chenyu 90c3ed17c5 move cast to before softmax in attention (#9213) ...

* move cast to before softmax in attention

saved some memory because exp (which is used for backward) are done in half. training bert seems fine and can fit BS=78 now (from 66)

* test

2025-02-24 17:24:59 -05:00

..

external

add test_onnx_ops.py (#8569)

2025-02-24 16:15:22 -05:00

imported

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

mockgpu

OSX GPUOcelot (#8209)

2025-02-13 12:24:29 +08:00

models

allow VIEW(BUFFER) in Tensor UOps [pr] (#9210)

2025-02-24 13:06:15 +01:00

testextra

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

unit

move cast to before softmax in attention (#9213)

2025-02-24 17:24:59 -05:00

web

Move efficientnet example (#8087)

2024-12-06 15:48:16 +01:00

__init__.py

All devices are equal! (#196)

2020-12-15 23:44:08 -08:00

Dockerfile

new cloud is cloudy [pr] (#7631)

2024-11-11 20:18:04 +08:00

helpers.py

add name uop (#9149)

2025-02-18 15:26:58 +08:00

test_arange.py

remove the reshapes in test_arange_2_reduce [pr] (#9063)

2025-02-13 12:33:25 -05:00

test_assign.py

KERNEL op try 3 (#9061)

2025-02-17 14:47:54 +01:00

test_compile_failures.py

enable a few tests that work now (#7721)

2024-11-15 14:30:52 -05:00

test_const_folding.py

rewriter -> devectorizer [pr] (#9147)

2025-02-18 12:42:08 +08:00

test_conv_shapetracker.py

remove inputs+outputs attributes from ScheduleItem [pr] (#9192)

2025-02-21 13:48:11 +01:00

test_conv.py

db in wal mode (#5388)

2024-07-12 20:43:36 -07:00

test_copy_speed.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_device_speed.py

add name uop (#9149)

2025-02-18 15:26:58 +08:00

test_dtype_alu.py

Autogen webgpu dawn, removing wgpu-py dependency (f16 support part 1) (#8646)

2025-02-07 15:16:59 +08:00

test_dtype.py

Solve get_grouped_dims does not split issue (#9085)

2025-02-16 19:57:29 -05:00

test_fusion_op.py

just schedule in test_recursive_pad [pr] (#8860)

2025-02-02 15:01:24 +02:00

test_fuzz_shape_ops.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_gc.py

remove unused ctx [pr] (#8751)

2025-01-26 17:59:15 +09:00

test_graph.py

create_schedule([x.lazydata]) -> x.schedule() in tests (#8449)

2024-12-31 03:15:52 +08:00

test_hcq.py

[TIP-9] rename Opt's amt to arg 2 (#8770)

2025-01-27 14:19:04 -05:00

test_image_dtype.py

remove inputs+outputs attributes from ScheduleItem [pr] (#9192)

2025-02-21 13:48:11 +01:00

test_jit.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_kernel_cache.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_linearizer_dumb.py

tc_select noop (#8801)

2025-01-29 13:53:23 -05:00

test_linearizer_failures.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_linearizer_overflows.py

move time_linearizer to extra.optimization.helpers [pr] (#9048)

2025-02-12 15:49:58 -05:00

test_linearizer.py

remove inputs+outputs attributes from ScheduleItem [pr] (#9192)

2025-02-21 13:48:11 +01:00

test_masked_st.py

multitensor start (#2676)

2023-12-07 17:07:05 -08:00

test_metal.py

Use MTLCompiler directly (v2) (#7920)

2024-12-04 16:36:48 +08:00

test_method_cache.py

simple LoadOps.ASSIGN (#3745)

2024-03-14 20:44:34 -07:00

test_multitensor.py

remove inputs+outputs attributes from ScheduleItem [pr] (#9192)

2025-02-21 13:48:11 +01:00

test_net_speed.py

hotfix: test_net_speed can't backward before realize

2024-12-19 20:32:59 -08:00

test_nn.py

remove SQRT hack in llvm (#9067)

2025-02-13 15:42:34 -05:00

test_ocl.py

Fix GPT2 with OpenCL backend. (#6821)

2024-10-01 16:57:22 +08:00

test_ops.py

fix gradient of pow(t, int) (#9217)

2025-02-23 17:42:09 -05:00

test_optim.py

raise RuntimeError in schedule_step if not Tensor.training [pr] (#8834)

2025-01-31 12:03:04 -05:00

test_pickle.py

simpler buf_uop [pr] (#9215)

2025-02-23 19:23:14 +01:00

test_profiler.py

metal: fix transfer profiling (#8659)

2025-01-17 23:47:01 +03:00

test_quantize_onnx.py

add ast to ProgramSpec + pre matcher [pr] (#9128)

2025-02-17 16:39:14 +08:00

test_randomness.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_rearrange_einops.py

Missing features from rearrange (#6184)

2024-08-19 11:19:07 -07:00

test_renderer_failures.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_sample.py

Autogen webgpu dawn, removing wgpu-py dependency (f16 support part 1) (#8646)

2025-02-07 15:16:59 +08:00

test_schedule.py

reorder expand (#9051)

2025-02-24 13:55:47 +01:00

test_search.py

remove inputs+outputs attributes from ScheduleItem [pr] (#9192)

2025-02-21 13:48:11 +01:00

test_setitem.py

delete forced_realize (#8615)

2025-01-20 09:40:36 -08:00

test_specific_conv.py

Solve get_grouped_dims does not split issue (#9085)

2025-02-16 19:57:29 -05:00

test_speed_v_torch.py

add test_sqrt to test_speed_v_torch (#9066)

2025-02-13 15:25:54 -05:00

test_subbuffer.py

remove contiguous in test_subbuffer_used [pr] (#9046)

2025-02-12 14:41:16 -05:00

test_symbolic_jit.py

UOp.variable (#7010)

2024-10-12 18:20:44 +08:00

test_symbolic_ops.py

few minor code cleanups [pr] (#8267)

2024-12-15 23:44:51 -05:00

test_symbolic_shapetracker.py

update test_merge_view_recursion_err2 [pr] (#8339)

2024-12-19 18:29:34 -05:00

test_tensor_data.py

[BUGFIX] Tensor([]).data() (#7884)

2024-11-24 16:42:57 -05:00

test_tensor_uop.py

remove const_arg (#9002)

2025-02-10 12:45:11 +01:00

test_tensor_variable.py

combine pad2d with pad (#7677)

2024-11-14 17:56:02 +08:00

test_tensor.py

hotfix: skip test_exclude_const_metadata (#9208)

2025-02-22 23:26:04 +02:00

test_tiny.py

preserve size in dtype ptr [pr] (#8898)

2025-02-05 14:38:57 +08:00

test_to_numpy.py

Apply ruff linting rules to tests (#2473)

2023-11-27 21:24:06 -08:00

test_transcendental.py

WebGPU f16 support (f16 bounty part 2) (#8653)

2025-02-12 19:46:53 +08:00

test_uop_graph.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_uops_stats.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_uops.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00

test_winograd.py

create_schedule([x.lazydata]) -> x.schedule() in tests (#8449)

2024-12-31 03:15:52 +08:00

test_zero_copy.py

CLANG -> CPU (#9189)

2025-02-20 18:03:09 -05:00