nimlgen/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity
13,471 Commits 334 Branches 15 Tags
python_prg
Commit Graph
13471 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Priyank PatelandGitHub 4f5e03bd60 better fix inplace detach (#9557) 2025-03-24 22:50:28 +08:00
qazalandGitHub 1c40873962 show buffer info in memory viz (#9562) 2025-03-24 22:12:30 +08:00
qazalandGitHub efaee75656 start viz of memory usage (#9561) 
* start viz of memory usage

* polygons/bars + use d3
 2025-03-24 19:05:35 +08:00
qazalandGitHub 1cfe6d02fe refactor uop_to_json to return a dict [pr] (#9560) 2025-03-24 16:38:17 +08:00
nimlgenandGitHub edf9e1bf8d am: move out soc21 to a sep module (#9551) 
* am: soc module is not part of am

* am: soc module is not part of am
 2025-03-24 14:17:42 +07:00
George HotzandGitHub 74d98eafb8 add onnx frontend stub [pr] (#9558) 2025-03-24 12:24:34 +08:00
geohot de7d6cec3a hotfix: DEBUG 5 prints the ast 2025-03-24 11:43:11 +08:00
chenyuandGitHub ba41076e94 update embedding test to not use dtypes.long [pr] (#9556) 2025-03-23 21:33:38 -04:00
chenyuandGitHub c965f4c20b update bert config (#9555) 
BEAM 4->5 for green, 2% faster
use AMD driver instead of AM for red, 5% faster
 2025-03-23 16:14:41 -04:00
chenyuandGitHub d734e24c01 minor WEBGPU_PATH cleanup [pr] (#9552) 
also mypy recognizes `sys.platform == 'win32'` but does not recognizes it if wrapped inside a helper...
 2025-03-23 09:10:02 -04:00
Ahmed HarmoucheandGitHub 7ce7fe0574 Refactor webgpu_dawn lib finding (#9547) 
* Refactor webgpu_dawn lib finding

* Fix ruff
 2025-03-23 08:23:29 -04:00
uuuvnandGitHub c631c72f22 HCQ: Increment timeline signal before submitting (#9550) 
`AMDComputeQueue.__del__` frees `hw_page` which is safe because
`AMDAllocator._free` does `self.dev.synchronize()` which is supposed
to wait for execution of IB to finish, however that doesn't happen if
AMDComputeQueue is dropped right after submit before timeline signal is
incremented, which it is in most places leading to a race if .bind() is
also used (required for multi-xcc because bug in mec fw treats all
PACKET3_PRED_EXECs outside IBs as if they had EXEC_COUNT of zero).
 2025-03-23 18:30:38 +07:00
nimlgenandGitHub d5667419af am: move out pte creation logic (#9548) 
* am: move out pte creation logic

* emu

* ops
 2025-03-23 18:29:10 +07:00
geohotstanandGitHub 309afa20b7 add Tensor.max_unpool2d (#9518) 
* why does max_unpool2d feel slower than out.gradient ...

* slightly cleaner

* what happened to ruff

* need to think about this some more

* slightly faster now?

* clean up, 1 more failing edge case

* ok good

* working TINY_BACKEND

* nit doc wording

* retry CI
 2025-03-22 12:11:33 -04:00
quortusandGitHub bdd44d4255 Fix DSP transcendentals (#9542) 2025-03-22 11:08:18 +08:00
Ignacio SicaandGitHub eddafb84e5 Bugfix for TC=3 (#9464) 
* wrong but uses less shared

* for size 8 tc1 with devectorize in 0 loads into local before wmma and works

* improvements over tc1 devectorize

* fix tc=3

* works for handcoded tc opts

* clean bugfix tc=3

* fix

* revert changes
 2025-03-21 16:43:42 -07:00
chenyuandGitHub 6da78164f9 assert Kernel ast.op to be Ops.SINK [pr] (#9539) 
rest of the code assumes self.ast is defined anyway
 2025-03-21 18:09:44 -04:00
chenyuandGitHub c33679c47b increase size in test_multinomial_counterexample (#9540) 
should be less flaky
 2025-03-21 17:46:52 -04:00
Francis LataandGitHub 1a1087e3a0 cleanups on losses and dataset tests (#9538) 2025-03-21 17:03:18 -04:00
Francis LataandGitHub 8cbe4009fc RetinaNet losses (#9536) 
* add sigmoid_focal_loss and l1_loss

