github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity
12,896 Commits 248 Branches 14 Tags
cb681da8405b0b3ffe91cdeee3f33dd06ecb7aa4
Commit Graph

1139 Commits

Author SHA1 Message Date
George Hotz
35e3983840 Add Q5_0, Q5_1, and bfloat16 GGUF types (#15644) 2026-04-08 17:16:19 +08:00
Christopher Milan
acf239e4d2 specify renderer in DEV, <dev>_<ren>=1 is deprecated (#15551) 2026-03-31 18:35:14 -04:00
nimlgen
5181c8e23a llm: fix nan in kvcache (#15552) 2026-04-01 00:38:45 +03:00
b1tg
a63392a565 llm: pairwise ranking topk for MoE expert selection (#15499) 2026-03-31 12:46:39 +08:00
nimlgen
0d6fc0f571 jit: graphing in uops (#15489) 
* jit: graphing as rewrite rule

* f

* +metal,cuda

* x

* cl

* x

* x

* simpler

* f

* m

* x

* revert?

* revert2

* back

* back

* t

* x

* m

* x

* c

* x

* l

* x

* comment

* smaller

* rv

* x

* x
 2026-03-27 19:09:02 +03:00
Christopher Milan
bc180a963c deprecate <dev>=1 in favor of DEV=<dev> (#15467) 
* start work on target

* add test

* update actions to use DEV

* update docs

* update readmes

* tests need that too

* update example

* update tests (comments)

* fix that test

* ruff

* mypy

* oops

* remove getenvs

* don't add Target yet

* and the test

* lint

* and docs

* more stuff

* assert

* few more fixes

* test assert
 2026-03-26 03:48:03 -04:00
George Hotz
fe2690399b llm: support assistant prefill + refactor to TransformerConfig (#15457) 
* llm: support assistant prefill

* refactor to ModelConfig

* TransformerConfig

* more
 2026-03-25 10:50:48 +08:00
George Hotz
a33ac869aa llm server: temperature + test client (#15444) 
* improvements to the llm server

* eval script

* eval llm

* better eval gets 58.71

* cleanups

* add temperature, but multinomial is absurdly slow

* claude is so smart

* lint

* remove slop

* no more stop
 2026-03-24 21:07:15 +08:00
chenyu
c491345766 pass device into Tensor._frompy (#15385) 
* pass device into Tensor._frompy

with this, canonicalize_device is the only usage of Device in tensor.py

* export_model.py
 2026-03-20 05:09:01 -04:00
George Hotz
3b75d8a7a2 fix double after bug in rangeify (#15381) 2026-03-20 14:53:46 +08:00
Christopher Milan
0c89340a1e automatically emulate unsupported (tiny) floats [skip_process_replay] (#15366) 2026-03-20 02:31:44 -04:00
chenyu
bf33c5f796 remove gradient materialize_grads (#15367) 
effectively default to True

and removed *0 hack in Tensor.copysign. now dy/dx=0 if y does not depend on x

remove
 2026-03-19 23:36:03 -04:00
chenyu
fceb21c315 Tensor(uop) uses device from uop (#15340) 2026-03-18 02:56:06 -04:00
George Hotz
9d95321be3 set allow_implicit=False by default (#15319) 
* set allow_implicit=False by default

* modernize beautiful mnist
 2026-03-17 17:14:38 +08:00
George Hotz
584ec75aa2 precompile backward (#15311) 
* add precompile backward support

* cleanups

* fix

* compact grad

* split v not split

* simpler

* no NOOPT
 2026-03-17 15:28:40 +08:00
b1tg
856a839efc llm: fix qwen3 moe topk renormalization (#15201) 2026-03-17 12:57:33 +08:00
George Hotz
3ff03be413 call always has tuple (#15297) 
* call always has tuple

* fix pre-commit and simplify

* update

* fix

* move that assert

* tuple

* fix multi

* cleanups

* fix merge
 2026-03-17 10:58:46 +08:00
wozeparrot
674c760974 embedded bwd vocab shard (#15001) 
* fix: remove more multi from call

* feat: embedding bwd vocab sharding

* clean: unused import

* clean: don't actually need this pattern
 2026-03-16 19:37:16 -07:00
chenyu
02afb45f29 remove UOp.assign [pr] (#15300) 
* remove UOp.assign [pr]

it's all store and after, UOp is immutable

* fix test
 2026-03-16 21:45:41 -04:00
chenyu
3e2b7803e6 view assign replaces at buffer identity (#15298) 
matches what functions capture
 2026-03-16 19:58:38 -04:00
George Hotz
476276f4b4 support grads on tuples (#15287) 
* support grads on tuples

* simpler

* grad_fxn works

* cleanups

* unused
 2026-03-16 17:39:34 +08:00
George Hotz
08662bc4ab add TUPLE/GETTUPLE, simple tests pass (#15286) 
* simple tuple stuff passes

* resolved
 2026-03-16 15:06:02 +08:00
chenyu
cd14e8e64b allocations contiguous is store+after (#15280) 2026-03-15 11:58:40 -04:00
Sieds Lykles
4b59083d7c assign into empty works (#15256) 2026-03-13 10:24:29 -04:00
chenyu
018c01508d test case for call precompile multi (#15254) 2026-03-13 06:28:43 -04:00
b1tg
18dc77ccab add fp8 fnuz dtypes with PYTHON backend support (#14945) 
* add fp8 fnuz dtypes with PYTHON backend support

* rm emu related change

* clarify fp8 fnuz zero handling

* Revert "rm emu related change"

This reverts commit efa4763c22.

