Code

[SYNC] Sync CentML -> hidet-org#455

Code

Merged

vadiklyutiy merged 98 commits intomainhidet-org/hidet:mainfrom sync-0-4-0hidet-org/hidet:sync-0-4-0Copy head branch name to clipboardJul 23, 2024

+22,145-1,791

Commits

Commits on Jul 22, 2024

• [FFI] Refactor CompiledFunction interface with ctypes (#79)

  Show description for 5ce601d
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  5ce601d

  Copy full SHA for 5ce601d

• [App] Complete UNet Definition (#99)

  Show description for 396ad09
  

  KTong821

  and

  vadiklyutiy

  committed
  396ad09

  Copy full SHA for 396ad09

• [Graph] Stable Diffusion Rope Module (#95)

  Show description for 08eef12
  

  KTong821

  authored and

  vadiklyutiy

  committed
  08eef12

  Copy full SHA for 08eef12

• [CI] Add clang-format script/action (#120)

  Show description for 860fc15
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  860fc15

  Copy full SHA for 860fc15

• [Graph] Add major UNet building components (#97)

  Show description for 68a3fe4
  

  KTong821

  authored and

  vadiklyutiy

  committed
  68a3fe4

  Copy full SHA for 68a3fe4

• [Models] Support for tokenizers in C++ runtime (#69)

  Show description for 9033c10
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  9033c10

  Copy full SHA for 9033c10

• [LLM App] LLM Application initial support (#121)

  

  yaoyaoding

  authored and

  vadiklyutiy

  committed
  0067ce6

  Copy full SHA for 0067ce6

• Stable Diffusion App Infra (#103)

  Show description for b63caf9
  

  KTong821

  authored and

  vadiklyutiy

  committed
  b63caf9

  Copy full SHA for b63caf9

• [Graph] Enhance forward debug instrument (#130)

  Show description for e57a9bb
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  e57a9bb

  Copy full SHA for e57a9bb

• [COMPTIME] Parallelize apply_prologue_epilog(fusion) and IR generation(implement*) (#127)

  Show description for 065a82d
  vadiklyutiy

  committed
  065a82d

  Copy full SHA for 065a82d

• [Ir][Primitives] fix __shfl_xor_sync (#155)

  Show description for a3b7dab
  

  2 peopleauthored and

  vadiklyutiy

  committed
  a3b7dab

  Copy full SHA for a3b7dab

• [Operator] Register missing math primitives (#134)

  Show description for 7d43780
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  7d43780

  Copy full SHA for 7d43780

• [Operator] Fix symbolic broadcasting (#131)

  Show description for bbf0530
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  bbf0530

  Copy full SHA for bbf0530

• [COMPTIME] Specialize Constant._binary() for compilation speedup (#148)

  Show description for aef8220
  vadiklyutiy

  committed
  aef8220

  Copy full SHA for aef8220

• [App] Minor bugfixes for LLM app (#157)

  Show description for fd63e52
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  fd63e52

  Copy full SHA for fd63e52

• [IR] [Primitives] Add thread cluster on sm_90 (#145)

  Show description for 527653a
  

  KTong821

  authored and

  vadiklyutiy

  committed
  527653a

  Copy full SHA for 527653a

• Gemma+torch.compile fixes(autocast, rtruediv) (#159)

  Show description for 4f80460
  vadiklyutiy

  committed
  4f80460

  Copy full SHA for 4f80460

• [Operator] triu + tril operators (#146)

  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  096bfcb

  Copy full SHA for 096bfcb

• [App] Fix LLM app tracing (#158)

  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  e4a0386

  Copy full SHA for e4a0386

• [Fixbug] Set _is_exiting correctly (#163)

  Show description for 43ec055
  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  43ec055

  Copy full SHA for 43ec055

• [App] Cleanup SD Implementation (#143)

  Show description for 76bf2f6
  

  KTong821

  authored and

  vadiklyutiy

  committed
  76bf2f6

  Copy full SHA for 76bf2f6

• Support Transpose2D (#77)

  Show description for 39cc879
  

  3 peopleauthored and

  vadiklyutiy

  committed
  39cc879

  Copy full SHA for 39cc879

• [Models] Gemma implementation (#132)

  

  jacklee1792

  authored and

  vadiklyutiy

  committed
  982b552

  Copy full SHA for 982b552

• Revive dynamic shape support with torch.compile (#162)

  Show description for b75e5d8
  vadiklyutiy

  committed
  b75e5d8

  Copy full SHA for b75e5d8

• [App] ResNet Compiled App (2/2) - Pipeline (#165)

