Commit Graph

Select branches

Hide Pull Requests

geohot/updates

gh-pages

master

remove_anchors

#1

#2

#3

v0.10.0

v0.10.1

v0.10.2

v0.10.3

v0.11.0

v0.12.0

v0.13.0

v0.4.0

v0.5.0

v0.6.0

v0.7.0

v0.8.0

v0.9.0

v0.9.1

v0.9.2

Mono Color

• d6d902afe9 wtf (#3944) uuuvnandGitHub 2024-03-27 02:49:28 +02:00
• 5530b0cbed fuzz_linearizer: reduce debug verbosity and make easier for CI usage (#3942) Francis LamandGitHub 2024-03-26 13:25:24 -07:00
• 8df6587c41 hotfix 97.3 for beautiful_mnist (#3941) chenyuandGitHub 2024-03-26 15:02:53 -04:00
• b1e3817e18 correctly handle Tensor.rand whwn default_float = bf16 (#3940) chenyuandGitHub 2024-03-26 14:56:16 -04:00
• f6ff76be21 check only upcast int amount in upcasted_axis (#3938) chenyuandGitHub 2024-03-26 12:54:57 -04:00
• e2d6f76723 _alloc and _free with options (#3934) nimlgenandGitHub 2024-03-26 19:11:41 +03:00
• 739f47eb0f check on cuEventSynchronize (#3933) nimlgenandGitHub 2024-03-26 16:14:38 +03:00
• 778d17fbd3 intel matmul (#3830) George HotzandGitHub 2024-03-25 22:37:20 -07:00
• ef537672bf bf16 support in metal (#3929) chenyuandGitHub 2024-03-25 23:17:36 -04:00
• 72d617a37d opencl on OSX does not support fp16 extension (#3931) chenyuandGitHub 2024-03-25 19:50:17 -04:00
• cb6e7b57a6 examples: Fix parameter bandwidth accounting for quantized LLama (#3930) Arseny KapoulkineandGitHub 2024-03-25 15:41:05 -07:00
• 4ecd5789ab #include <tgmath.h> in ops_clang (#3927) chenyuandGitHub 2024-03-25 17:48:57 -04:00
• 514c43201d Fix issues with pointer provenance in load/store through ALU (#3916) Arseny KapoulkineandGitHub 2024-03-25 14:41:05 -07:00
• d651835ef5 verify beautiful_mnist.py eval acc and put into benchmark ci (#3926) chenyuandGitHub 2024-03-25 16:47:49 -04:00
• dc508022a9 clean up clang src header (#3925) chenyuandGitHub 2024-03-25 15:18:35 -04:00
• 2080325e8d output_buffer isn't used anymore (#3919) uuuvnandGitHub 2024-03-25 15:03:56 +02:00
• f2a9ea4ea9 lru allocator for copyin host buffers (#3918) nimlgenandGitHub 2024-03-25 15:57:18 +03:00
• e0e234bf94 hotfix, str compare version for cuda geohot 2024-03-24 20:35:24 -07:00
• 715850aef9 Fix sm89 PTX=1 compilation (#3915) Arseny KapoulkineandGitHub 2024-03-24 20:32:29 -07:00
• 83f39a8ceb env var to change default float (#3902) chenyuandGitHub 2024-03-24 20:33:57 -04:00
• 03899a74bb increase atol on reset train geohot 2024-03-24 15:17:31 -07:00
• d8fafca13a assign regression (#3907) 3 people 2024-03-25 00:12:31 +02:00
• 2d0bfdf01c ptx cleanup (#3893) Szymon OżógandGitHub 2024-03-24 22:54:45 +01:00
• 2e39f57594 move lines around in ops_python wmma (#3911) chenyuandGitHub 2024-03-24 17:14:26 -04:00
• e27129a798 Fix linearizer failure 26 test (#3906) 3 people 2024-03-24 16:34:13 -04:00
• 10673d1447 tiny search cleanup (#3910) chenyuandGitHub 2024-03-24 14:20:55 -04:00
• 9a9cac58f9 add lars to nn (#3750) 3 people 2024-03-24 11:43:12 -04:00
• 8c8b57fd5f cleanup ops python (#3908) chenyuandGitHub 2024-03-24 11:36:31 -04:00
• 2c69888654 include negative float in test_dtype (#3884) chenyuandGitHub 2024-03-24 02:39:15 -04:00
• e22d78b3d2 training cifar with BF16 on CUDA (#3905) chenyuandGitHub 2024-03-24 01:37:47 -04:00
• 0145366323 wmma: fix the AMD TC threads to split the first 16 threads (#3904) Francis LamandGitHub 2024-03-23 18:17:42 -07:00
• 7c3632fd1e add --minimal flag to nvrtc (#3899) sekstiniandGitHub 2024-03-24 00:38:31 +01:00
• a2b2597fc2 replace dtype.name str with render_dtype (#3903) chenyuandGitHub 2024-03-23 19:25:48 -04:00
