nimlgen/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity
14,164 Commits 334 Branches 15 Tags
475ad15f2829dbd8afa53d596af01e981fcdc538
Commit Graph
14164 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
geohot 475ad15f28 27b 935/47 2026-08-02 00:36:02 +00:00
geohot 6d526f0252 /raid/models/Qwen3.6-27B-IQ4_XS.gguf works at 704/40 2026-07-31 21:25:09 +00:00
geohot 8e86cfa60e /raid/models/Qwen3.6-35B-A3B-UD-IQ4_XS.gguf 2026-07-31 14:33:37 +00:00
geohot 3ae41d3f58 move the kernels 2026-07-31 04:38:59 +00:00
geohot 5696ecb5bc fix thinking for qwen 2026-07-31 03:38:07 +00:00
geohot d05770194c kernel cleanup 2026-07-31 02:51:23 +00:00
geohot 87485820f3 delete 2026-07-31 02:29:49 +00:00
geohot d432948533 make llm launch metadata explicit 2026-07-30 18:13:31 +00:00
geohot 55ebf56c69 simplify llm core integration 2026-07-30 16:22:25 +00:00
geohot 5dbd9a3020 cleanups 2026-07-30 15:55:10 +00:00
geohot 6a8bb39f3d no junk ops 2026-07-30 15:02:12 +00:00
geohot 5331889b06 speed for testing 2026-07-30 13:54:01 +00:00
geohot 3af1d62571 tests passing 2026-07-30 03:24:48 +00:00
geohot f0295493a8 Merge remote-tracking branch 'origin/master' into llm_262k_minimal 
# Conflicts:
#	extra/gemm/amd_flash_attention.py
#	tinygrad/codegen/__init__.py
#	tinygrad/runtime/ops_cpu.py
#	tinygrad/uop/ops.py
 2026-07-30 00:51:47 +00:00
geohot 6cd7cc0888 unused 2026-07-29 23:27:06 +00:00
sirhcmandGitHub 060f447db6 qcom: match cl for SP_CS_INSTR_SIZE (#17289) 2026-07-29 18:47:02 -04:00
chenyuandGitHub fd912b348c generic c0*x<c1 [pr] (#17291) 2026-07-29 18:45:23 -04:00
George HotzandGitHub 138676ab81 improve fragment example + index unshard (kimi) (#17288) 
* fix dtypes in fragment example

* match tilelang

* flip locals

* fix index on unshard

* test fixes

* kimi needs more taste
 2026-07-29 15:38:38 -07:00
chenyuandGitHub 027907a544 fix c0*x<c1 symbolic [pr] (#17290) 
use `//` and not math.ceil
 2026-07-29 18:32:31 -04:00
geohot d79daa6acb all kernels in tinygrad uop language and fast 2026-07-29 22:30:59 +00:00
nimlgenandGitHub d4ba8b6e0f hcq2: use stack (#17286) 2026-07-29 22:09:34 +03:00
George HotzandGitHub b30c7e00d4 support 2d on UNSHARD (kimi) (#17285) 
* support 2d on UNSHARD

* fixes

* Fix test and spec

* single barrier

* 2d sharding works for devices too

* cleanups

* no _rewrap
 2026-07-29 12:01:59 -07:00
sirhcmandGitHub aab51fb7b6 64-bit UOp.variable support, try 2 (#17256) 2026-07-29 14:39:36 -04:00
chenyuandGitHub dd86a30798 test case for weakfloat cast to weakint INDEX (#17287) 2026-07-29 14:15:52 -04:00
George HotzandGitHub 52c9e5a99e rename LOOP -> WEAK and STRONGLOOP -> LOOP (#17283) 2026-07-29 10:38:36 -07:00
chenyuandGitHub 6c2b9fac08 weak const cleanups [PR] (#17282) 2026-07-29 13:14:20 -04:00
George HotzandGitHub bd296a7359 enable alloc_fragment support with UNSHARD (kimi) (#17272) 
* enable alloc_fragment support with UNSHARD (kimi)

* cleaner with implicit barrier

* cleanups

* cleaner

* strongloop

* dcount cleanups
 2026-07-29 09:46:45 -07:00
Noah SchiroandGitHub 3df1b07c86 Patch half precision ops for older gpu archs (#17274) 2026-07-29 09:42:24 -07:00
chenyuandGitHub 3803f1583b fix add_raw_barrier [pr] (#17278) 
* fix add_raw_barrier [pr]

a kernel might need multiple barriers

* smaller?
 2026-07-29 09:30:00 -07:00
qazalandGitHub 6ea7d366fa test permuted input in custom Ops.PROGRAM test (#17279) 
* test permuted input in custom Ops.PROGRAM test