  Show description for 742a6b6
  

  KTong821

  authored and

  vadiklyutiy

  committed
  742a6b6

  Copy full SHA for 742a6b6

• [OPS] Add torch.Tensor.sin, torch.Tensor.cos and torch._C._nn.pad (#175)

  Show description for cafaeed
  vadiklyutiy

  committed
  cafaeed

  Copy full SHA for cafaeed

• [Ir][Primitives] add hopper instructions (#83)

  Show description for 3131abe
  

  4 peopleauthored and

  vadiklyutiy

  committed
  3131abe

  Copy full SHA for 3131abe

• [App] SyncLLM + AsyncLLM interface (#166)

  Show description for b938f74
  

  2 peopleauthored and

  vadiklyutiy

  committed
  b938f74

  Copy full SHA for b938f74

• optimize grouping method (#174)

  Show description for bee8c2a
  

  2 peopleauthored and

  vadiklyutiy

  committed
  bee8c2a

  Copy full SHA for bee8c2a

• [COMPTIME] Add chunksize arg to pool.imap (#178)

  Show description for f1e5162
  vadiklyutiy

  committed
  f1e5162

  Copy full SHA for f1e5162

• [COMPTIME] Reduce the number of fork in multithreading.Pool (#180)

  Show description for dbc130a
  vadiklyutiy

  committed
  dbc130a

  Copy full SHA for dbc130a

• [Bug] Fix number of groups under certain case (#181)

  Show description for 459cab4
  

  maxyanghu

  authored and

  vadiklyutiy

  committed
  459cab4

  Copy full SHA for 459cab4

• [Operator] Adding meshgrid operator support (#183)

  Show description for 067b155
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  067b155

  Copy full SHA for 067b155

• [Fix] Remove YOLOv7 from tests/benchmarks/run_configs.json (#187)

  Show description for c7d827a
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  c7d827a

  Copy full SHA for c7d827a

• __getitem__ with N dimensional index tensor (#185)

  Show description for 70c15b2
  

  2 peopleauthored and

  vadiklyutiy

  committed
  70c15b2

  Copy full SHA for 70c15b2

• feat: parallel job execution for tests (#147)

  Show description for 8c17b1c
  

  c-fteixeira

  authored and

  vadiklyutiy

  committed
  8c17b1c

  Copy full SHA for 8c17b1c

• [OPTIONS] Don't create hidet config if it's not exist (#203)

  Show description for ebd3d85
  vadiklyutiy

  committed
  ebd3d85

  Copy full SHA for ebd3d85

• [BUG] Clear _job_queue in parallel_imap for tests (#204)

  Show description for ec93d68
  vadiklyutiy

  committed
  ec93d68

  Copy full SHA for ec93d68

• [Ir] add utilities for CuTe (#107)

  Show description for 7f11d07
  

  2 peopleauthored and

  vadiklyutiy

  committed
  7f11d07

  Copy full SHA for 7f11d07

• [BUGFIX] Init cuda info before run forks for IR generation (#208)

  Show description for 6adc3d1
  vadiklyutiy

  committed
  6adc3d1

  Copy full SHA for 6adc3d1

• [BENCHs] Refactor transformers tests. Add llama2, mistral, gemma, gpt2 to script (#210)

  Show description for 0881032
  vadiklyutiy

  committed
  0881032

  Copy full SHA for 0881032

• Fix issues related to mistral model (#213)

  Show description for f8bb7dc
  

  2 peopleauthored and

  vadiklyutiy

  committed
  f8bb7dc

  Copy full SHA for f8bb7dc

• steal_weight option fixes && fixes for mistral model (#209)

  Show description for 156268b
  

  2 peopleauthored and

  vadiklyutiy

  committed
  156268b

  Copy full SHA for 156268b

• Xiaocenxiaocen/expose more ldst instructions (#216)

  Show description for eda4e61
  

  2 peopleauthored and

  vadiklyutiy

  committed
  eda4e61

  Copy full SHA for eda4e61

• [Ir][CuTE] lower cute dialect (#109) (#230)

  Show description for 5c7bfe2
  

  2 peopleauthored and

  vadiklyutiy

  committed
  5c7bfe2

  Copy full SHA for 5c7bfe2

• [Operator] Registering torch.Tensor.argmax (#234)

  Show description for 9d14cc1
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  9d14cc1

  Copy full SHA for 9d14cc1

• [Operators] Registering torch.as_tensor (#235)

  Show description for 4add4b9
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  4add4b9

  Copy full SHA for 4add4b9

• Inherit mode argument from torch.compile and set corresponding options (#237)

  Show description for 4d95978
  vadiklyutiy

  committed
  4d95978

  Copy full SHA for 4d95978

• Delete options use_fp16 and use_fp16_reduction (#239)

