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
11,378 commits 280 branches 15 tags 325 MiB
Commit graph

4,744 commits

Author SHA1 Message Date
George Hotz
1b879967c4 png test 2025-12-15 15:22:16 -04:00
George Hotz
d11f6d316d png works 2025-12-15 12:56:30 -04:00
wozeparrot
7ef7ce2856
 		tk reg local store (#13689) 2025-12-14 23:07:30 -08:00
George Hotz
572ca80046
 		fast tinygrad.apps.llm (#13685) 
* llm: add --benchmark support

* fix speed

* debug logging

* fix test attention
 2025-12-14 21:05:21 -05:00
chenyu
ed962786d6
 		use assign in Tensor.backward (#13674) 
preserve the grad object so that jit works
 2025-12-13 22:43:06 -05:00
George Hotz
55845f7de7
 		schedule: cache unbinds for consistent cache keys (#13664) 
* schedule: cache unbinds for consistent cache keys

strip BIND values before computing cache key so different bound values
(e.g. KV cache positions) hit the same schedule cache entry.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* spec: allow single-src BIND for schedule cache key normalization

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* docs: add lessons learned to CLAUDE.md

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* more claude.md

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
 2025-12-12 17:27:42 -05:00
George Hotz
8c87a0bf8d Revert "schedule: cache unbinds for consistent cache keys (#13662)" 
This reverts commit af86cae10c.
 2025-12-12 16:49:50 -05:00
George Hotz
af86cae10c
 		schedule: cache unbinds for consistent cache keys (#13662) 
* schedule: cache unbinds for consistent cache keys

different bound variable values (e.g. kv cache positions) now produce
the same schedule cache key by unbinding BIND(DEFINE_VAR, CONST) before
computing the cache key and rebinding after lookup.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* schedule: cache unbinds for consistent cache keys

When scheduling, BIND(DEFINE_VAR, CONST) nodes are now unbound to
tagged DEFINE_VARs before computing the cache key. This ensures that
the same computation with different bound values (e.g., different
KV cache positions in LLM) gets the same cache key and reuses the
cached schedule.

The fix:
- pm_pre_sched_cache: replaces BIND with tagged DEFINE_VAR
- pm_post_sched_cache: restores tagged DEFINE_VAR back to original BIND
- pm_remove_rangeify_tags: excludes DEFINE_VAR to preserve tags through rangeify
- var_vals extracted from BINDs before cache key computation

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* schedule: fix BIND handling and add CLAUDE.md

- Handle BIND to RANGE in create_schedule (not matched by CONST pattern)
- Assert all BINDs on same variable have same value
- Add CLAUDE.md codebase guide

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
 2025-12-12 16:40:10 -05:00
George Hotz
316da9f7ff
 		llm: add created/model fields, non-streaming support, and tests (#13660) 
* llm: add created/model fields, non-streaming support, and tests

- Add `created` timestamp and `model` fields to response (required by OpenAI spec)
- Add non-streaming mode support for /v1/chat/completions
- Add `send_data` helper to HTTPRequestHandler for responses with Content-Length
- Refactor viz/serve.py to use send_data
- Add integration tests using real OpenAI client

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* add openai to testing

* toml

* Remove 'openai' from dependencies

Removed 'openai' from the dependencies list.

* bump cache

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
 2025-12-12 14:50:36 -05:00
nimlgen
e36385e570
 		am: support xgmi systems (#13659) 
* am: support xgmi systems

* fake_am
 2025-12-12 18:55:45 +03:00
Jakob Sachs
ab2220b834
 		Handle missing bfloat16 natives on CPU architectures (#13553) 
* CPU: fix compiler-rt libcall by adding intermediate casts for bfloat16

* fix lint

* remove old manual bypass of bf16 for CPU tests, and add diversion converstion from bf16 to/from fp16

---------

Co-authored-by: Jakob Sachs <jakobs99@purelymail.com>
 2025-12-11 15:38:43 -05:00
chenyu
03600aef1e
 		failed test case when init jit with empty inputs (#13641) 
not related to bert grad acc, but still seems to be a bug
 2025-12-10 22:03:06 -05:00
Nino Risteski
76d465dbc3
 		optim empty shard #13513 (#13598) 
* optim empty shard

