sirhcm/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests 2 Actions 8 Activity
14,232 Commits 336 Branches 15 Tags
revert_wrong_ck_fix
Commit Graph
14232 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
geohot eec82b0db4 revert wrong custom kernel fix 2026-08-05 18:27:40 -07:00
George HotzandGitHub 470c032a5e fix slice + non contig kernels (#17423) 
* movement: resolve negative int slice bounds against symbolic sizes

negative int bounds in a slice against a symbolic dim were passed through
unresolved, giving wrong views. resolve them against the (possibly
symbolic) size, like slice.indices does for int dims

* schedule: realize custom kernel inputs that don't resolve to a buffer state

rangeify assigns ranges backward from consumers and CALL contributes none,
so the subgraph above a custom kernel input gets no ranges unless something
in it is realized, and reduce conversion crashes with a KeyError. realize
call inputs that don't resolve to a buffer state.

only view-only movement ops preserve the underlying buffer: anything
computed (ALU, REDUCE, ...) must be realized even if one of its sources
resolves to a buffer, since the whole subgraph above the call has no
ranges. unwrapping src[0] unconditionally missed const branches hanging
off non-src[0] children and silently resolved REDUCEs to their source
buffer. includes regression tests for pure const, mixed buffer+const, and
view-over-buffer inputs
 2026-08-05 16:18:17 -07:00
geohot a8a8030bc9 add benchmark_llm script 2026-08-05 15:59:51 -07:00
George HotzandGitHub 581bfdd94f merge track_rewrites and profile_matches into rewrite_group [PR] (#17420) 
* merge track_rewrites and profile_matches into rewrite_group

* bug

* flip ctx polarity
 2026-08-05 14:41:38 -07:00
chenyuandGitHub 07ac911665 few weak and decomp tweaks [PR] (#17419) 2026-08-05 15:55:04 -04:00
chenyuandGitHub c2f1e5ae2a fix weak cast to strong dtype [pr] (#17418) 
weak can mean higher than that strong dtype, so always use that strong dtype is wrong
 2026-08-05 15:32:23 -04:00
George HotzandGitHub 757a727808 move callify into tensor (#17416) 2026-08-05 11:48:56 -07:00
George HotzandGitHub 2cce85a606 chat: display reasoning_content from streamed responses (#17414) 
* chat: display reasoning_content from streamed responses

The server's StreamRouter emits reasoning_content deltas for think blocks,
but the chat UI was only reading delta.content, silently dropping all
reasoning. Now reasoning is shown in gray (#888) and included in the
message history sent back to the server.

* fix
 2026-08-05 10:49:24 -07:00
nimlgenandGitHub 9b27ea8523 hcq2: cleaner (#17413) 
* hcq2: cleaner

* x
 2026-08-05 19:45:04 +03:00
chenyuandGitHub 6cb419b9b7 regression test for bert nan with weak (#17412) 2026-08-05 12:14:26 -04:00
nimlgenandGitHub 5b0b68ec55 remove debug from test (#17410) 2026-08-05 15:47:40 +03:00
qazalandGitHub ad32bd272b viz/cli: faster and more complete rewrites print (#17411) 
* viz/cli: faster and more complete matches print

* kwargs
 2026-08-05 20:02:27 +09:00
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
chenyuandGitHub 3bf9e70b19 Revert "don't cast weak in _broadcasted [pr] (#17408)" (#17409) 
This reverts commit b45058b5ec.
 2026-08-05 02:40:33 -04:00
RaineGitHubGeorge Hotz
 		77e124e455 fix AMD WMMA emulation and test in CI (#17184) 
* fix SPEC=1 test_tensor_cores

* implement i32 WMMA for RDNA3, add regression test to CI

* gfx950 scaled mfma llvmir fix

* detect VOP3PX2 in emu

* start cdna4 scaled mfam emu

* fix gfx12 llvmir signatures

* oops

* fix

* fix src2 const field extraction (gpt)

* scaled mfma fixes cdna4 (kimi)

* fp8 out breaks mfma (glm)

* fix const signature

* another

* fix f string for linter

* lint

* clean

* and a final lint

* .

* fix mypy

* skip slow tests on ci

* reduce unroll tensor shape -> 64x64

---------

Co-authored-by: George Hotz <[email protected]>
 2026-08-04 23:25:32 -07:00
chenyuandGitHub b45058b5ec don't cast weak in _broadcasted [pr] (#17408) 
* don't cast weak in _broadcasted [pr]

* fine now?
 2026-08-05 02:19:31 -04:00
George HotzandGitHub 46f0003776 more KernelCountException (#17407) 2026-08-04 22:55:38 -07:00
sirhcmandGitHub de57be1f26 kill nvidia pids at benchmarks start (#17406) 2026-08-04 23:45:49 -04:00
chenyuandGitHub 9b508dfafc remove invalid special case in cast [PR] (#17405) 2026-08-04 23:08:07 -04:00
George HotzandGitHub e1f42681fa add new schedule tests + format better (#17402) 
* add new schedule tests + format better