  Show description for d91446b
  vadiklyutiy

  committed
  d91446b

  Copy full SHA for d91446b

• [Operator] Adding support to operators torch.Tensor.max and torch.Tensor.new_full (#238)

  Show description for 99b12b3
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  99b12b3

  Copy full SHA for 99b12b3

• [Operator] Adding torch.Tensor.expand_as support (#250)

  Show description for 66bf235
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  66bf235

  Copy full SHA for 66bf235

• [Operator] Adding support for torch.Tensor.div (#249)

  Show description for 16eb4a2
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  16eb4a2

  Copy full SHA for 16eb4a2

• [Operator] Registering torch.sigmoid_ (#258)

  Show description for 4be2fd9
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  4be2fd9

  Copy full SHA for 4be2fd9

• [OPTIONS] Use Attention by default (#261)

  Show description for 836e3ac
  vadiklyutiy

  committed
  836e3ac

  Copy full SHA for 836e3ac

• [Operator] Registering torch.Tensor.copy_ (#259)

  Show description for d0877d5
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  d0877d5

  Copy full SHA for d0877d5

• [PERF] Increase accuracy of pick up the best candidate (#269)

  Show description for 8add3b7
  vadiklyutiy

  committed
  8add3b7

  Copy full SHA for 8add3b7

• [Operator] Adding support to repeat_interleave and more (#270)

  Show description for a94c0c6
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  a94c0c6

  Copy full SHA for a94c0c6

• [Operator] Added advanced tensor indexing (#251)

  Show description for 4706425
  

  2 peopleauthored and

  vadiklyutiy

  committed
  4706425

  Copy full SHA for 4706425

• [Fix] Handling getitem special case (#281)

  Show description for 58c351c
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  58c351c

  Copy full SHA for 58c351c

• [SCRIPTS] Adopt our scripts to use mode from torch.compile (#274)

  Show description for 6ddca7f
  vadiklyutiy

  committed
  6ddca7f

  Copy full SHA for 6ddca7f

• fix: handles race condition on parallel config directory creation (#285)

  Show description for 33bb7a3
  

  c-fteixeira

  authored and

  vadiklyutiy

  committed
  33bb7a3

  Copy full SHA for 33bb7a3

• adding support for torch.any (#277)

  Show description for a446ba8
  

  2 peopleauthored and

  vadiklyutiy

  committed
  a446ba8

  Copy full SHA for a446ba8

• [Graph][Ops] fp32 accumulation for matmul_f16 (#268)

  Show description for 39b44ab
  

  3 peoplecommitted
  39b44ab

  Copy full SHA for 39b44ab

• [Operator] torch.any (#287)

  Show description for a87de77
  

  2 peopleauthored and

  vadiklyutiy

  committed
  a87de77

  Copy full SHA for a87de77

• [Fix] Handling Tensor.to(..., device=....) on symbolic tensors (#284)

  Show description for ed0c5e6
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  ed0c5e6

  Copy full SHA for ed0c5e6

• Removing constant tensors that are not needed after subgraph rewrite pass (#252)

  Show description for 6793a09
  

  2 peopleauthored and

  vadiklyutiy

  committed
  6793a09

  Copy full SHA for 6793a09

• [Perf] support vectorized epilogue fusion (#220)

  Show description for 0e76d41
  

  2 peopleauthored and

  vadiklyutiy

  committed
  0e76d41

  Copy full SHA for 0e76d41

• [Graph][Ops] fp32 accumulation for cute matmul (#292)

  Show description for 2e5b827
  

  2 peopleauthored and

  vadiklyutiy

  committed
  2e5b827

  Copy full SHA for 2e5b827

• [BUG] when device is None, device_from_torch returns 'cpu' by default. Fixed (#311)

  Show description for d3aa812
  

  2 peopleauthored and

  vadiklyutiy

  committed
  d3aa812

  Copy full SHA for d3aa812

• Setitem with tensor values. And Boolean type promotion (#290)

  Show description for e1fb8b0
  

  2 peopleauthored and

  vadiklyutiy

  committed
  e1fb8b0

  Copy full SHA for e1fb8b0

• Increase batch size for bert to decrease fluctuations (#236)

  Show description for 3a6b8f1
  vadiklyutiy

  committed
  3a6b8f1

  Copy full SHA for 3a6b8f1

• [PERF] Reduce fixed overhead for model run (#310)

  Show description for bc868f6
  vadiklyutiy

  committed
  bc868f6

  Copy full SHA for bc868f6

• Handle dtype and device in hidet.ones_like op (#316)

  Show description for a6405ac
  

  2 peopleauthored and

  vadiklyutiy

  committed
  a6405ac

  Copy full SHA for a6405ac

• [Operator] Extending the functionality support for einsum (#312)

