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
4,871 commits 280 branches 15 tags 325 MiB
Commit graph

4,871 commits

This branch
This branch
All branches
Author SHA1 Message Date
chenyu
d8dc43ad06
 		remove JIT_BATCH_SIZE=4 from gpt2 NV benchmark (#5198) 
this no longer helps
 2024-06-27 15:20:34 -04:00
George Hotz
345bcc2099
 		move graph_dedup out of class [run_process_replay] (#5197) 2024-06-27 12:04:00 -07:00
George Hotz
d094a6828f
 		single pass rewrite (#5159) 
* single pass rewrite

* claude cleanups

* claude cleanups

* skip those tests

* restrict that to ints

* comment

* asserts i don't expect to fail do fail

* simplest...rewrite...ever

* simplest...rewrite...ever

* add that rule back

* tests pass?

* only collapse reduce loops

* second SHL/SHR arg must be 4 bytes

* fix verify

* no SHL/SHR in ptx

* put that back

* skip them in PTX...bad tests
 2024-06-27 11:36:05 -07:00
Roelof van Dijk
1ff9bbaa61
 		ruff: close file handle (#5180) 
* close file handle

* some more open file handles

* must stay open

* remove this close, stays open
 2024-06-27 11:29:47 -07:00
chenyu
83da8b3558
 		use NV instead of CUDA in benchmark (#5192) 
also reenabled mixtral on green
 2024-06-27 13:52:58 -04:00
chenyu
0c6c7c5f7b
 		CACHELEVEL=0 -> IGNORE_BEAM_CACHE=1 in benchmark (#5191) 
ignoring beam cache but using compile cache should be fine, saved some benchmark time.

also updated `beam_search` to check flag value before accessing diskcache
 2024-06-27 13:15:18 -04:00
chenyu
c12de4f47d
 		benchmark use JITBEAM for llama and gpt2 (#5189) 2024-06-27 12:56:02 -04:00
chenyu
ad91962dcf
 		CACHECOLLECTING -> CAPTURING and don't capture clear_l2 (#5190) 
fixed first time BEAM slowness
 2024-06-27 12:32:28 -04:00
Roelof van Dijk
01e8838b65
 		ruff: suppressible-exception (#5182) 
* fix: use contextlib to suppress errors

* enable rule SIM105

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2024-06-27 08:23:44 -07:00
Roelof van Dijk
9704c7d4d4
 		ruff rule if-exp-instead-of-or-operator (FURB110) (#5178) 
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-06-27 08:22:19 -07:00
chenyu
5b8fda3c65
 		fix: JIT=0 means no JIT (#5188) 2024-06-27 10:31:37 -04:00
qazal
3af17849bf
 		safely parse quoted titles [run_process_replay] (#5183) 2024-06-27 16:39:48 +03:00
Roelof van Dijk
975b811ad9
 		names shadowing builtins (#5179) 
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-06-27 08:15:01 -04:00
Roelof van Dijk
26e254c42b
 		ruff: else-raise and else-return (#5175) 
* ruff: enable else-raise and else-return

* ruff: add error names

* fix order

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-06-27 07:54:59 -04:00
Roelof van Dijk
f88f71d73a
 		ruff: unnecessary-comprehension (#5174) 
* enable ruff C416 unnecessary-comprehension

* already a list
 2024-06-27 07:45:29 -04:00
reddyn12
f1c7944c44
 		Fix batchnorm shapes for resnet.load_pretrained (#5167) 
* Fix batchnorm shapes

* make it general reshape
 2024-06-26 18:44:10 -04:00
George Hotz
396ce6cfc9
 		clean up graph dedup function [run_process_replay] (#5169) 2024-06-26 15:07:34 -07:00
kormann
3a04e518ec
 		print_tree UPat +fix (#5132) 
* fix and extend print_tree

* typing

* typing

* fix upat

* fix none

* ws

* rm prefix

* mv luop dag

* typo

* test print_tree
 2024-06-26 15:02:19 -07:00
chenyu
0ba093dea0
 		hotfix: only validate stable diffusion when using threefry (#5166) 2024-06-26 16:50:38 -04:00
chenyu
e4a5870b36
 		validate stable_diffusion output (#5163) 
changed default steps, forgot to update validation
 2024-06-26 16:42:21 -04:00
nimlgen
21b225ac45
 		llama3 download works (#5160) 2024-06-26 22:45:13 +03:00
wozeparrot
c91b3c4079
 		shard llama3 on 0 sometimes (#5157) 2024-06-26 11:50:57 -07:00
Roelof van Dijk
294bd1a9ff
 		refactor: name check [run_process_replay] (#5158) 2024-06-26 11:39:41 -07:00
Roelof van Dijk
2c80583e14
 		perf: cache const UOp creation [run_process_replay] (#5156) 2024-06-26 11:13:14 -07:00
George Hotz
eda2824cd8
 		freeze uop [run_process_replay] (#5155) 2024-06-26 10:18:15 -07:00
Elias Wahl
e267f3161d
 		Add MLLogger (#5125) 
* add MLPerf logger

* eval steps

* start with step 1

* compliance for 3.1.0 and 4.0.0

* more compliance

* assert, comment and contiguous
 2024-06-26 12:23:56 -04:00
nimlgen
16405b973a
 		fix hcq sync (#5062) 
* fix hcq sync

