nimlgen/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity
14,335 Commits 334 Branches 15 Tags
rewrite_rangeify2
Commit Graph
6094 Commits
Author SHA1 Message Date
chenyuandGitHub 5ca87f1bac fix cast to weak twice [pr] (#17548) 
also no gradient for weak target
 2026-08-15 12:39:54 -04:00
chenyuandGitHub 26cbadd69a no pm_fold_cast_const in full_rewrite_to_sink [PR] (#17547) 2026-08-15 10:18:59 -04:00
chenyuandGitHub fae893753b no pm_fold_cast_const in get_kernel_graph [pr] (#17546) 
* no pm_fold_cast_const in get_kernel_graph [pr]

* maybe
 2026-08-15 09:57:05 -04:00
chenyuandGitHub 539a03343a no casted const from sub and div [pr] (#17543) 2026-08-15 07:47:40 -04:00
qazalandGitHub a57569349c renumber invalids before callify (#17542) 
* renumber invalids before callify

* change

* Revert "change"

This reverts commit 6f4df1541e.

* renumber in tensor

* scope renumber_invalid_outputs

* cleanup
 2026-08-15 18:00:54 +09:00
qazalandGitHub 5c43a89fb1 precompile_backward tests for sched_cache (#17544) 
* work

* back

* work

* keep +
 2026-08-15 15:25:31 +09:00
qazalandGitHub e6f5bb9c09 simple test for Invalid clone cache miss regression (#17541) 
* simple test for Invalid clone cache miss regression

* xfail

* _
 2026-08-15 11:00:21 +09:00
chenyuandGitHub 4b0525e594 no pm_fold_cast_const in UOp.simplify and hcq2 [pr] (#17540) 2026-08-14 21:39:47 -04:00
chenyuandGitHub 13c381b0c0 remove pm_fold_cast_const from initial symbolic [pr] (#17535) 
interestingly it gives more accurate numerics when composing const like log10
 2026-08-14 14:14:40 -04:00
chenyuandGitHub ac7067ac60 fix deconstruct_function for python 3.11 (#17534) 2026-08-14 13:38:34 -04:00
chenyuandGitHub 80169c6758 remove where push cast to branches from sym [pr] (#17533) 
* remove where push cast to branches from sym [pr]

not really needed and one less place that generates casted weak const when it's not needed

* fix
 2026-08-14 12:57:56 -04:00
chenyuandGitHub 89ab344c42 fix assign into bitcast with no explicit realize (#17531) 2026-08-14 09:23:01 -04:00
nimlgenandGitHub 6b35220622 cpu hcq2 (#17503) 
* cpu hcq2

* temp

* slop

* test with backpressure

* x

* x

* x

* x

* x

* x

* Dx

* save reverts

* um?

* x

* x

* call from py

* x?

* x

* submitters gone

* x

* x

* z

* Dx

* Dx

* x

* x

* fixes

* repl

* x

* f

* for now keep hcqbuffer
 2026-08-14 15:06:03 +03:00
George HotzandGitHub b1859805b1 remove Ops.BIND (#17511) 
* remove Ops.BIND

* param arg

* simplify that

* simplify

* cleaner

* props, not functions

* param and buffer can share
 2026-08-13 23:52:16 -07:00
qazalandGitHub faba071b1d don't enter CALL body in assign fixups (#17527) 
* fix python time regression in mxfp4

* this saves even more time

* s_nop test

* itertools count

* cleanup
 2026-08-14 15:05:04 +09:00
qazalandGitHub 16c5ff2490 add external_benchmark_all2all.py (#17507) 
* add external_benchmark_all2all.py

* mv

* more minimal

* less

* fix space
 2026-08-13 15:50:43 +09:00
George HotzandGitHub e103fb2a10 more lil llm improvements (#17514) 
* more lil llm improvements

* default float
 2026-08-12 20:13:20 -07:00
George HotzandGitHub 2297118541 lil llm improvements (#17513) 2026-08-12 19:29:32 -07:00
geohot ff0cb28c21 skip slow whisper tests 2026-08-12 13:12:37 -07:00
qazalandGitHub ed8297a102 kerenl opts test from nan in llama 8b (#17510) 
* all2all

* nan

* remove that

* less

* has_local

* only the nan change here

* use nice getitem syntax for INDEX

* work

* remove

* even simpler
 2026-08-13 04:07:30 +09:00
RaineandGitHub de04781b36 simplify equivalent const max (#17505) 
* add const max folds

* add regression test

* move
 2026-08-12 08:39:29 -07:00
qazalandGitHub 2e5a9a4121 no hardcoded device names in test_sliced_buffer_function (#17501) 2026-08-12 15:19:25 +09:00
sirhcmandGitHub 479ffb0cda remove Ops.SLICE (#17492) 2026-08-11 18:50:04 -04:00
George HotzandGitHub 7edf80a48a small changes from new rangeify + remove flaky tc tests (#17486) 
* small changes from new rangeify

* remove test/opt/test_tensor_cores.py
 2026-08-10 15:58:49 -07:00
RaineandGitHub d41ca5e60f Fix WMMA CI (#17479) 
* init

* split into sub tests

* trigger ci
 2026-08-10 08:39:14 -07:00
nimlgenandGitHub e29606f07e hcq2: copy kernel (#17480) 
* hcq2: copy with kernel

* test

* x
 2026-08-10 17:28:46 +03:00
nimlgenandGitHub 8611fe22a7 fix hevc (#17477) 
* hevc tests

* x
 2026-08-10 13:33:49 +03:00
qazalandGitHub 44f1f45cd5 llama: custom silu kernels (#17462) 
* start by copying the C