  Show description for 3f413bf
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  3f413bf

  Copy full SHA for 3f413bf

• [Chore] replace copyrights with citations (#315)

  Show description for 61fe266
  

  3 peopleauthored and

  vadiklyutiy

  committed
  61fe266

  Copy full SHA for 61fe266

• [Fix] Fix the bug in tensor_expand caused by attempting to modify immutable_list (#320)

  Show description for aa2dccc
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  aa2dccc

  Copy full SHA for aa2dccc

• [Operators] Adding PyTorch operators encountered while compiling DALLE2_pytorch (#319)

  Show description for ff1bfc6
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  ff1bfc6

  Copy full SHA for ff1bfc6

• [Fix] Fixing a RuntimeError triggered by tensor_reshape function in register_functions.py (#328)

  Show description for dd134d1
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  dd134d1

  Copy full SHA for dd134d1

• [Fix] Handling hidet errors caused by device difference in getitem (#322)

  Show description for afee434
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  afee434

  Copy full SHA for afee434

• [Fix] Fixing an error triggered by ClampOp (#329)

  Show description for c7f3f61
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  c7f3f61

  Copy full SHA for c7f3f61

• [Operator] Adding __ge__ method for the Tensor class (#330)

  Show description for f839be6
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  f839be6

  Copy full SHA for f839be6

• [OPTIONS] Inherit options from torch.compile() (#260)

  Show description for 44e5162
  vadiklyutiy

  committed
  44e5162

  Copy full SHA for 44e5162

• [Operators] Adding support for torch.nn.TransformerEncoder (#327)

  Show description for 70d1bb2
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  70d1bb2

  Copy full SHA for 70d1bb2

• [Operator] Adding support for torch.Tensor.view_as (#334)

  Show description for 06ab1a0
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  06ab1a0

  Copy full SHA for 06ab1a0

• [OPTIONS] Remove dynamo_config['search_space'] (#342)

  Show description for 4f738b7
  vadiklyutiy

  committed
  4f738b7

  Copy full SHA for 4f738b7

• [OPS] Dissallow in fxgraph not supported functions (#317)

  Show description for 7a1466a
  vadiklyutiy

  committed
  7a1466a

  Copy full SHA for 7a1466a

• [BUG] Fixed search_space bug in bench_op.py (#348)

  Show description for ec3af9d
  vadiklyutiy

  committed
  ec3af9d

  Copy full SHA for ec3af9d

• [Fix] Handling special cases in setitem regarding dtype and device (#332)

  Show description for d2266d7
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  d2266d7

  Copy full SHA for d2266d7

• [Fix] Fixing a bug in register_methods (#331)

  Show description for 7fbaa8d
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  7fbaa8d

  Copy full SHA for 7fbaa8d

• [Fix] Added missing torch.multiply and torch.nn.functional.unfold ops for conv-bert-base model (#351)

  Show description for 97a0c98
  

  2 peopleauthored and

  vadiklyutiy

  committed
  97a0c98

  Copy full SHA for 97a0c98

• [Fix] type casting for attention mask from fp32 -> f16 (#323)

  Show description for 32a2255
  

  2 peopleauthored and

  vadiklyutiy

  committed
  32a2255

  Copy full SHA for 32a2255

• [CI] Promote nvidia docker container to version 24.4 (#354)

  Show description for c6ce9fb
  vadiklyutiy

  committed
  c6ce9fb

  Copy full SHA for c6ce9fb

• [PERF] Introduce add_hint_pass (#355)

  Show description for a5a5373
  vadiklyutiy

  committed
  a5a5373

  Copy full SHA for a5a5373

• [Operators] Registering tensor methods whose PyTorch function equivalents are supported by Hidet (#347)

  Show description for 3ab84b4
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  3ab84b4

  Copy full SHA for 3ab84b4

• [PERF] Remote workaround for loops in add_hints_pass (#356)

  Show description for eb67b6a
  vadiklyutiy

  committed
  eb67b6a

  Copy full SHA for eb67b6a

• [Fix] Fixing an error triggered while compiling the torch.nn.Upsample module with align_corners=True (#344)

  Show description for 7db8629
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  7db8629

  Copy full SHA for 7db8629

• [Operators] Adding leaky_relu support (#360)

  Show description for b9551c4
  

  BolinSNLHM

  authored and

  vadiklyutiy

  committed
  b9551c4

  Copy full SHA for b9551c4

• [CI] Repeat start_instance (#361)

  Show description for f5f528f
  vadiklyutiy

  committed
  f5f528f

  Copy full SHA for f5f528f