* remove tuple

* simplify

* lint

* lint2

* test

* remove original buffer unique id

* new rule

* reset shard

* update

* reset shard
 2025-12-09 12:28:36 -05:00
ayanhan
47a170be2e
 		test: enable cummax scalar IndexError test (#13625) 2025-12-09 12:25:56 -05:00
Christopher Milan
a17077d1d9
 		skip test_double_assign in CI LVP (#13620) 2025-12-08 14:54:02 -05:00
Christopher Milan
1c16b6e082
 		Mesa: freedreno (#12746) 
* ir3 init

* got a program

* 1 + 1 works

* use isa_disasm instead of shader_disasm

* wip

* matmul works

* works on py3.14

* fix const loading

* skip QCOM failing tests

* cleanup

* args actually work

* add compile-only tests

* fix typo and install tinymesa

* IR3 NULL backend

* (float32) images work

* autogen fix

* fix compile only test

* typo

* mypy happy

* compile-only uses py3.14

* bump mesa

* unify qcom disassembler

* float16 works

* disasm shows in viz

* save a line

* add real del

* variable workgroup sizes

* simplify diff

* bump line count

* properly set wgsz

* regen mesa

* no preamble

* bump lines
 2025-12-08 14:02:08 -05:00
Douglas Nyberg
947c6eefc3
 		add Swish op (#13541) 
* add Swish ONNX operator

* add Swish regression test

* remove trailing whitespace

* upgrade ONNX to 1.20, add excludes for unimplemented ops

* upgrade ONNX to 1.19, add Swish op

* upgrade ONNX to 1.19, TensorFlow to 2.18, add Swish op

* exclude attention_3d and attention_4d_gqa tests

* exclude attention fp16 tests

* exclude all attention tests

* retrigger CI

* retrigger CI - worker crash
 2025-12-08 12:41:18 -05:00
Christopher Milan
94d7646bdc
 		fix anonymous struct fields (#13610) 2025-12-07 12:56:38 -05:00
nimlgen
ac5f1e115d
 		autogen: repro for the bug (#13607) 
* autogen: repro for the test

* mute
 2025-12-07 15:51:03 +03:00
wozeparrot
93f1baca77
 		feat: tk fa in tensor (#13580) 2025-12-05 14:36:29 -08:00
George Hotz
c5bd28e21d
 		start work on schedule cache (#13529) 
* start work on schedule cache

* local unique

* schedule cache works

* schedule cache cleanup

* fix tests

* preserve metadata

* oops, fix cache

* put that there

* fix spec

* always miss

* why is that broken?

* src[0].op

* fix process replay

* delete abstractions2

* reenable the actual schedule cache

* metadata is best effort

* fix JIT in examples/gradaccum_mnist.py

* full jit

* fixed and test is real
 2025-12-04 17:24:49 -08:00
chenyu
42f6cf3a90
 		tighter test_real_world mem and kernel count bounds (#13573) 
also check if actual usage is within 20% of set limit, the old limits are too big to be useful
 2025-12-04 13:35:39 -05:00
chenyu
89f9e1dcd5
 		add SGD to beautiful_mnist (#13571) 2025-12-04 12:17:29 -05:00
Rory Clear
6eab756578
 		fix and test loading num_batches_tracked (#13538) 
* fix and test loading num_batches_tracked

* add failing reverse case

* try reshape state dict if mismatch

* reshape for () and (1,)

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
 2025-12-04 01:22:49 -08:00
Douglas Nyberg
a8a62bc08e
 		add max/min reduction support to ScatterND (#13562) 2025-12-04 00:53:47 -08:00
ayanhan
edf929ec9d
 		fix: add __delitem__ to Tensor with proper TypeError (#13561) 2025-12-04 00:53:08 -08:00
ayanhan
92b40290c7
 		fix: add test_sum_int and remove outdated TODO in test_custom_kernel (#13560) 2025-12-03 21:51:58 -05:00
Christopher Milan
0a54434b15
 		mitigate ctypes c_bool bitfield bug (#13558) 
* mitigate ctypes c_bool bitfield bug

