chenyu c07daf40e7 move attention upcast (#7830) ...

still upcast before softmax, but faster because intermediate buffer can be stored in half (as long as qk is within half range).

2024-11-22 17:10:51 -05:00

..

__init__.py

more test cleanups (#2631)

2023-12-05 16:17:57 -08:00

test_attention.py

move attention upcast (#7830)

2024-11-22 17:10:51 -05:00

test_device.py

Device method cleanup [pr] (#7375)

2024-10-29 12:49:47 -04:00

test_disk_cache.py

helpers.diskcache_clear (#4436)

2024-05-05 14:19:01 -04:00

test_disk_tensor.py

move is_dtype_supported to device [pr] (#7575)

2024-11-07 20:38:03 +08:00

test_elf.py

run unit tests separate from others (and only once) (#6020)

2024-08-10 11:17:56 -07:00

test_gguf.py

move is_dtype_supported to device [pr] (#7575)

2024-11-07 20:38:03 +08:00

test_graph_rewrite.py

Remove UnaryOps, BinaryOps, TernaryOps, MetaOps [pr] (#7725)

2024-11-16 20:56:56 +08:00

test_helpers.py

move polyN to helper [pr] (#7405)

2024-10-30 10:09:57 -04:00

test_pattern_matcher.py

Remove UnaryOps, BinaryOps, TernaryOps, MetaOps [pr] (#7725)

2024-11-16 20:56:56 +08:00

test_qcom.py

BufferSpec and ProgramSpec [pr] (#7814)

2024-11-21 12:18:05 +08:00

test_shapetracker_math.py

enable a few tests that work now (#7721)

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

test_shapetracker.py

rewrite idx again in real_strides after uop_given_valid (#7600)

2024-11-08 14:30:32 -05:00

test_shm_tensor.py

line reduction [pr] (#7296)

2024-10-25 17:05:09 +07:00

test_simplify_valid_idx.py

different valid order (#7589)

2024-11-07 20:27:56 -05:00

test_tqdm.py

Context manager support for tqdm (#7770)

2024-11-18 14:12:03 -05:00

test_transcendental_helpers.py

update payne_hanek_reduction [pr] (#7455)

2024-10-31 18:41:22 -04:00

test_uop_resolve.py

rename MathTraits to maximum [pr] (#7356)

2024-10-29 16:43:04 +08:00

test_uop_symbolic.py

cond.where(True, False) is cond (#7733)

2024-11-16 09:44:17 -05:00

test_uop_vmin_vmax.py

s/UOps/Ops (#7500)

2024-11-03 11:26:10 +08:00

test_verify_ast.py

delete ReduceOps, only use REDUCE_AXIS (#7667)

2024-11-13 19:04:27 +08:00

test_view.py

check contiguous in View.create after canonicalizing mask and offset (#4770)

2024-05-29 11:31:13 -04:00