* update ref implementation comment
 2025-03-21 15:52:54 -04:00
Francis LataGitHubchenyu
 		e6389184c5 update comment for retinanet dataloader implementations (#9534) 
Co-authored-by: chenyu <[email protected]>
 2025-03-21 15:07:45 -04:00
chenyuandGitHub ee3d313b34 Revert "update ruff to 0.11.2 (#9531)" (#9535) 
This reverts commit d8d65e2747.
 2025-03-21 14:52:25 -04:00
chenyuandGitHub b46b8ee15e add a flag to log when beam surpassed max limit [pr] (#9533) 2025-03-21 13:37:02 -04:00
Francis LataandGitHub eb95825eea RetinaNet dataloader (#9442) 
* retinanet dataloader

* remove batch_size from generate_anchors

* refactor kits19 dataset tests

* add tests for dataloader

* fix testing setup and cleanups

* remove unused import
 2025-03-21 13:36:41 -04:00
b1tgGitHubb1tgchenyu
 		58206fa8a9 add amd llvm compiler (#9519) 
Co-authored-by: b1tg <[email protected]>
Co-authored-by: chenyu <[email protected]>
 2025-03-21 23:13:27 +08:00
chenyuandGitHub d8d65e2747 update ruff to 0.11.2 (#9531) 
0.11.2 fixed the false alert from 0.11.1. also pinned the version in setup for now to prevent broken CI from ruff upgrade
 2025-03-21 10:32:59 -04:00
qazalandGitHub ee3ed73ed1 add reorder_view matcher to scheduler [pr] (#9528) 2025-03-21 17:46:20 +08:00
George HotzandGitHub 8e555c586c switch quantization to unsigned/unsigned + add Ops.REDUCE (#9527) 
* switch quantization to unsigned/unsigned + add Ops.REDUCE

* tests

* nhwc + replay pkl
 2025-03-21 17:02:37 +08:00
nimlgenandGitHub a35b0a88bf am: just rename and reorder ip init funcs (#9504) 2025-03-21 15:57:32 +08:00
nimlgenandGitHub 8a131ab271 am: allow allocations as small as a page (#9523) 
* am: fix allocs

* bettermsg

* comment

* next time
 2025-03-21 15:53:32 +08:00
Sieds LyklesandGitHub 3ad3ac4d1e Change dtypes.int to dtypes.ints (#9517) 2025-03-20 17:24:26 -04:00
chenyuandGitHub b9fab9b914 pin ruff to 0.11.0 in CI (#9520) 
0.11.1 had a bug https://github.com/astral-sh/ruff/issues/16874 that breaks ci
 2025-03-20 13:12:50 -04:00
George HotzGitHubqazal
 		3c5161b4cb add validation of the bounds of Ops.INDEX (#9503) 
* add validation of the bounds of Ops.INDEX

* do mask properly

* more validation

* correct

* fix gated

* add CAST support to vmin/vmax

* fix ptx and image

* ptx no diff

* upat.index also stays

---------

Co-authored-by: qazal <[email protected]>
 2025-03-20 12:15:55 +08:00
qazalandGitHub 0b20f91ce7 remove move_mask from the devectorizer (#9511) 
* remove move_mask from the devectorizer

* add (wrong) ptx

* reason

* enable index addition in PTX, we won't have the INDEX anyways

* space
 2025-03-20 11:53:12 +08:00
qazalandGitHub 9302738263 hotfix: more consistent wgsl.py spacing + cleanups [pr] (#9515) 
* hotfix: more consistent wgsl.py spacing + cleanups [pr]

* free things up
 2025-03-20 11:07:15 +08:00
George HotzGitHubComma Device
 		68053d0510 dsp stuff / sniff ioctls from snpe (#9490) 
* sniff ioctls from snpe

* dump input buffers

* snpe logs from dsp

* NHWC support

* knum 3

* this run?

