Commit graph

  • 4b5aa3645b Logging: isamu-isozaki 2024-02-27 19:59:23 -0500
  • bd704bee90 Layernorm dino test isamu-isozaki 2024-02-27 19:55:21 -0500
  • 8fecf12e55 Testing with layernorm isamu-isozaki 2024-02-27 19:53:23 -0500
  • b66aa709d0 Try out layernorm isamu-isozaki 2024-02-27 19:30:39 -0500
  • f27e8002aa Fixed default embed isamu-isozaki 2024-02-27 16:45:10 -0500
  • bebe7b57bc Fix name isamu-isozaki 2024-02-27 14:24:40 -0500
  • 6f55ff99f0 From scratch isamu-isozaki 2024-02-27 14:23:05 -0500
  • 1175a8e9fe Logging isamu-isozaki 2024-02-27 14:21:07 -0500
  • 17717e5e8b Logging isamu-isozaki 2024-02-27 14:19:45 -0500
  • 1fdc496b2e Try random crop isamu-isozaki 2024-02-27 13:41:46 -0500
  • 9f7953b97b Printing lda isamu-isozaki 2024-02-27 13:40:03 -0500
  • 0b8d274004 Logging isamu-isozaki 2024-02-27 13:36:13 -0500
  • 6209f7d223 Precompute fp32 isamu-isozaki 2024-02-27 13:33:28 -0500
  • 78f62d3e9c Remove logs isamu-isozaki 2024-02-27 13:32:24 -0500
  • 8a5f2a1fc7 Logging isamu-isozaki 2024-02-27 13:19:43 -0500
  • 916c1e18a8 Fixed config isamu-isozaki 2024-02-27 13:15:47 -0500
  • 52b8eed036 Fixed config isamu-isozaki 2024-02-27 13:08:00 -0500
  • 7f97546922 Remove random crop isamu-isozaki 2024-02-27 12:56:53 -0500
  • 6277dbceed initial commit Antony 2024-02-27 18:37:15 +0100
  • 5501713a7f Fixed double tokenizer isamu-isozaki 2024-02-27 11:31:23 -0500
  • 9b0d69d487 Fix TextEncoderWithPoolingL isamu-isozaki 2024-02-27 11:19:34 -0500
  • 711f3a714c Checking grad dtype isamu-isozaki 2024-02-27 10:06:15 -0500
  • 872fafacd1 Rescaler sanity check isamu-isozaki 2024-02-27 08:19:18 -0500
  • c8c8c2ee56 Lower lr isamu-isozaki 2024-02-26 22:19:06 -0500
  • 0af79848b4 Use correct text encoder isamu-isozaki 2024-02-26 19:13:01 -0500
  • 9226b8446c Remove pool isamu-isozaki 2024-02-26 19:12:23 -0500
  • 4e102de397 Testing minsnr isamu-isozaki 2024-02-26 19:08:56 -0500
  • be0cfe3e79 Image cross attention should be 0 isamu-isozaki 2024-02-26 18:58:01 -0500
  • 3894f1d0d7 Reporting value of pooled text embedding isamu-isozaki 2024-02-26 18:56:15 -0500
  • 9fc79871a0 Printing context for sanity check isamu-isozaki 2024-02-26 18:51:23 -0500
  • 3b48fe8f4e Printing size isamu-isozaki 2024-02-26 18:49:24 -0500
  • d0dc56512f Fixed loggin isamu-isozaki 2024-02-26 18:47:47 -0500
  • 0ac86ecc72 Printing pooled text linear weight isamu-isozaki 2024-02-26 18:45:56 -0500
  • 062ec1bb4e logging isamu-isozaki 2024-02-26 18:43:34 -0500
  • 198010a042 Testing if pooled text linear is set correctly isamu-isozaki 2024-02-26 18:37:37 -0500
  • 51ac12c26a Sanity check isamu-isozaki 2024-02-26 18:32:18 -0500
  • 86ef7e3b93 No pooled text isamu-isozaki 2024-02-26 18:30:36 -0500
  • ca6b92c50a Increase eval interval isamu-isozaki 2024-02-26 18:26:10 -0500
  • 35f73502a7 Removed unnecessary logic from pooled text emb isamu-isozaki 2024-02-26 18:25:07 -0500
  • dc0e2bd0d8 No palp test isamu-isozaki 2024-02-26 18:13:54 -0500
  • e6da37f797 Test with palp isamu-isozaki 2024-02-26 16:00:47 -0500
  • 529bd0b63a Increase bs test isamu-isozaki 2024-02-26 15:56:36 -0500
  • 4f41322f91 Test diabling palp isamu-isozaki 2024-02-26 15:53:42 -0500
  • ce2070c4b7 Do palp only once in microbatch isamu-isozaki 2024-02-26 15:48:09 -0500
  • f04b3ddbdf Lower batchsize isamu-isozaki 2024-02-26 15:34:16 -0500
  • 0972d3cc6d Fixed to iteration logging isamu-isozaki 2024-02-26 15:31:42 -0500
  • 7fb204594e Fixed uncond isamu-isozaki 2024-02-26 15:15:30 -0500
  • c09bceb586 logging check shapes isamu-isozaki 2024-02-26 15:10:05 -0500
  • 4e5bce1599 Fixed pooled emb issue isamu-isozaki 2024-02-26 14:28:51 -0500
  • daf61f2d2a Expand dim test isamu-isozaki 2024-02-26 14:24:28 -0500
  • 24806b9ae6 Remove logging isamu-isozaki 2024-02-26 14:20:35 -0500