* assert_kernel_count
 2026-08-04 18:46:38 -07:00
chenyuandGitHub 3eab809e06 update minimum to not create strong type const [PR] (#17401) 2026-08-04 21:29:24 -04:00
George HotzandGitHub 6122b3c98f use check_schedule in tests where possible (#17400) 2026-08-04 18:17:29 -07:00
chenyuandGitHub f295f9fc99 use weak 0 in convert_pad_to_where_to_keep_behavior_local [pr] (#17398) 2026-08-04 19:44:34 -04:00
chenyuandGitHub d79772f057 fix pow on extreme inputs (#17397) 
* fix pow on extreme inputs

* WEBGPU
 2026-08-04 19:30:43 -04:00
chenyuandGitHub c1a10e0726 fix _min_max for CAST from float to int [pr] (#17396) 
* fix _min_max for CAST from float to int [pr]

* fix
 2026-08-04 18:06:50 -04:00
chenyuandGitHub 7b6d2ddf23 more weak const without cast in const_like [PR] (#17395) 
* more weak const without cast in const_like [PR]

* that?
 2026-08-04 15:10:43 -04:00
George HotzandGitHub f489f4bc36 add test_eye + color INDEX (#17394) 2026-08-04 11:08:08 -07:00
wozeparrotandGitHub 80d2073a11 fa: fix dq hazard with D=64 (#17393) 2026-08-04 09:38:01 -07:00
chenyuandGitHub 85e9440446 keep weak consts weak in symbolic [PR] (#17392) 2026-08-04 12:16:53 -04:00
chenyuandGitHub 0db63e13cf update dtype_from_uop for IMAGE INDEX [PR] (#17391) 2026-08-04 11:50:24 -04:00
chenyuandGitHub 6eedca57ce dtype_from_uop for CUSTOM, CUSTOMI, PYLITERAL [PR] (#17390) 
these do not change dtype
 2026-08-04 11:02:13 -04:00
b1tgandGitHub 0796853845 support symbolic shapes in allreduce (#17364) 2026-08-04 07:53:12 -07:00
geohot 568bfb6a37 Revert "usb: wait for PCIe link after power on (#17380)" 
This reverts commit 104ee90ccf.
 2026-08-04 06:42:26 -07:00
qazalandGitHub f993228b73 llama: accurate mxfp4 mfu (#17388) 
* llama: accurate mxfp4 mfu

* train_llama3 import
 2026-08-04 18:41:57 +09:00
qazalandGitHub c9cd44b751 more custom kernel contig input edge case tests (#17387) 2026-08-04 18:20:50 +09:00
chenyuandGitHub 13fff4f258 const_like cleanups [PR] (#17386) 2026-08-03 23:57:20 -04:00
chenyuandGitHub 0170a3075a use python bitcast in fold_bitcast [PR] (#17385) 2026-08-03 23:09:53 -04:00
George HotzandGitHub c21a552f3d llm: bugfixes + warmup (#17384) 2026-08-03 18:23:14 -07:00
chenyuandGitHub 67dc02d7e7 bitcast in python for _bits_to_rand [PR] (#17383) 
* bitcast in python for _bits_to_rand [PR]

const in mixin would be weak only without width, so not bitcast
 2026-08-03 21:14:58 -04:00
George HotzandGitHub 3cb786f447 llm: update test_llm_server tests (#17382) 2026-08-03 16:06:17 -07:00
geohot 87289a7410 hotfix: revert test_scalar_alu_index, violates spec 2026-08-03 15:25:51 -07:00
George HotzandGitHub c2625c78cb scalar ALU index fix + llm: preserve_thinking (#17381) 
* cstyle: scalar ALU index fix, serve: preserve_thinking, test: fix Handler import

- cstyle.py: return scalar directly when ALU buffer has 1 element
- cli.py: add preserve_thinking param to FallbackTemplate.render
- serve.py: pass preserve_thinking=True when rendering chat completions
- test_llm_server.py: fix import to use Handler from llm.serve

* real fix
 2026-08-03 15:10:21 -07:00
chenyuandGitHub 33755a3465 improve threefry codegen [pr] (#17379) 
decomp uint64 can handle part of it
 2026-08-03 17:24:25 -04:00
YassineYousfiandGitHub 104ee90ccf usb: wait for PCIe link after power on (#17380) 2026-08-03 14:19:05 -07:00
chenyuandGitHub a2385ae21d MAX_LINE_COUNT=26000 (#17378) 
oh well
 2026-08-03 15:37:58 -04:00
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
qazalandGitHub be5f62d269 llama: refactor amax stuff and skip in fp4 (#17375) 2026-08-03 20:05:52 +09:00
nimlgenandGitHub e22935c758 hcq2: inputs table (#17374) 
* revert this

* x

* simpler

* fix
 2026-08-03 13:40:43 +03:00
chenyuandGitHub 314df72b5f deflake test_hcq with MOCKGPU (#17370) 
for MOCKGPU we compare with e2e wall time which would be device agnostic
 2026-08-03 13:22:59 +03:00