nimlgen/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity
13,996 Commits 334 Branches 15 Tags
loop_is_rng
Commit Graph
5937 Commits
Author SHA1 Message Date
geohot 048f510b51 cleanups 2026-07-21 18:29:13 -07:00
geohot cacba3f4d5 cleanups 2026-07-21 18:11:20 -07:00
geohot 01c6f396b1 upd 2026-07-21 17:36:05 -07:00
geohot d51003bb61 LOOP is srcless RANGE (kimi) 2026-07-21 17:18:17 -07:00
chenyuandGitHub 92f9c850b4 fix pow(int, float) (#17126) 
* fix pow(int, float)

* onnx
 2026-07-21 18:48:24 -04:00
chenyuandGitHub b1060ca708 don't promote dtype in _pad_constant [pr] (#17125) 2026-07-21 18:13:50 -04:00
chenyuandGitHub f19a2ad771 single where mixin [pr] (#17118) 
* single where mixin [pr]

no shape broadcasting in ufix and _broadcasted anymore

* QCOM vectorized bool is broken
 2026-07-21 17:38:36 -04:00
Armand du Parc LocmariaandGitHub ef37830d13 allow freeing buffers when pickling/unpickling (#16799) 
* allow pickling out of band buffers

* also need to release when loading

* test peak ram

* lint

* sync before yielding next buffer for backends with async copy in

* skip on mock devices

* reason

* or always bytearray, always free?

* Revert "or always bytearray, always free?"

This reverts commit a017bb68742985a5b7431e0b4e973c2997c92b6a.

* one less copy
 2026-07-21 16:11:23 -04:00
nimlgenGitHubgeohot
 		b764599d87 add Ops.LOOP + conditional Ops.END (kimi) (#17117) 
* add Ops.LOOP + conditional Ops.END (kimi)

* c

* x

---------

Co-authored-by: George Hotz <[email protected]>
 2026-07-21 22:54:14 +03:00
chenyuandGitHub 46b82d4755 don't auto cast cond for WHERE (#17115) 
no or_casted all WHEREs with single mixin, matched torch
 2026-07-21 13:00:07 -04:00
chenyuandGitHub 76dade5a11 implicit broadcast gradient based on shape only [pr] (#17114) 
fixed gradient for shape () UOp, enabled unify WHERE mixin
 2026-07-21 12:51:54 -04:00
chenyuandGitHub f64f96ec59 broadcast_axes [PR] (#17112) 
prerequisite to simplify broadcasting logic and make it implicit
 2026-07-21 11:53:12 -04:00
sirhcmandGitHub f3a5337825 correct spelling of coalesce (#17103) 2026-07-20 21:44:04 -04:00
chenyuandGitHub 95f5c85bf3 some realize and corealize for slow tests (#17106) 2026-07-20 21:43:23 -04:00
chenyuandGitHub 13ca9bd8a6 remove dtypes.index again (#17104) 
also reverted some dtype change, the split made things needlessly complicated
 2026-07-20 20:30:04 -04:00
sirhcmandGitHub 980748ccfc add multiple_of to ParamArg (#17101) 2026-07-20 20:11:54 -04:00
George HotzandGitHub f7ce7f330d llm: minor fixes + tests (#17099) 
* llm: minor fixes + tests

* error
 2026-07-20 14:31:20 -07:00
chenyuandGitHub 4b8db13e01 rdna int8 wmma (#17098) 
nice to fix _wmma_name, also more generic tests
 2026-07-20 17:00:50 -04:00
Pol Puigdemont PlanaandGitHub ef77963cfd derivative of logsumexp is independent of max (#17088) 
same as #7009 but for logsumexp and logcumsumexp.
fwd+bwd kernel count 5 -> 3 for both. gradients unchanged
(ties, -inf masks, torch-compared at grad_atol=1e-7).
 2026-07-20 06:52:16 -07:00
qazalandGitHub 1cf8f2f68c llama: inplace amax update (#17064) 
* llama: inplace amax update

* remove amax_out return

* work

* fit

* work

* work

* keep

* diff cleanup
 2026-07-20 15:05:41 +09:00
chenyuandGitHub ac3f56a1a2 more shift tests (#17083) 2026-07-19 16:05:13 -04:00
chenyuandGitHub 89117d8b9e use real shift in l2i decomp [pr] (#17080) 
works for variable shift distace too, also fixed signed arithmetic fill
 2026-07-19 13:15:06 -04:00
chenyuandGitHub 9970a0aad0 fix Tensor << Tensor for x86 (#17082) 
* fix Tensor << Tensor for x86

* torch
 2026-07-19 12:31:05 -04:00
chenyuandGitHub 0146a30125 improve cast to unsign min_max [pr] (#17078) 2026-07-18 21:58:41 -04:00
George HotzandGitHub b53cd35cff llm: make tokenizer fast (kimi) (#17077) 
* llm: make tokenizer fast