* uop kernel

* cleanup tests

* estimates is part of SPEC
 2026-08-10 16:43:01 +09:00
George HotzandGitHub d4d537c8ae add SPEC checking for the kernel graph (#17432) 
* add SPEC checking for the kernel graph

* skip test

* raise there, not None

* handwritten

* issue with unshard

* multi works

* and bitcast

* fix new tests

* fix linear

* remove call index

* fix shrink

* fixes
 2026-08-08 10:00:06 -07:00
b1tgandGitHub abe2256299 fix symbolic sharded reshape (#17463) 2026-08-08 09:18:02 -07:00
b1tgandGitHub 8c49a7a34b support symbolic shapes in copy (#17461) 
* pad_to is no-op when same shape

* support symbolic shapes in copy
 2026-08-08 09:16:59 -07:00
nimlgenandGitHub 4c206a52b1 fix ci emu (gpt) (#17437) 
* fix ci emu

* x
 2026-08-07 22:36:59 +03:00
chenyuandGitHub 4a3b8f6501 better _drop_valid_stmts [pr] (#17454) 2026-08-07 15:18:35 -04:00
chenyuandGitHub f76422b8af fix cast to float _min_max [pr] (#17451) 2026-08-07 11:56:19 -04:00
Robert JosephandGitHub 0c96cdc300 fix prod gradients at zero (#17404) 2026-08-07 09:56:52 -04:00
chenyuandGitHub baa6148066 fix var of large half input (#17444) 
* fix var of large half input

similar to mean, we use sum_acc_dtype for denominator

* mypy
 2026-08-06 23:14:32 -04:00
qazalGitHubgeohot
 		f253c4469d remove contiguous from custom_kernel (#17149) 
* no user contig on custom_kernel

* clean up

* non removable

* test MXFP4 llama without hipcc

* use compiler-free HIPCC renderer in llama CI

* move llama coverage to AMD tests

* run llama coverage in AMD test matrix

* respect configured ROCm path in llama profile

* work

* clean up

* fix

* add views back

* remove that

* update test

* test_double_permute one less kernel

* test_shrink less kernels

---------

Co-authored-by: George Hotz <[email protected]>
 2026-08-07 11:20:59 +09:00
chenyuandGitHub 28195d51fb fix f2f from fp8e5m2fnuz to half (#17442) 
* fix f2f from fp8e5m2fnuz to half

* it works if it's supported
 2026-08-06 21:19:46 -04:00
chenyuandGitHub 9020a88f03 truncate float in DType.const [pr] (#17439) 2026-08-06 20:19:08 -04:00
qazalandGitHub f258708d7d llama: custom quantize_mxfp4+transpose kernel (codex) (#17434) 
* llama: custom quantize_mxfp4+transpose kernel (codex)

* rename to cpp

* inline

* cleanup

* lds load_bf16x4

* more tests, add Estimates
 2026-08-06 16:13:28 +09:00
chenyuandGitHub 28e6ef6937 fix postopt symbolic [pr] (#17433) 
REDUCE with src simplied to const would become unparented
 2026-08-06 00:12:30 -04:00
chenyuandGitHub 7a9cd8e329 move weak function and pm to uop/weak [PR] (#17429) 2026-08-05 22:27:40 -04:00
George HotzandGitHub b4372df9c6 revert wrong custom kernel fix (#17427) 2026-08-05 18:31:35 -07:00
sirhcmandGitHub be25207a7a scope variable names inside CALLs (#17424) 2026-08-05 20:59:09 -04:00
chenyuandGitHub d726e5f7f3 split pm_fold_cast_const [PR] (#17425) 
need to delete this rule that writes strong typed CONST
 2026-08-05 19:56:04 -04:00
George HotzandGitHub 470c032a5e fix slice + non contig kernels (#17423) 
* movement: resolve negative int slice bounds against symbolic sizes

negative int bounds in a slice against a symbolic dim were passed through
unresolved, giving wrong views. resolve them against the (possibly
symbolic) size, like slice.indices does for int dims

* schedule: realize custom kernel inputs that don't resolve to a buffer state

rangeify assigns ranges backward from consumers and CALL contributes none,
so the subgraph above a custom kernel input gets no ranges unless something
in it is realized, and reduce conversion crashes with a KeyError. realize
call inputs that don't resolve to a buffer state.

only view-only movement ops preserve the underlying buffer: anything
computed (ALU, REDUCE, ...) must be realized even if one of its sources
resolves to a buffer, since the whole subgraph above the call has no
ranges. unwrapping src[0] unconditionally missed const branches hanging
off non-src[0] children and silently resolved REDUCEs to their source
buffer. includes regression tests for pure const, mixed buffer+const, and
view-over-buffer inputs
 2026-08-05 16:18:17 -07:00
George HotzandGitHub 581bfdd94f merge track_rewrites and profile_matches into rewrite_group [PR] (#17420) 
* merge track_rewrites and profile_matches into rewrite_group

* bug

* flip ctx polarity
 2026-08-05 14:41:38 -07:00
chenyuandGitHub c2f1e5ae2a fix weak cast to strong dtype [pr] (#17418) 
weak can mean higher than that strong dtype, so always use that strong dtype is wrong
 2026-08-05 15:32:23 -04:00
chenyuandGitHub 6cb419b9b7 regression test for bert nan with weak (#17412) 2026-08-05 12:14:26 -04:00
chenyuandGitHub 3bf9e70b19 Revert "don't cast weak in _broadcasted [pr] (#17408)" (#17409) 
This reverts commit b45058b5ec.
 2026-08-05 02:40:33 -04:00