nimlgen/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity
14,335 Commits 334 Branches 15 Tags
rewrite_rangeify2
Commit Graph
2025 Commits
Author SHA1 Message Date
nimlgenandGitHub 6b35220622 cpu hcq2 (#17503) 
* cpu hcq2

* temp

* slop

* test with backpressure

* x

* x

* x

* x

* x

* x

* Dx

* save reverts

* um?

* x

* x

* call from py

* x?

* x

* submitters gone

* x

* x

* z

* Dx

* Dx

* x

* x

* fixes

* repl

* x

* f

* for now keep hcqbuffer
 2026-08-14 15:06:03 +03:00
George HotzandGitHub b1859805b1 remove Ops.BIND (#17511) 
* remove Ops.BIND

* param arg

* simplify that

* simplify

* cleaner

* props, not functions

* param and buffer can share
 2026-08-13 23:52:16 -07:00
geohot 95ca5081fe hotfix: update extra/runbook_digitalocean_mi350x 2026-08-13 17:26:39 -07:00
wozeparrotandGitHub 1b7f040984 fa: paas through window (#17517) 2026-08-13 14:45:57 +08:00
qazalandGitHub 39d144546e fix mxfp4 mem estimate (#17515) 
* add mem estimates

* rename

* move
 2026-08-13 14:48:15 +09:00
qazalandGitHub 04c271ac41 simplify digitalocean_mi350x (#17502) 
* simplify digitalocean_mi350x

* no hardcoded rocm path
 2026-08-12 16:00:04 +09:00
George HotzandGitHub e11df72e0f notes from digitalocean_mi350x (#17494) 
* notes from digitalocean_mi350x

* cleanup

* revert non-doc changes on digitalocean_mi350x branch
 2026-08-11 13:22:11 -07:00
nimlgenandGitHub 55e4f9d4f3 hcq2: proper unmap (#17489) 2026-08-11 15:41:28 +03:00
nimlgenandGitHub e29606f07e hcq2: copy kernel (#17480) 
* hcq2: copy with kernel

* test

* x
 2026-08-10 17:28:46 +03:00
qazalandGitHub 44f1f45cd5 llama: custom silu kernels (#17462) 
* start by copying the C

* uop kernel

* cleanup tests

* estimates is part of SPEC
 2026-08-10 16:43:01 +09:00
nimlgenandGitHub 8c8b43de62 hcq2: fix beam (#17467) 
* fix beam

* x
 2026-08-09 16:53:47 +03:00
nimlgenandGitHub e17c21e102 hcq2: timings (#17464) 
* hcq2: timings

* Dx

* x

* x

* x

* x

* align

* x
 2026-08-08 22:00:32 +03:00
wozeparrotandGitHub 1827ec57f7 gptoss: fix sharded invalids (#17450) 2026-08-07 08:42:02 -07:00
wozeparrotandGitHub 1fd6b1035f fa: swa support (#17367) 2026-08-06 08:07:30 -07:00
qazalandGitHub 9636dd1a25 test MXFP4 llama without hipcc (#17435) 
* test MXFP4 llama without hipcc

* first pythonpath then dev
 2026-08-06 17:31:40 +09:00
qazalandGitHub f258708d7d llama: custom quantize_mxfp4+transpose kernel (codex) (#17434) 
* llama: custom quantize_mxfp4+transpose kernel (codex)

* rename to cpp

* inline

* cleanup

* lds load_bf16x4

* more tests, add Estimates
 2026-08-06 16:13:28 +09:00
geohot a8a8030bc9 add benchmark_llm script 2026-08-05 15:59:51 -07:00
nimlgenandGitHub 874d33128b hcq2 benchmark (#17235) 
* hcq2 in ci?

