nimlgen
dc10187fc0
am: add am_smi ( #8739 )
...
* am: start monitor
* cleanups
* fixes
* hmm
* progress
* cleanup
2025-01-24 20:16:19 +03:00
George Hotz
7a2223a6c6
add merge views to ops_folding [pr] ( #8051 )
...
Co-authored-by: qazal <qazal.software@gmail.com >
2025-01-24 17:45:11 +02:00
qazal
0814a79cb4
cleanup the merge_views upats [pr] ( #8738 )
2025-01-24 16:49:54 +02:00
qazal
07069b9988
rename to tensor_uop [pr] ( #8737 )
2025-01-24 13:42:25 +02:00
George Hotz
e82ba1454b
MultiLazyBuffer is UOp [pr] ( #8662 )
...
* MultiLazyBuffer is UOp [pr]
* this is new mlb
* this is the idea
* progress
* multitensor works
* more movement ops
* this
* MultiLazyBuffer is UOp
* cleanups
* multi axis
* fix more tests
* work
* not that
* add multi grad and move shard to ops
* mops not views
* no double contig
* sweet, all mt tests passing
* port old logic
* remove lbs
* fix realized
* whitespace
* assign tweak
* test_assign_kv_cache_multi passes
* fix is_realized
* fix JIT for multi
* just a few more lines i'll pay them back soon i swear please bro just a few more
* no split reduceop for multi
2025-01-24 13:28:55 +09:00
chenyu
eb77488f85
update llama3 70B to use R1 ( #8733 )
2025-01-23 19:06:05 -05:00
George Hotz
3e987fc856
add device print with -m tinygrad.device [pr] ( #8729 )
...
* add device print with -m tinygrad.device [pr]
* fix linter
2025-01-24 05:46:27 +09:00
geohotstan
04846b91aa
reorder and categorize onnx_ops ( #8731 )
...
* new order
* remove a todo
* constant node is definitely requires_grad false
* one new line spacing
* property and graph
* oops linter
2025-01-23 13:18:54 -05:00
qazal
8e5bd0cd7a
fix buffer init and skip test_swizzle_failure_permute [pr] ( #8732 )
...
* fix buffer init and skip test_swizzle_failure_permute [pr]
* replace preload with just load
* add
2025-01-23 17:21:38 +02:00
nimlgen
e4512baea4
am: cleanup mm ( #8730 )
...
* am: cleanup mm
* cle
* ops
* entries
2025-01-23 15:49:37 +03:00
qazal
07ec99001a
keep VIEW in big_sink + copy of buffer view spec [pr] ( #8727 )
...
* keep views in sink [pr]
* tests
* things from the gpt2 bug
2025-01-23 11:29:30 +02:00
qazal
6cb74bb630
fix using clone with shrink [pr] ( #8724 )
...
* fix using clone with shrink [pr]
* remove extra arg, add test_clone_with_shrink_realized
2025-01-23 08:28:07 +02:00
chenyu
af65331b76
update beam params for bert green [pr] ( #8726 )
...
increase BEAM_UPCAST_MAX and BEAM_LOCAL_MAX to default and matched red. 3% faster step
2025-01-22 22:00:05 -05:00
qazal
907dfa0e82
image buffer realization spec [pr] ( #8420 )
...
* image buffer realization spec [pr]
* redo the spec
* work
2025-01-22 20:25:22 +02:00
chenyu
49b914ee69
simpler bert acc [pr] ( #8714 )
...
logit.log_softmax().argmax(-1) is equivalent to logit.argmax(-1)
2025-01-22 10:32:19 -05:00
nimlgen
93fb50ce77
allreduce: add flags ( #8713 )
2025-01-22 17:44:31 +03:00
qazal
891436853d
remove buffer size check in schedule item [pr] ( #8712 )
2025-01-22 13:36:30 +02:00
qazal
2dae467b75
scheduler + process_replay import cleanup ( #8711 )
2025-01-22 12:44:07 +02:00
qazal
e3d1464ba4
move assign preload out of schedule item [pr] ( #8710 )
...
* move assign preload out of schedule item [pr]
* fix that
2025-01-22 12:43:57 +02:00
chenyu
9a9079118e
envvar BERT_LAYERS [pr] ( #8709 )
...
default is 24 for large
2025-01-21 22:49:19 -05:00
chenyu
9f6d545a16
bert log global_norm in training step [pr] ( #8708 )
...
* bert log global_norm in training step [pr]
and minor cleanups
* .item()
2025-01-21 20:36:27 -05:00
nimlgen
c5e46c5eee
am: recover from any boot interrupt ( #8703 )
...
* am: recover from any load interrupt
* add fuzzer
* nu
2025-01-21 22:22:23 +03:00
chenyu
1e283c33d3
remove realize in bert model init [pr] ( #8707 )
2025-01-21 14:11:03 -05:00
George Hotz
018edd934b
don't use view in copy [pr] ( #8704 )
...
* don't use view in copy [pr]
* oh, remove double contig
* fix reps
2025-01-21 09:57:47 -08:00
qazal
d6bf1feaab
remove the "no copy" line from copy_to_device ( #8702 )
...
* delete the no copy one
* add tests
2025-01-21 17:09:33 +02:00
nimlgen
3628f89929
fix deallocate for subbuffers ( #8701 )
...
* fix deallocate for subbuffers
* forgot this
* rm name
* hmm
2025-01-21 16:34:19 +03:00
nimlgen
6733a3a96b
am: fix typo ( #8700 )
2025-01-21 14:35:15 +03:00
qazal
f0d424ecdf
Tensor UOps can become a buffer or const after scheduling ( #8698 )
...