  • d36560cf3e Expand dim test isamu-isozaki 2024-02-26 14:19:25 -0500
  • 74387ac346 Logging isamu-isozaki 2024-02-26 14:15:42 -0500
  • e122e9dab5 Fixed typo isamu-isozaki 2024-02-26 14:11:07 -0500
  • 4702e07c66 Put pooled text embedding in tensor isamu-isozaki 2024-02-26 14:08:30 -0500
  • 5dafc5d88a Logging isamu-isozaki 2024-02-26 13:29:23 -0500
  • 3794b423a1 Fixed to bfloat16 isamu-isozaki 2024-02-26 13:21:33 -0500
  • e55a3d8e1f Fixed pooled text query isamu-isozaki 2024-02-26 13:19:57 -0500
  • a5ff5785a2 Remove diffs isamu-isozaki 2024-02-26 10:35:37 -0500
  • 8a9e034a2a Merge branch 'grad_scaler' of https://github.com/isamu-isozaki/refiners into grad_scaler isamu-isozaki 2024-02-26 10:33:59 -0500
  • 7c7c3ff739
    Merge branch 'main' into grad_scaler Isamu Isozaki 2024-02-26 10:33:35 -0500
  • f9d45c99ff case __getitem__ Pierre Colle 2024-02-26 15:33:23 +0000
  • 07185bec09 Merge branch 'grad_scaler' of https://github.com/isamu-isozaki/refiners into grad_scaler isamu-isozaki 2024-02-26 10:32:03 -0500
  • e206b7a368 Linting isamu-isozaki 2024-02-26 10:27:27 -0500
  • 72e8216ef4 Remove accelerate isamu-isozaki 2024-02-26 10:10:56 -0500
  • 123b4d4dfb More explicit name isamu-isozaki 2024-02-26 10:05:13 -0500
  • 4df3741135 Fix default isamu-isozaki 2024-02-26 10:02:07 -0500
  • 3ed2c29167 Cleaner amp isamu-isozaki 2024-02-26 09:59:26 -0500
  • 03f43523ed Remove comments isamu-isozaki 2024-02-26 09:56:02 -0500
  • 78e3a521c5 Removed grad scaler for non-amp training isamu-isozaki 2024-02-26 09:54:31 -0500
  • acba9cf61d Remove overflow grad check isamu-isozaki 2024-02-23 10:06:54 -0500
  • 5d3cd51fa9 Remove comment isamu-isozaki 2024-02-23 10:04:55 -0500
  • 9e7b65a39c Removed not implemented error isamu-isozaki 2024-02-23 09:54:14 -0500
  • 486de2ab53 Fixed always true condition isamu-isozaki 2024-02-23 09:11:47 -0500
  • dce3cd4d9b Remove diffs isamu-isozaki 2024-02-23 09:09:28 -0500
  • 691aeb5a5b Remove diff isamu-isozaki 2024-02-23 09:08:10 -0500
  • bc20875ab1 Attempt file fix isamu-isozaki 2024-02-23 09:07:26 -0500
  • 1cf64e3296 Style fixes isamu-isozaki 2024-02-23 08:58:24 -0500
  • a82daf4f71 Amp config isamu-isozaki 2024-02-23 08:53:15 -0500
  • d78525d328 Fixed typo isamu-isozaki 2024-02-23 07:58:13 -0500
  • b547ee16cd Testing with no api change isamu-isozaki 2024-02-22 10:35:32 -0500
  • 9876caf424 Remove dtype isamu-isozaki 2024-02-22 10:17:05 -0500
  • 5dc798ad42 Added option for None as dtype for mixed precision training isamu-isozaki 2024-02-21 12:34:19 -0500
  • ce364e1130 Fixed mixed precision training isamu-isozaki 2024-02-21 12:30:47 -0500
  • ae0800988f Disable grad scaler for bf16 isamu-isozaki 2024-02-21 09:43:49 -0500
  • e1f309dae9 Removed diffs isamu-isozaki 2024-02-20 22:00:56 -0500
  • e227f0da42 Cleaned up code isamu-isozaki 2024-02-20 21:57:41 -0500
  • aa6f0a4c9b Fixed styles isamu-isozaki 2024-02-20 21:53:44 -0500
  • b4da3ef496 Fixed accordign to review isamu-isozaki 2024-02-20 21:41:51 -0500
  • c121dfbb1a Fixed double backward Isamu Isozaki 2024-02-16 13:48:44 -0500
  • be99c30bd1 zero grad isamu-isozaki 2024-02-16 14:16:13 -0500
  • 05e77b6f07 Adding grad scaler isamu-isozaki 2024-02-16 12:27:23 -0500
  • 2d64832f5b Linting isamu-isozaki 2024-02-26 10:27:27 -0500
  • 362bd2d7ab Remove accelerate isamu-isozaki 2024-02-26 10:10:56 -0500
  • fdd2153f3f stylistic nits Pierre Colle 2024-02-26 15:05:38 +0000
  • 2959c30bd8 More explicit name isamu-isozaki 2024-02-26 10:05:13 -0500
  • e289d64d5e Fix default isamu-isozaki 2024-02-26 10:02:07 -0500
  • 903fbfefd4 Cleaner amp isamu-isozaki 2024-02-26 09:59:26 -0500
  • ab00953cee Remove comments isamu-isozaki 2024-02-26 09:56:02 -0500
  • 73a469215a Removed grad scaler for non-amp training isamu-isozaki 2024-02-26 09:54:31 -0500