIEW(BUFFER) to contiguous

* in image base mutates, how do we rewrite the view?

* cast post realize

* now ucache errors

* how strict can this be?

* put constraints on EMPTY

* merge

* save lines

* import import

* overloaded assign target

* more strict

* fine don't overload it

* more

* actually, this is better

* and it even exists

* this way it works for BUFFER

* Revert "this way it works for BUFFER"

This reverts commit 71c15f6b14.

* make it like linearize.py

* assign take 4

* minor

* all int, space and that's already base

* target

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2024-12-20 23:47:40 +08:00
qazal
5776ea9386 hotfix: account for all changes in process_replay early stopping [pr] (#8348) 2024-12-20 23:46:46 +08:00
chenyu
e63c7818dc few type cleanups [pr] (#8347) 2024-12-20 01:56:01 -05:00
George Hotz
82833f1b3c a little more typing [pr] (#8346) 
* a little more typing [pr]

* few more
 2024-12-19 22:09:52 -08:00
George Hotz
62e5d96446 more typing work [pr] (#8345) 2024-12-19 21:46:35 -08:00
George Hotz
9c77e9f9b7 replace Tuple with tuple [pr] (#8344) 
* replace Tuple with tuple [pr]

* replace List with list [pr]

* replace Dict with dict [pr]

* replace Set with set [pr]
 2024-12-19 21:27:56 -08:00
George Hotz
adcdc583a2 small cleanups [pr] (#8343) 
* small cleanups [pr]

* GPU suppress
 2024-12-19 21:20:46 -08:00
George Hotz
9f306e12ac hotfix: test_net_speed can't backward before realize 2024-12-19 20:32:59 -08:00
George Hotz
aa9462c29b fix (some) requires_grad [pr] (#8342) 2024-12-19 19:34:14 -08:00
Harald Schäfer
7059459648 Openpilot compile: fix for openpilot use (#8338) 
* compile3 changes

* merge conflict

* merge conflict

* give dm npy for now

* Revert "give dm npy for now"

This reverts commit bfd980da7d2c2bab5b073127442c361922032ba1.

* updates

* Always float32 floats

* Update compile3.py

* Update compile3.py

---------

Co-authored-by: ZwX1616 <zwx1616@gmail.com>
 2024-12-19 19:43:15 -05:00
chenyu
7153f7709f update test_merge_view_recursion_err2 [pr] (#8339) 
the view was not created through View.create, updated the test to show the expected behavior
 2024-12-19 18:29:34 -05:00
chenyu
2bf47b75da temp fix for symbolic shape view add [pr] (#8337) 
something is still wrong with symbolic shape shrink, but it should not recurse forever
 2024-12-19 16:10:42 -05:00
chenyu
791a80a1c7 add failed merge view example to test_simplify_valid_idx [pr] (#8334) 
* add failed merge view example to test_simplify_valid_idx [pr]

* !=True is fine
 2024-12-19 12:54:03 -05:00
qazal
8e266091fb tensor const spec [pr] (#8331) 2024-12-19 22:41:30 +08:00
George Hotz
0ad264ed2d new from uops [pr] (#8330) 
* new from uops [pr]

* mem_estimate is it's own thing
 2024-12-18 23:42:58 -08:00