* dtype
 2026-07-29 18:29:05 +09:00
George HotzandGitHub 527e57300c fix smu reset for kernel >= 7 (#17277) 
* fix smu reset for kernel >7

* match kernel
 2026-07-28 23:11:21 -07:00
qazalandGitHub e684fcc689 pm_reduce_collapse fix for re enabling stack for cat of same shape (codex) (#17251) 
* change

* stack cat

* remove nop

* change
 2026-07-29 14:47:28 +09:00
George HotzandGitHub 451120c6e1 make .barrier implicit (kimi) (#17275) 
* make .barrier implicit (kimi)

* simplier

* lil

* remove tinygrad stock barriers

* readable

* lil
 2026-07-28 22:34:57 -07:00
chenyuandGitHub 2f8f2d2d37 remove no-op explicit dtype= or cast [PR] (#17276) 2026-07-29 00:47:35 -04:00
George HotzandGitHub dd16d5aead apply shrink bugfix for 3.11 (#17271) 2026-07-28 19:03:06 -07:00
chenyuandGitHub 291ee435bb qwen3.6 for 27b and 35b-a3b (#17268) 2026-07-28 21:11:49 -04:00
George HotzandGitHub a17387d6a0 add UNSHARD to spec (#17269) 2026-07-28 17:00:21 -07:00
George HotzandGitHub 57ae1bc7a7 rename MULTI to UNSHARD (#17267) 
* rename MULTI to UNSHARD

* comment updates (glm)

* rename method to unshard
 2026-07-28 16:51:41 -07:00
George HotzandGitHub 1757067fa6 add device range as src[1] to multi (kimi) (#17264) 
* add device range as src[1] to multi (kimi)

* cleanup

* fix tests
 2026-07-28 16:39:34 -07:00
chenyuandGitHub f11f884f49 update dtypes.md (#17266) 2026-07-28 19:30:13 -04:00
chenyuandGitHub 23e9e76e8c DEFAULT_FLOAT/DEFAULT_INT ContextVar [pr] (#17265) 2026-07-28 19:08:18 -04:00
George HotzGitHubteeny-runner
 		755dfb243b rename CPU_COUNT to NUM_CPU_THREADS with cgroup awareness (#17263) 
Rename CPU_COUNT to NUM_CPU_THREADS so it can be overridden via env var.
Default uses _get_cpu_count() which respects cgroup limits:
  - os.process_cpu_count() on Python 3.13+
  - /sys/fs/cgroup/cpu.max on cgroup v2
  - /sys/fs/cgroup/cpu/cpu.cfs_quota_us on cgroup v1
  - os.sched_getaffinity(0) fallback

Use NUM_CPU_THREADS.value in the dataloader instead of cpu_count(),
and update export_model.py and all renderer references.

Co-authored-by: teeny-runner <runner@teeny>
 2026-07-28 15:33:57 -07:00
George HotzandGitHub 0cdddf3495 remove unneeded default args in renderers (#17262) 2026-07-28 14:53:04 -07:00
chenyuandGitHub fde3a8faff CAPTURE_PROCESS_REPLAY=0 default chaging test [PR] (#17261) 
* CAPTURE_PROCESS_REPLAY=0 default chaging test [PR]

quick patch before making default dtype context vars, which is quite annoying

* few more
 2026-07-28 17:32:35 -04:00
George HotzandGitHub 9ce65b71c0 delete dsp_pm_late (#17260) 2026-07-28 14:05:41 -07:00
George HotzandGitHub 7d48926293 switch _device_num to AxisType.DEVICE range (kimi) (#17252) 
* switch _device_num to AxisType.DEVICE range (kimi)

* improve

* don't exclude device for ranges

* current multi ends the range
 2026-07-28 13:52:47 -07:00
drkandGitHub d1c3ae0f91 preserve typed ranges through reshape (#17259) 2026-07-28 16:51:06 -04:00
chenyuandGitHub 749e002286 remove some dtype= when construct UOp [pr] (#17257) 
prep for removing the dtype field
 2026-07-28 16:05:35 -04:00
chenyuandGitHub c9e11544df delete explicit casts [pr] (#17255) 
handled by broadcast
 2026-07-28 13:57:13 -04:00
George HotzandGitHub a9ad08064e make the github actions runners generic for gitea (#17254) 2026-07-28 10:48:49 -07:00