---------

Co-authored-by: b1tg <b1tg@users.noreply.github.com>
Co-authored-by: chenyu <chenyu@fastmail.com>
 2026-03-11 22:30:18 -04:00
George Hotz
4f3f55328b do not patch on invalid tensor tests (#15226) 
* do not patch on invalid tensor tests

* cleanup
 2026-03-12 09:35:20 +08:00
Christopher Milan
2fb8a7f60f fix test_invalid_tensor when before values are nan (#15215) 2026-03-10 23:51:19 -04:00
Christopher Milan
ffaafd391a Invalid in Tensor (#15154) 2026-03-10 02:49:54 -04:00
chenyu
a53187eef7 fix TestPartialAssignToSharedBuffer (#15202) 
bufferize_to_store issue with assign
 2026-03-09 23:14:23 -04:00
b1tg
891a73befc llm: fix chunked prefill (#15182) 
* llm: fix chunked prefill

* less lines

---------

Co-authored-by: b1tg <b1tg@users.noreply.github.com>
 2026-03-07 22:08:31 +08:00
Ananta Ranganathan
5bdad8ee41 update mxfp4 tests to use the same patterns as the others (#15177) 
* update mxfp4 tests to use the same patterns as the others

* fix typo in test call not sure how it committed
 2026-03-06 13:21:40 -05:00
Ananta Ranganathan
5c50035e0d avoid using arithmetic for mxfp4 (#15172) 
* avoid using arithmetic for mxfp4

* update tests to use assert equal

* no longer todo
 2026-03-06 11:17:56 -05:00
Roelof van Dijk
059c6326c0 metal uint32 icb offset overflow (#15156) 
* metal uint32 icb offset overflow

fix: diff

supports_exec_item

GraphRunner.supports_exec_item

tests

fix: can't import on non-metal

stricter

* also test the non-metal buffer case

* imports on non-mac
 2026-03-06 00:54:39 +03:00
Ananta Ranganathan
8ef656324e FIXED TEST Q5_K GGUF dequant (#15147) 
* q5_k gguf support as separate pr

* fix the problematic gemv test for q5_k

* add assert to make sure the gemv test cant fail with warning instead of error
 2026-03-05 16:32:36 +08:00
George Hotz
e97922a57c LLM speedup with two jits, prefill/rollout (#15153) 
* START_TIME

* print cleanup

* fix tests
 2026-03-05 16:21:09 +08:00
George Hotz
fb43b415f9 fix symbolic shape call + chunked prefill (#15149) 
* fix precompile for symbolic shape

* chunked prefill

* cleaner

* test that
 2026-03-05 14:02:26 +08:00
George Hotz
ac1847cbf7 fully symbolic llm (#15097) 
* work

* llm symbolic (almost)

* work

* revert that

* llm sym

* works

* cleanups

* cache tokens with the kv cache

* cleanups

* cleanups
 2026-03-05 10:22:11 +08:00
chenyu
34594bcaaf Revert "bug in metal: offset is stored as uint32, overflow (#15129)" (#15136) 
This reverts commit 9c58db16fa.
 2026-03-04 16:54:42 -05:00
Roelof van Dijk
9c58db16fa bug in metal: offset is stored as uint32, overflow (#15129) 
* metal uint32 icb offset overflow

* fix: diff

* supports_exec_item

* GraphRunner.supports_exec_item

* tests

* fix: can't import on non-metal
 2026-03-04 22:52:12 +03:00
chenyu
fae400d300 update assign tests to also test the expected behavior (#15132) 2026-03-04 11:34:43 -05:00
chenyu
1f96cc2b51 update non-contiguous buffer error message [pr] (#15131) 
* update non-contiguous buffer error message [pr]

also cleaned up the tests

* order
 2026-03-04 11:13:26 -05:00
George Hotz
01ddb4c267 add precompile to call (#15099) 
* add precompile to call

* put get back

* something

* after structure

* alt

* keep it call

* resolve call

* resolve linear call

* precompile works with llm

* revert rangeify

* color for debugging

* getenv PRECOMPILE

* clean up deco pattern

* fully recursive sink scheduling

* revert llama

* fix SPEC=2
 2026-03-03 22:32:42 +08:00
chenyu
5dcf29b1a0 use clone in test_swap_slices (#15096) 2026-03-02 22:05:12 -05:00
George Hotz
d483e4153a buffer view is like buffer (#15082) 
* buffer view is like buffer

* fix

* swap_reshape_shrink

* contiguous on gguf, fix overlap

* revert that

* _device_supports_view

* this

* fix that test

* 0 buffers

* that test was wrong

* this

* check correct size

* contig BUFFER_VIEW

* this

* fix tests

* buffer view tests

* om

* fix torch

* no MOCKGPU

* skip
 2026-03-03 09:52:33 +08:00
chenyu
14d1c5fdfd assign fusion tests on detach and contiguous_backward (#15092) 2026-03-02 15:21:51 -05:00
chenyu
103ea16ec0 add contiguous back to svd (#15074) 
can cause infinite loop
 2026-02-28 16:49:26 -05:00
George Hotz
bb84e389cf functions for llama trainer (#15045) 
* functions for llama trainer

* function there

* axis match

* fix multi

* lil cleaner

* there's a bug with HK_FLASH_ATTENTION

* training functions

* for commit
 2026-02-28 12:15:18 +08:00
chenyu
5fd06f4f02 differentiable setitem (#15054) 
* differentiable setitem

go through the where path for bw

* no return
 2026-02-27 17:25:15 -05:00
chenyu
c9f6d8751b don't remove_bufferize for Invalid (#15053) 
* don't remove_bufferize for Invalid

* replaced
 2026-02-27 15:16:09 -05:00