tinygrad

sirhcm/tinygrad

Fork 0

forked from tinygrad/tinygrad

Files

T

History

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

amd

start dtypes.index [PR] (#16949 )

2026-07-09 15:12:23 -04:00

backend

llama: inplace amax update (#17064 )

2026-07-20 15:05:41 +09:00

device

remove copyin/copyout from Buffer (#17020 )

2026-07-14 19:47:22 +03:00

external

benchmarks: test usbgpu copy speeds on comma (#17060 )

2026-07-17 02:02:21 -04:00

mockgpu

llm: make tokenizer fast (kimi) (#17077 )

2026-07-18 17:31:59 -07:00

models

Tensor.train -> TRAINING [PR] (#16705 )

2026-06-22 15:13:22 -04:00

null

derivative of logsumexp is independent of max (#17088 )

2026-07-20 06:52:16 -07:00

opt

remove copyin/copyout from Buffer (#17020 )

2026-07-14 19:47:22 +03:00

speed

remove intel and amx support (#16482 )

2026-06-02 18:53:05 -07:00

testextra

more Tensor.training -> TRAINING [PR] (#16770 )

2026-06-26 20:11:59 -04:00

unit

more weak dtype materialization raise (#17071 )

2026-07-17 23:15:14 -04:00

web

remove stale / untested viz related files (#13785 )

2025-12-21 16:42:48 +08:00

__init__.py

All devices are equal! (#196 )

2020-12-15 23:44:08 -08:00

helpers.py

hcq2: simpler sync (#17069 )

2026-07-18 16:27:44 +03:00

README

move tests to test/backend (#14691 )

2026-02-12 11:09:44 +08:00

test_tiny.py

add test_hgemm to test_tiny (#17050 )

2026-07-16 13:12:10 -07:00

README

Three groups of tests run in CI

backend -- tests that run on each backend
null -- tests that don't require any backend
unit -- tests that only run on a single backend in CI