* rewrite

* linter + comment

* fix profiler

* no default dict

* correct sync of unjitted transfer

* fix test
 2024-06-26 17:50:37 +03:00
David Hou
3604642847
 		Llama shard axis 0 sometimes (#5123) 
* make buffer view optional with a flag [run_process_replay]

* do not view when sharding to save memory [run_process_replay]

* llama shard axis=0 sometimes

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
Co-authored-by: chenyu <chenyu@fastmail.com>
 2024-06-26 10:35:25 -04:00
nimlgen
fd27f19e92
 		graph tests (#5153) 
* graph tests

* add test

* cleanup
 2024-06-26 16:31:20 +03:00
George Hotz
7b709c3ccd
 		switch tensorcoreoptions to tuple [run_process_replay] (#5143) 
* switch tensorcoreoptions to tuple [run_process_replay]

* localbuffer can stay namedtuple for now

* freeze LocalBuffer

* remove NamedTuple

---------

Co-authored-by: qazal <qazal.software@gmail.com>
 2024-06-26 14:12:53 +03:00
qazal
6ca7b13ed1
 		limit pickled objects [run_process_replay] (#5154) 
* limit pickled objects

* delete uop from the list

* debug metal

* need self.opts for TC

* dont need device

* [run_process_replay]

* minor
 2024-06-26 13:51:32 +03:00
George Hotz
ee4f080a14
 		rewrite div const [run_process_replay] [no_assert] (#5151) 
* rewrite div const [run_process_replay] [no_assert]

* Update uops.py
 2024-06-25 20:23:14 -07:00
David Hou
666a9c1448
 		don't view origin buffer when sharding (#5122) 
* make buffer view optional with a flag

* do not view when sharding to save memory
 2024-06-25 20:19:09 -07:00
George Hotz
89e106686a
 		simpler unmatch [run_process_replay] (#5149) 2024-06-25 19:57:40 -07:00
George Hotz
c98ca23cb9
 		test pickle variable (#5150) 
* test pickle variable

* fix process replay
 2024-06-25 19:49:21 -07:00
David Hou
8fcc41582f
 		make buffer view optional with a flag (#5120) 2024-06-25 19:13:20 -07:00
George Hotz
63ba2d05d1
 		uops dfs cleanup (#5147) 
* uops dfs cleanup

* Update uops.py
 2024-06-25 18:51:42 -07:00
George Hotz
6841ea3baf
 		don't allow duplicate variables (#5148) 2024-06-25 18:47:29 -07:00
George Hotz
cc7fafcd8b
 		sink folding rule [run_process_replay] (#5145) 2024-06-25 18:34:44 -07:00
Jhenner Tigreros
fa78755f19
 		Add new patterns to unfold division (#5139) 
* Add new patterns to unfold division

* Create regression test and fix pattern
 2024-06-25 18:07:47 -07:00
qazal
c4fdb9c725
 		second iteration on verify_lazyop (#5140) 2024-06-25 09:44:32 +03:00
chenyu
dade7677cf
 		validate llama3 output only with model "LLaMA-3/8B-SF-DPO" (#5138) 2024-06-24 20:58:25 -04:00
qazal
981afb114f
 		safely fold NEG in lazy.py (#5135) 
* safe

* add test
 2024-06-24 19:40:37 -04:00
chenyu
7948b05738
 		fix uneven shard with shrink and pad args on sharded axis (#5131) 
it's incorrect to assume all first (len(device)-1) shards would have the same size. e.g. size 2 shard 4 -> (1, 1, 0, 0)
 2024-06-24 16:55:50 -04:00
qazal
18e70deec3
 		verify_lazyop (#5124) 
* start verify_lazyop

* bfs order

* assert

* assert shapetrackers 2

* refactor

* more iteration

* skips

* that ast was wrong too
 2024-06-24 13:45:35 -07:00
qazal
fe707bc968
 		hotfix: don't use is for comparing dtype (#5128) 2024-06-24 14:12:34 -04:00
Jhenner Tigreros
dfa562dbc1
 		DEFINE_ACC takes UOps.CONST in vin instead of arg (#4975) 
* Change DEFINE_ACC to receive UOps.CONST in vin

* Use localtype instead of acc dtype

* Fix idp

* Fix copy list

* Fix warp

* Fix error

* Fix merge

* Fix testing

* Fix merge

* Use deepcopy

* Change to copy of inp

* Fix lint

* Move const to first place

* Fix issue upat

* Fix upat patterns

* Change to list, to test permutations

* Add condition

* Change pm

* Revert change pm

* Remove unused rule

* Fix

* Change of float4 DEFINE_ACC values

* Cast on PM to correct dtype

* Improve assert message

* Move IFs

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2024-06-24 09:25:33 -07:00
nimlgen
d84beaa6dd
 		tiny profiler cleanups (#5126) 2024-06-24 17:02:31 +03:00
chenyu
4a7d403777
 		cleanup test_multitensor (#5118) 
renamed d_zero, d0, d1, d2, ... to d0, d1, d2, d3 and reused some multi device tuples
 2024-06-23 20:54:22 -04:00
chenyu
c0ba5e0dfb
 		multi copy_to_device return the copy on same device if possible (#5117) 
previously it always returns from the first device
 2024-06-23 20:25:56 -04:00