* revert those

---------

Co-authored-by: Comma Device <[email protected]>
 2025-03-20 10:38:23 +08:00
qazalandGitHub 2223b93338 add UPat.or_casted [pr] (#9513) 2025-03-20 10:08:32 +08:00
qazalandGitHub 1839e8c9b3 place masks in INDEX for TestGatedStoreRewrite [pr] (#9512) 2025-03-20 09:46:53 +08:00
b1tgGitHubb1tg
 		bd731a8624 AMDCompiler refactor (no_comgr prereq) (#9497) 
* add amdgpu_disassemble to helpers

* refactor hip compiler

---------

Co-authored-by: b1tg <[email protected]>
 2025-03-20 09:44:07 +08:00
geohotstanandGitHub 8c0d0a122c Add return_indices to max_pool (#9506) 
* wow argmax is so good

* 1 less line

* clean up and better variable names

* is this torch thing right...?

* add more tests

* slap a TODO on it

* clean ups

* prettier looking code and fix ceil mode test

* add return types and some docs

* ok that was a bad example since indices == value, just no example
 2025-03-19 15:25:37 -04:00
chenyuandGitHub 189f62d44f add rounding to tqdm unit scale (#9507) 
fixed `AssertionError: ' 1.00/10.0  1000it/s]' != ' 1.00/10.0  1.00kit/s]'`
 2025-03-19 12:08:46 -04:00
nimlgenandGitHub a5c971ff3a am: prereqs for rdna4 1/n (#9495) 
* am: ip_ver rename for acc

* am: refactor this

* fix version

* ugh
 2025-03-19 17:14:57 +08:00
Francis LamandGitHub 1e5d9ad8f7 extra/gemm/max_matmul: start of custom kernels for GEMM (#6926) 
* extra/gemm/max_matmul: start of custom kernels for GEMM

* add an unoptimized FP16/FP16 MMA example

* add slow 3-stage fp16 acc example

* add correct 3-stage pipeline with unswizzled/flat smem input (slow)

* add acc fp16 example with 3 stages and swizzle (no bank conflicts)

* add max version of NV fp16_fp16_fp16

* fix up comments and removed unused code in max variations

* add start of no_xor example

* fix to account for UOps to Ops
 2025-03-19 15:04:57 +08:00
geohot 865f23dd7b olmoe memory usage cleanups 2025-03-19 12:28:18 +08:00
b1tgGitHubb1tg
 		2c87a22cf2 fix prg size calculation when there are adjacent mapped ranges (#9498) 
Co-authored-by: b1tg <[email protected]>
 2025-03-19 11:55:03 +08:00
b1tgGitHubb1tg
 		1d71436e6a use libllvm19 in ci (#9494) 
Co-authored-by: b1tg <[email protected]>
 2025-03-19 11:53:32 +08:00
b1tgGitHubGeorge Hotz
 		a95b489a55 nanoGPT train works with tiny torch backend (#9283) 
* train_shakespeare_char.py works

* move aten.where.self_out to tiny_backend_out

* fix memory leak

* corealize in the backward_hook

* Update backend.py

---------

Co-authored-by: George Hotz <[email protected]>
 2025-03-19 11:51:02 +08:00
chenyuandGitHub f8976dd2eb enable more webgpu tests (#9502) 
OSX has larger buffer number limit, and it supports fp16 now
 2025-03-18 23:03:54 -04:00
qazalandGitHub ae688e4103 simple failing test for scheduling parallel reduce [pr] (#9501) 
* simple failing test for scheduling parallel reduce [pr]

* atol
 2025-03-19 10:52:13 +08:00
leopfGitHubGeorge Hotzchenyu
 		e4dad99145 nn.state docs cleanup (#8332) 
* doc cleanup

* extension cleanup

* manual definition

* bring back accept_filename for gguf_load

---------

Co-authored-by: George Hotz <[email protected]>
Co-authored-by: chenyu <[email protected]>
 2025-03-18 17:16:40 -04:00