• 24d004a89b hotfix check ckpts before writing achieved model (#3901) chenyuandGitHub 2024-03-23 17:16:38 -04:00
• 4d566f12b1 touchup einsum (#3900) chenyuandGitHub 2024-03-23 16:46:39 -04:00
• 556dcfb8f2 Fix the result permutation in einsum (#3895) 3 people 2024-03-23 15:48:19 -04:00
• 4e18dd78d3 faster program start in llvm (#3897) nimlgenandGitHub 2024-03-23 15:20:15 +03:00
• 46a3501cec nv ioctl sniffer (#3892) George HotzandGitHub 2024-03-23 00:29:30 -07:00
• 18e0cef14d cheap less lines in ptx (#3890) chenyuandGitHub 2024-03-23 01:12:31 -04:00
• f0c4e06ffd fix cuda sync (#3888) George HotzandGitHub 2024-03-22 19:02:30 -07:00
• 2d3ce53348 touchup test_dtype.test_gradient_dtype (#3887) chenyuandGitHub 2024-03-22 20:56:45 -04:00
• fc11808a79 initialize Tensor grad same type as self (#3613) 3 people 2024-03-22 17:33:18 -07:00
• 8db7a6bbcc debug: add optional detailed BEAM_LOG logging (#3883) Francis LamandGitHub 2024-03-22 16:23:31 -07:00
• f7f67e0cc5 simple fix llama shard with quantize (#3882) chenyuandGitHub 2024-03-22 18:15:37 -04:00
• ee502c8055 fixup to_movement_ops and add back to CI (#3881) chenyuandGitHub 2024-03-22 18:14:49 -04:00
• 16e31f7f0d init multidevice cuda graph (#3858) 3 people 2024-03-22 23:49:48 +03:00
• 0c197b9cf3 hotfix: hip bfloat formatting geohot 2024-03-22 11:52:05 -07:00
• 54dc48aa47 fix assign (#3878) George HotzandGitHub 2024-03-22 11:48:48 -07:00
• 5587594a00 fuzz_linearizer: add --ast and --file params to read kernels (#3877) Francis LamandGitHub 2024-03-22 11:27:40 -07:00
• c5467e5bd6 diverse test value in test_dtype DATA based on dtype (#3864) chenyuandGitHub 2024-03-22 14:22:06 -04:00
• 86ee36e697 preschedule all (#3875) George HotzandGitHub 2024-03-22 11:20:06 -07:00
• d8c3f1894a Use UOpGraph in test (#3876) Szymon OżógandGitHub 2024-03-22 19:12:38 +01:00
• 1c51d586ea replace raise Exception with specific errors (#3874) chenyuandGitHub 2024-03-22 12:32:21 -04:00
• 8ef5490ec8 cuda tranfer + async copyin (#3873) nimlgenandGitHub 2024-03-22 19:01:37 +03:00
• 624bc89910 PTX - implement float 4, ptr arithmetics and other speed improvements (#3775) 3 people 2024-03-22 16:54:02 +01:00
• f4055439dc don't include hip common (#3851) George HotzandGitHub 2024-03-22 08:50:50 -07:00
• 4a27ce6ec9 tiny version of amd_hip_bfloat16 (#3868) qazalandGitHub 2024-03-22 17:37:30 +02:00
• 82ce60e172 use JIT_BATCH_SIZE=4 for GPT2 3090 benchmark (#3870) chenyuandGitHub 2024-03-22 00:40:06 -04:00
• fe6ceff15f proposal: multioutput JIT spec (#3856) qazalandGitHub 2024-03-22 06:28:30 +02:00
• a26090d404 search: change to use "spawn" and limit the number of tasks per child (#3862) Francis LamandGitHub 2024-03-21 21:23:36 -07:00
• dca69df197 hot fix use DEBUG >= 3 for allreduce message (#3869) chenyuandGitHub 2024-03-21 23:40:44 -04:00
• 6729f20aab Ring allreduce try 2 (#3852) 5 people 2024-03-22 01:17:51 +02:00
• 3c0478bfab fuzz_linearizer: add additional DEBUG info for comparison errors (#3866) Francis LamandGitHub 2024-03-21 15:58:10 -07:00
• bc482729d0 lower hlb_cifar acc to 93.3 (#3865) chenyuandGitHub 2024-03-21 17:58:53 -04:00
• e50b7abe4f diversed buf inputs based on dtype in fuzz_linearizer (#3863) chenyuandGitHub 2024-03-21 16:23:11 -04:00
• c40f78499f reuse fuzz_linearizer.compare_linearizer in test_linearizer_failures (#3861) chenyuandGitHub 2024-03-21 14:23:37 -04:00
• 30fa03243e reuse fuzz_linearizer.compare_linearizer in test_linearizer_failures (#3861) chenyuandGitHub 2024-03-21 14:12:27 -04:00
