sirhcm/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests 2 Actions 3 Activity
11,106 Commits 336 Branches 15 Tags
python_speed
Commit Graph
11106 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
qazalandGitHub 0c59c6b8c7 remove replace from Tensor assign [pr] (#10127) 
* remove replace from Tensor assign

* assign is contiguous

* allow chaining view

* only assert axis
 2025-05-01 19:37:55 +08:00
nimlgenandGitHub 9caceda79a amd: comgr is not required (#10128) 2025-05-01 13:41:44 +03:00
nimlgenandGitHub c3d2e4a6e1 amd: use sdma to copy program (#10126) 
* amd: use sdma to copy program

* rm

* ensure prog is copies

* match nv style
 2025-05-01 13:04:22 +03:00
nimlgenandGitHub 09f5be9bcb amd: finalize device in case of failures (#10124) 2025-05-01 10:41:15 +03:00
George HotzandGitHub ef011ff5f9 flip Ops.COPY order [pr] (#10122) 
* flip Ops.COPY order [pr]

* fix copy and support multi device copy in _device
 2025-05-01 00:26:24 -04:00
chenyuandGitHub 145e51247a split CAST and BITCAST in PYTHON [pr] (#10123) 
CAST only needs truncate and does not require dtype fmt. added bfloat16 tests can run locally
 2025-04-30 23:27:35 -04:00
Ignacio SicaandGitHub bf5fb97498 fix AMD_LLVM bf16 tc for gfx1100 (#10102) 
* fix amd_llvm bf16 tc

* cleanup pattern
 2025-04-30 20:06:38 -03:00
George HotzandGitHub dd0070daab Revert "flip Ops.COPY order [pr] (#10120)" (#10121) 
This reverts commit 984f09ac74.
 2025-04-30 17:25:21 -04:00
George HotzandGitHub 984f09ac74 flip Ops.COPY order [pr] (#10120) 2025-04-30 16:50:18 -04:00
chenyuandGitHub 17d4d258ea simple symbolic slice in llama [pr] (#10112) 
support slice that has step None and stop > start
 2025-04-30 14:36:35 -04:00
nimlgenandGitHub b583ece8f3 amd: replace AMD_DRIVERLESS with AMD_IFACE (#10116) 
* amd: replace AMD_DRIVERLESS with AMD_IFACE

* docs

* print direct err for amd_iface

* print for all
 2025-04-30 20:22:02 +03:00
nimlgenandGitHub 0e1beaf44f nv: align copies + better test (#10118) 2025-04-30 20:09:53 +03:00
Ignacio SicaandGitHub 2941537250 cast is noop if src has dtypes.void (#10110) 2025-04-30 13:55:41 -03:00
nimlgenandGitHub fcdda4fc09 am: move boot memory to vram start (#10115) 2025-04-30 19:12:19 +03:00
nimlgenandGitHub 844d5577d8 hcq: make copy_bufs and kernargs_size params configurable per device (#10114) 2025-04-30 18:43:50 +03:00
nimlgenandGitHub 2ec3b722e2 nv: fix copies larger than 4g (#10117) 2025-04-30 18:43:17 +03:00
George HotzandGitHub d81acbeef6 multi: move shrink after copy (#10109) 
* multi: move shrink after copy

* passing now
 2025-04-30 10:29:51 -04:00
qazalandGitHub 67bd8489ad grouper cleanups [pr] (#10113) 2025-04-30 18:54:47 +08:00
nimlgenandGitHub b4c9a3d8f4 hcq: use mmio iface in copies (#10111) 
* hcq: use mmio iface in copies

* linter

* fix_am

* am
 2025-04-30 11:05:13 +03:00
nimlgenandGitHub 5c7d004da5 hcq: refactor int ptrs to hcqbuffers (#10105) 
* hcq: refactor int ptrs to hcqbuffers

* more refactors

* linter

* use in allocator

* test fiz

* fx

* ops

* final?

* simpler

* keep this for now
 2025-04-30 00:12:18 +03:00
chenyuandGitHub 573bbb9746 Revert "remove TransformerBlock contiguous in llama (#10104)" (#10108) 
This reverts commit b8d07dcc54.
 2025-04-29 15:28:38 -04:00
chenyuandGitHub 4a04098389 fix llama3 with nf4 quantize (#10107) 
also int8 outputs is wrong
 2025-04-29 15:14:36 -04:00
George HotzandGitHub 9c1b80499f names for graph rewrites + null device supports exp and friends (#10106) 2025-04-29 14:28:20 -04:00
chenyuandGitHub b8d07dcc54 remove TransformerBlock contiguous in llama (#10104) 2025-04-29 14:15:39 -04:00
Ignacio SicaandGitHub 9d5677c12c fix ptx linearizer bug 2 [pr] (#9967) 
* check for local buffer

* hotfix

* add test_tensor_cores_emulation run for ptx
 2025-04-29 14:30:07 -03:00
qazalandGitHub a59d18da21 hack for VIZ=1 with examples/llama (#10103) 
* hack for VIZ=1 with examples/llama