* spec
* work
* update test_viewed_consts_do_not_realize
* remove
2025-01-21 12:33:19 +02:00
qazal
e2008c98c3
allow symbolic shape in tensor const parents [pr] ( #8699 )
2025-01-21 12:01:25 +02:00
nimlgen
2b239db5d2
temp() with usernames ( #8697 )
2025-01-21 12:26:43 +03:00
qazal
66ac0087e8
more high level contiguous tests + scheduler deletions [pr] ( #8695 )
...
* delete those
* move the upat too
* rename ops_folding to just sym
* keep that
2025-01-21 01:52:58 +02:00
qazal
08eb1f1f56
simplify tensors before scheduling [pr] ( #8580 )
...
* delete forced_realize
* put that back
* work
* remove forced_realize
* expectedFailures
* contiguous(buffer)
* multi
* expectedFailures
* cleaner create_subbuffer
* more comments
* remove that
* note
* realizes
* work
* one upat and image is back
* remove
* cleaner
* fix test_complex_backward for now
---------
Co-authored-by: George Hotz <geohot@gmail.com >
2025-01-20 23:42:42 +02:00
qazal
02ad450e22
add failing assert for gradient realization [pr] ( #8692 )
2025-01-20 22:50:09 +02:00
qazal
b14c9848cc
small changes to make the tensor_map_simple diff cleaner [pr] ( #8691 )
2025-01-20 22:25:59 +02:00
Sieds Lykles
1a15c0e89d
Move define_acc down an unrolled add chain ( #8404 )
...
* Move define_acc down an unrolled add chain
* Prevent possible infinite recursion
* Add test
* Fix typo in test
* Move mulacc_unrolled to devoctorize + load_store_indexing pass
* Add test for mulacc_unrolled by itself
* undo formatter
* import from ops, not rewriter
* Add a const version
---------
Co-authored-by: chenyu <chenyu@fastmail.com >
2025-01-20 14:56:27 -05:00
geohotstan
dd82b4c913
make onnx runner a class ( #8647 )
...
* this
* clean up
* more clean ups and improve debug msg
* more correct training toggler
* remove manual training toggling
* change some variable names
* actually just add the training toggle for LIMIT envvar too
* more refinement
* __call__ and OnnxRunner
* fix half pylint, other half is importing from onnx while this file is onnx.py, figure out later
* ahhhh found another mistake
* remove limit from __call__
---------
Co-authored-by: chenyu <chenyu@fastmail.com >
2025-01-20 10:11:05 -08:00
George Hotz
46a8c5e1e5
delete forced_realize ( #8615 )
...
* delete forced_realize
* put that back
* expectedFailures
* cleaner create_subbuffer
* more comments
---------
Co-authored-by: qazal <qazal.software@gmail.com >
Co-authored-by: qazal <77887910+Qazalin@users.noreply.github.com >
2025-01-20 09:40:36 -08:00
chenyu
679b1ad058
move softmax upcast to after subtracting max ( #8684 )
...
* move softmax upcast to after subtracting max
max can always be done in the same dtype without any numerical loss, so this is better when explicitly upcasting in softmax
* skipUnless half
2025-01-20 12:16:32 -05:00
nimlgen
08ca871d77
am: remove pm block ( #8688 )
...
* am: remove pm block
* hm
* oops
2025-01-20 18:05:22 +03:00
nimlgen
9d3c40601f
am: fast memory manager ( #8654 )
...
* start
* progress
* fixes
* smth
* mini fixes
* fix2
* ugh, need this for now
* faster
* cleanups
* tiny linters
* make mypy happier
* test & free pts
* ops
* linter
* cleanup vm
* fix
* remove map_from
* tiny fixes
* add test to ci
2025-01-20 16:58:22 +03:00
qazal
9e55495b4d
fold double contiguous [pr] ( #8687 )
2025-01-20 14:38:33 +02:00
qazal
ed63ff2372
Remove contiguous on buffer ( #8676 )
...
* remove contiguous on buffer
* spec
* make things that can't be images not images
2025-01-20 13:48:33 +02:00
qazal
3499a2c72d
start moving image things to rewrite rules ( #8678 )
...
* start moving image things to rewrite rules [pr]
* that too
* as expected
* fix
* Revert "fix"
This reverts commit fd03c9464b
2025-01-20 13:34:29 +02:00
qazal
b1847d561f
smaller do_realize and some cleanups [pr] ( #8685 )
...
* do_realize cleanups [pr]
* cleanup assign
* unwrap ShapeTracker as we expect it to exist
2025-01-20 12:47:01 +02:00
qazal
689bf68cfc
remove GroupOp.Meta [pr] ( #8686 )
2025-01-20 12:24:19 +02:00
George Hotz
4198bce150
_apply_map_to_tensors [pr] ( #8683 )
2025-01-19 17:56:04 -08:00
George Hotz
98d01a059d
rename uopgraph to rewriter [pr] ( #8682 )
2025-01-19 17:03:12 -08:00
Ignacio Sica
f532c78889
minor space hotfix ( #8679 )
2025-01-19 17:00:24 -08:00
chenyu
2d0842386d
fix parse_valid for float uop ( #8681 )
...
x < c -> X <= c-1 only works for int
2025-01-19 18:15:49 -05:00
George Hotz
168c16646a
change create_schedule_with_vars api to big_sink [pr] ( #8677 )
2025-01-19 13:30:26 -08:00