* fix

* traning

* x

* x

* x

* recover

* debug

* impler

* x

* x

* x

* hcq2: group input scatter plans by destination

* hcq2: simplify input scatter tables

* x
 2026-08-05 10:00:42 +03:00
wozeparrotandGitHub 80d2073a11 fa: fix dq hazard with D=64 (#17393) 2026-08-04 09:38:01 -07:00
George HotzandGitHub c21a552f3d llm: bugfixes + warmup (#17384) 2026-08-03 18:23:14 -07:00
wozeparrotandGitHub 3331944547 gptoss: fix moe routing (#17377) 2026-08-03 11:16:44 -07:00
nimlgenandGitHub 7c1ce50f63 hcq2: epoch (#17376) 
* hcq2: epoch

* x

* minor
 2026-08-03 16:37:25 +03:00
chenyuandGitHub b502fc1367 more const arg -> val (#17350) 2026-08-01 02:18:10 -04:00
nimlgenandGitHub 1095bbe409 hcq2: fix ib reuse (#17330) 
* hcq2: initialize IB reuse counters at link

* x
 2026-07-31 18:43:23 +03:00
chenyuandGitHub 8e2f175542 const(dtype, b) -> const(b, dtype) [PR] (#17328) 
prep for dtype removal
 2026-07-31 09:46:37 -04:00
qazalandGitHub f7964acb64 llama with MXFP4 (#17321) 
* mxfp4 in llama

* less

* name
 2026-07-31 18:25:25 +09:00
qazalandGitHub 0a3325f9c2 add mxfp4 quantize and layout kernels (#17320) 2026-07-31 14:50:18 +09:00
qazalandGitHub a8c1e89500 fp4 asm gemm 6+ pflops (#17315) 
* fp4 gemm

* better kernargs structure

* move to .s files

* work

* work

* p2

* style

* use .py

* move to dsl

* cleanup

* add MFMA_SCALE_X2_ENCODING

* cleanup mfma

* fma docs

* gemm_mxfp4

* more cleanup

* move

* move to cdna_asm_gemm

* change

* rm

* change

* mx
 2026-07-31 14:25:23 +09:00
George HotzandGitHub d65ea465ed cleanup gemm fragment + add store unshard (#17313) 
* cleanup gemm fragment + add store unshard

* multi

* fix
 2026-07-30 20:53:28 -07:00
wozeparrotandGitHub b5a2a5666a gptoss moe routing (#17284) 2026-07-30 07:49:14 -07:00
sirhcmandGitHub 060f447db6 qcom: match cl for SP_CS_INSTR_SIZE (#17289) 2026-07-29 18:47:02 -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
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
George HotzandGitHub 52c9e5a99e rename LOOP -> WEAK and STRONGLOOP -> LOOP (#17283) 2026-07-29 10:38:36 -07:00
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
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
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 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
nimlgenandGitHub 97a2265362 hcq2: amd indirect (#17220) 
* ind

* mock
 2026-07-26 21:36:01 +03:00
chenyuandGitHub a60b5f77ac fix torch backend out= into a view (#17210) 2026-07-25 21:04:42 -04:00
nimlgenandGitHub 3946df787d hcq2, cpu is hcq2-ish (#17197) 
* m

* i

* x

* x

* Df

* x

* x
 2026-07-26 02:38:15 +03:00
chenyuandGitHub 076b37e1ae failing batch norm test (#17206) 
* failing batch norm test

running stats does not schedule in training now since there's no reader

* not that
 2026-07-25 18:46:49 -04:00
chenyuandGitHub f902513355 derive torch backend dispatch from the aten schema (#17203) 2026-07-25 14:21:38 -04:00
chenyuandGitHub 983ad3bd95 fix torch backend batchnorm backward (#17201) 2026-07-25 13:15:09 -04:00
chenyuandGitHub 9f78504304 checked cast in torch backend unwrap (#17199) 2026-07-25 12:34:08 -04:00
chenyuandGitHub 8a10892f5a fix torch backend as_strided (#17195) 
0 means 0 offset
 2026-07-25 02:24:17 -04:00
chenyuandGitHub 4c58b260fb less wrong calculate_storage_offset (#17194) 
initially for speed, then realized it's just wrong
 2026-07-25 01:53:54 -04:00
wozeparrotandGitHub c9b60caf8c gptoss: moe gemm kernels (#17178) 2026-07-24 07:25:15 -07:00
nimlgenandGitHub 0ecef210bb hcq2 cleanup 2 (#17177) 
* hcq uops

* x
 2026-07-24 13:04:21 +03:00