• 33dd99acf4 remove helper_add_store from test_linearizer_failures (#3860) chenyuandGitHub 2024-03-21 12:53:31 -04:00
• 6bf0b82267 alloc new output in fuzz_linearizer between baseline and real one (#3859) chenyuandGitHub 2024-03-21 11:36:05 -04:00
• b78352b423 do not create structs every call in CUDAProgram (#3855) nimlgenandGitHub 2024-03-21 17:51:40 +03:00
• e5745c1a0d fix nan on multigpus cuda (#3854) nimlgenandGitHub 2024-03-21 15:21:55 +03:00
• 4e0819e40b fixing the benchmark not printing in handcode resnet50 opt example (#3850) Anurag LamsalandGitHub 2024-03-21 10:40:31 +05:45
• 85691c8e20 fix hsa sync issue (#3847) nimlgenandGitHub 2024-03-21 04:00:30 +03:00
• f271cd682b user _resolve_dim in argmax (#3846) chenyuandGitHub 2024-03-20 20:17:30 -04:00
• 5c4cf62d2c fix View.pad arg type (#3845) chenyuandGitHub 2024-03-20 19:36:02 -04:00
• 6d5dec2fef log optimized kernels and a script to compare with non-optimized ones (#3829) Francis LamandGitHub 2024-03-20 16:22:08 -07:00
• 9d1d08fbb0 show llama bandwith with timing (#3844) chenyuandGitHub 2024-03-20 17:19:15 -04:00
• 7ff47e45a1 cifar TARGET_EVAL_ACC_PCT=93.5 (#3843) chenyuandGitHub 2024-03-20 16:56:51 -04:00
• 92c5067439 conceptual small refactor (#3842) qazalandGitHub 2024-03-20 22:46:14 +02:00
• 519336cfea factor out partial in SumNode div int (#3841) chenyuandGitHub 2024-03-20 16:34:33 -04:00
• 8cb5215885 Revert "Ring allreduce in multitensor (#3000)" (#3840) George HotzandGitHub 2024-03-20 11:41:49 -07:00
• c5bf9e4c96 Ring allreduce in multitensor (#3000) 5 people 2024-03-20 20:20:01 +02:00
• 455f7bea9b test example from half resnet that idx has number outside of int32 (#3838) chenyuandGitHub 2024-03-20 13:44:20 -04:00
• 727de5ba1e llama 7B on 3090 benchmark (#3837) chenyuandGitHub 2024-03-20 12:48:22 -04:00
• 9452994201 add a better error message for resnet training (#3836) qazalandGitHub 2024-03-20 18:22:15 +02:00
• 47b9cc2dfe use float32 for rand buffer in test_beam_search and test in metal (#3831) chenyuandGitHub 2024-03-19 23:22:58 -04:00
• d17900bc45 use int32 instead of default_int in simplify_phi_loops (#3828) chenyuandGitHub 2024-03-19 17:49:58 -04:00
• 2d54e4d747 clean up hsa driver (#3818) nimlgenandGitHub 2024-03-20 00:17:41 +03:00
• 99cbc24390 use dtypes.int32 as return dtype for functions that return indices (#3827) chenyuandGitHub 2024-03-19 17:06:57 -04:00
• fa1921ec7d move test_dtype tests to test dtype and output value (#3826) chenyuandGitHub 2024-03-19 16:31:27 -04:00
• 131bbb6563 test_linearizer_failure: add failure 27 from a gpt2 kernel (#3825) Francis LamandGitHub 2024-03-19 13:29:50 -07:00
• 3fb13ff892 HIP -> HSA in docs/env_vars (#3824) nimlgenandGitHub 2024-03-19 22:53:33 +03:00
• 9851e2c3b9 test_linearizer_failure: add failure 26 from a gpt2 kernel (#3821) Francis LamandGitHub 2024-03-19 10:19:54 -07:00
• b436c9792f Fix factoring bug (O(n) arange related) (#3817) 3 people 2024-03-19 08:49:42 -07:00
• e12bc85014 use BS=128 and BS=768 for resent benchmark (#3815) chenyuandGitHub 2024-03-18 23:49:55 -04:00
• e6d55932ca hotfix: this makes beautiful mnist work again, not okay geohot 2024-03-18 18:22:44 -07:00
• a6ed2ae3c6 use old cumsum optimization for arange (#3813) chenyuandGitHub 2024-03-18 20:01:03 -04:00
• ac866eaf5a disable simplify_phi_loops (#3812) chenyuandGitHub 2024-03-18 19:25:26 -04:00
• 4c4d3cb3e3 restrict assignment to base (#3809) George HotzandGitHub 2024-03-18 15:33:06 -07:00
• 20681d5c4a remove old dist multigpu (#3811) chenyuandGitHub 2024-03-18 18:31:05 -04:00