* simpler

* re.escape + qcom mypy fix
 2026-07-18 17:31:59 -07:00
nimlgenandGitHub 232529ce88 hcq2: simpler sync (#17069) 
* x

* y

* n
 2026-07-18 16:27:44 +03:00
chenyuandGitHub 47629f4bcf more weak dtype materialization raise (#17071) 2026-07-17 23:15:14 -04:00
chenyuandGitHub f315df29a0 no weak Tensor from and to real buffer (#17067) 
* no weak Tensor from and to real buffer

creation, assign, safe_save

* is_numpy_ndarray to tensor

* one more
 2026-07-17 16:09:10 -04:00
George HotzandGitHub 86a6ad8ed2 llm: split cli.py into serve.py with the HTTP server (#17065) 
* llm: split cli.py into serve.py with the HTTP server

* min edit
 2026-07-17 10:45:37 -07:00
George HotzandGitHub 3ee2baf71d llm: add tool calling support (kimi) (#17061) 
* llm: add tool calling support

* simpler

* cls

* gpt cleanup

* more gpt cleanups

* tests for tools calling
 2026-07-17 10:20:01 -07:00
qazalandGitHub 7dd3422c63 llama: replace two stage amax with atomics (#17063) 
* atomic amax in c kernels

* quantize fp8 UOp kernel

* diff
 2026-07-17 19:27:10 +09:00
sirhcmandGitHub 6f1176ea90 benchmarks: test usbgpu copy speeds on comma (#17060) 2026-07-17 02:02:21 -04:00
George HotzandGitHub 46172bb7c7 llm: add optional jinja template support (kimi) (#17058) 
* add jinja template support (kimi)

* fix tests

* lil

* more crap to fallback
 2026-07-16 19:02:58 -07:00
chenyuandGitHub 88826a6f35 no weak dtype for randn_like either (#17055) 2026-07-16 18:29:12 -04:00
chenyuandGitHub 3bfd62e915 fix 0 size tolist to match numpy (#17054) 2026-07-16 17:42:52 -04:00
chenyuandGitHub 6fa0b2b19e materialize weak dtype casts to default (#17051) 
in clone and _buffer
 2026-07-16 16:12:33 -04:00
George HotzandGitHub 4d8c3d3fc9 add test_hgemm to test_tiny (#17050) 
* add test_hgemm to test_tiny

* dsp skip
 2026-07-16 13:12:10 -07:00
chenyuandGitHub f6a92d0a16 sum_acc_dtype(weak) is weak (#17047) 
also no explicit weak for rand
 2026-07-16 14:32:37 -04:00
George HotzandGitHub 61e104bdfb use UOp.wmma everywhere (#17045) 
* use UOp.wmma everywhere

* fix
 2026-07-16 10:40:48 -07:00
chenyuandGitHub 5a4156c5d1 bitcast and element_size raise for weak dtypes (#17046) 2026-07-16 13:07:45 -04:00
chenyuandGitHub dba8b6b505 allow weak alu operands (#17044) 2026-07-16 09:33:20 -04:00
jk20342GitHubJacob Kitchen
 		810d8732f9 fix n^2 in limit_bufs by memoizing reachable loads [PR] (#17017) 
* fix n^2 in limit_bufs by memoizing reachable loads [pr]

* Update test_schedule.py

---------

Co-authored-by: Jacob Kitchen <[email protected]>
 2026-07-15 23:54:04 -07:00
chenyuandGitHub e8d3047a50 dtype_from_uop cleanup [PR] (#17036) 2026-07-15 21:52:21 -04:00
chenyuandGitHub be075b200a weak dtypes in dtype_from_uop [PR] (#17032) 
* weak dtypes in dtype_from_uop [PR]

* no weak in spec_program

* weak const fold tests
 2026-07-15 16:54:31 -04:00
nimlgenandGitHub d6fddb066f usb: keep only custom (#17029) 
* usb: keep only custom

* mockgpu by gpt

* gpt said sorry

* revert

* reset

* fix

* flash
 2026-07-15 22:40:52 +03:00
chenyuandGitHub c23d8188e1 remove _ensure_float [pr] (#17030) 
do this cast late. allow `SQRT(int)`
 2026-07-15 11:19:20 -04:00
chenyuandGitHub 0d19970edc least_upper_dtype in dtype_from_uop [PR] (#17028) 2026-07-15 09:27:53 -04:00
chenyuandGitHub ebe26420a7 update where Invalid rules [pr] (#17026) 
fixed TestInvalidTensor.test_tensor_index
 2026-07-15 00:00:30 -04:00
chenyuandGitHub 47ddf94f17 remove InvalidType lt and gt (#17023) 
not really used
 2026-07-14 21:59:15 -04:00
sirhcmandGitHub c9baa2ef79 use pattern matcher in contiguous_view_offset [PR] (#17022) 2026-07-14 19:37:36 -04:00