* move it alongside BEAM=0
 2025-04-29 23:42:17 +08:00
qazalandGitHub 93bf8764f2 do not open devices in lowering (#10101) 
* do not open devices in lowering [pr]

* ctx=opts

* ctx

* fuzz test
 2025-04-29 23:18:16 +08:00
George HotzandGitHub c3ff308abb range has only one src now [pr] (#10100) 
* range has only one op now

* fix z3 checker

* ci fix

* needs shell

* try pip ensure update

* that ensurepip is useless

* upgrade pip before cache

* windows happy?
 2025-04-29 10:31:05 -04:00
geohot 427471550a hotfix: amd tflops to 74 and some external_benchmark_sdxl_softmax stuff 2025-04-29 09:02:27 -04:00
Ignacio SicaandGitHub 58cf8cd493 add support for "shared_mem" for LLVM (#10093) 
* init llvm shared

* add test_tensor_cores_emulation run for llvm
 2025-04-29 08:56:36 -04:00
qazalandGitHub ad7546c931 assert in test_indexing_two_bind instead of silent fail (#10099) 
* assert in test_indexing_two_bind instead of silent fail

* debuggable

* skip test_simple_train
 2025-04-29 20:23:25 +08:00
George HotzGitHubqazal
 		cee220a1ab always expand ssa on wheres (#9697) 
Co-authored-by: qazal <[email protected]>
 2025-04-29 20:08:41 +08:00
qazalandGitHub 3b67f56c02 kernelize some llama realizes (#10098) 2025-04-29 18:39:56 +08:00
qazalandGitHub cbf7347cd6 display viz rewrites with tabbing if they are subrewrites (#10097) 
* display viz rewrites with tabbing if they are subrewrites

* update viz api
 2025-04-29 17:57:21 +08:00
George HotzandGitHub 73c2f6602f test sdxl softmax (#10096) 2025-04-28 21:55:50 -04:00
George HotzandGitHub eaceafecae do fusion locally (#10095) 
* do fusion locally

* oops, that's the right way

* explicit delete closure
 2025-04-28 20:45:37 -04:00
chenyuandGitHub 3eba3d6ee9 don't pass model in convert_from_huggingface and convert_from_gguf (#10094) 
it only needs n_layers
 2025-04-28 20:11:19 -04:00
George HotzandGitHub a2d0684fc1 test_attention_simple_view (#10092) 
* test_attention_simple_view

* correct comment
 2025-04-28 20:01:22 -04:00
Ignacio SicaandGitHub bda116d773 fix use_tensor_cores propagation (#10048) 
* propagate use_tensor_cores

* add use_tensor_core to arg in test and search

* bugfix

* get TC val from ContextVar in search

* revert minor space change

* add tc emulation test to ci and benchmark

* revert

* revert whitespace change

* remove test for ptx

* add comment and remove llvm test run
 2025-04-28 19:30:50 -03:00
George HotzandGitHub d32f5e9f3a improve rendering of shapes in viz + investigate symbolic [pr] (#10091) 2025-04-28 16:44:09 -04:00
Sieds LyklesandGitHub dbb7aee02e Split constant in div with negative x (#10088) 
* add rule

* change test

* lower complexity limit

* remove offset in fold_unrolled_divs

* remove import

* add one more condition
 2025-04-28 16:24:14 -04:00
chenyuandGitHub 610ee79b22 cherry pick mlperf5.0 branch to master (#10089) 2025-04-28 15:36:56 -04:00
chenyuandGitHub 459a223202 simpler Literal annotation in code_for_workitem [pr] (#10087) 2025-04-28 14:59:25 -04:00
nimlgenandGitHub dcd9a633c3 am: load minimum fw (#10083) 
* am: load minimum psp parts

* try thos

* remove me & pfp
 2025-04-28 21:28:05 +03:00
George HotzandGitHub ecff82a698 fixing single kernel softmax: resolve (#10086) 
* fixing single kernel softmax: resolve

* add failing lin test
 2025-04-28 13:46:20 -04:00
geohot 4c242b0483 hotfix: tests all pass on metal local 2025-04-28 12:09:00 -04:00
George HotzGitHubSieds Lykles
 		690dac79b5 don't modify the ranges on reduce rewrite (#10062) 
* bug in div range folding

* simpler

* oh, this is right for indexing, but the div mod folding needs to be fixed

* reenable

* Passing test_complexity_w_unroll2 (#10068)

* Passing

* remove non_folded_divs

* Add check for negative tern in div folding

* Add test

* bump that limit

* fix casted

---------

Co-authored-by: Sieds Lykles <[email protected]>
 2025-04-28 12:01:19 -04:00
quortusandGitHub 5130759605 Make sure clang always inline batched functions (#10037) 2025-04-28 10:48:24 -04:00
George HotzandGitHub c4a50f9d89 fix full shape in kernel.py [pr] (#10085) 
* fix full shape in kernel.py

* fix that heuristic

* full shape in shapetracker is fast

* fix process replay [pr]

* simpler

* this

* i'm just going to ignore that one
 2025-04-28 09:32:58 -04:00
qazalandGitHub ac37510f60 remu: only write v_cmp result if exec is set (#10084) 2025-04-28 20:31:52 +08:00