* don't delete old test
 2025-12-03 20:46:04 -05:00
George Hotz
24ca8eeaa7
 		small fixups from schedule_cache (#13557) 2025-12-03 15:41:16 -08:00
Douglas Nyberg
f5abd38132
 		remove tfa dependency: use keras.optimizers.Lamb and tf.raw_ops for LARS (#13555) 2025-12-03 17:48:27 -05:00
George Hotz
a4c4e48385
 		add LUNIQUE op (#13554) 2025-12-03 14:34:34 -08:00
chenyu
22777a89ea
 		minor test_uop_symbolic updates (#13551) 2025-12-03 13:17:44 -05:00
chenyu
a205f98ef4
 		tighter bound for MOD (#13550) 2025-12-03 11:24:29 -05:00
nimlgen
549f3287a8
 		fix caching for fetch (#13544) 2025-12-03 14:34:14 +03:00
George Hotz
6bd355fa26
 		add needs_second_gpu decorator (#13543) 
* add needs_second_gpu decorator

* more skips

* two more fixes
 2025-12-02 19:08:23 -08:00
wozeparrot
0d55aec605
 		fix after end (#13542) 2025-12-02 18:42:58 -08:00
George Hotz
055d5aeb7f add external_test_process_count 2025-12-02 17:26:30 -08:00
chenyu
e8879f7e31
 		match torch clamp backward (#13533) 
* match torch clamp backward

* fix PYTHON
 2025-12-02 17:58:32 -05:00
Roelof van Dijk
c158e3c988
 		add cifar gated uop_given_valid regression test (#13536) 2025-12-02 16:02:47 -05:00
Roelof van Dijk
e329baffa7
 		fix cifar while keeping openpilot fused (#13528) 
* this works

* test now passes
 2025-12-02 12:05:56 -08:00
nimlgen
0874ba8cc8
 		test_hevc: do not download the whole file (#13531) 
* test_hevc: do not download the whole file

* fix
 2025-12-02 21:31:28 +03:00
qazal
366badaa68
 		require renderer argument in get_program, removes device opening in process replay [pr] (#13524) 2025-12-03 02:05:31 +08:00
Douglas Nyberg
6a7c58abf1
 		fix(onnx): unwrap list/tuple value in Pad op (#13500) 
* fix(onnx): unwrap list/tuple value in Pad op

* add regression test for Pad list value

* remove trailing whitespace

* use _resolve_const for Pad constant_value
 2025-12-02 07:47:20 -08:00
nimlgen
77a76d1b13
 		device: respect compiler ContextVars (#13523) 
* device: envvars for cc

* fix

* fix

* x

* um

* fix

* remote

* em

* cleanup

* typing

* fix

* debug

* lvp?

* ugh

* singl

* rm

* lol

* fix

* ?

* this?

* why?

* rev

* mod test

* l
 2025-12-02 14:42:04 +03:00
wozeparrot
1b7dbfb37f
 		tk: named kernels + per kernel range id (#13522) 2025-12-01 22:51:04 -08:00
nimlgen
455dd88236
 		nv: minimal hevc (#13502) 
* nv: minimal hevc

* validate

* not needed

* tralin

* var

* cpu

* fxi

* desc

* move

* cleanup
 2025-11-30 16:46:55 +03:00
George Hotz
fd373fea7a
 		fix a few tests [pr] (#13498) 2025-11-29 13:43:45 -08:00
George Hotz
6a140f74fe
 		split out unique_const and cache const [pr] (#13493) 
* split out unique_const

* add cache to const

* call const in unique_const
 2025-11-29 10:44:28 -08:00
George Hotz
c38b7684dc
 		improve microbenchmarks (#13492) 
* improve microbenchmarks

* bugfix + ubench

* lil

* no src in const method
 2025-11-29 10:15:22 -08:00
kamilisjon
3d76ef9ba8
 		Update tests (#13479) 2025-11-28 18:35:28 -08:00