Commit Graph

  • c243ba8251
    Update README.md petals-readme-title Alexander Borzunov 2022-08-02 17:11:20 +0400
  • 443b2c0261 add pseudo efficient inference: debug dbaranchuk 2022-08-01 00:50:50 +0300
  • 564f6edb07 set throughput=1.0 to fix zero throughput problem dbaranchuk 2022-07-31 06:17:48 +0300
  • 8b3d66167f set throughput=1 to fix 0 throughput problem dbaranchuk 2022-07-31 06:13:59 +0300
  • 123e57a5fc update initial peer dbaranchuk 2022-07-31 04:29:39 +0300
  • 2aa236a88e fix typos dbaranchuk 2022-07-31 04:28:38 +0300
  • 1ea397a88b black dbaranchuk 2022-07-31 04:00:24 +0300
  • fced9a8f30 fix typo dbaranchuk 2022-07-31 03:54:19 +0300
  • 805e71a35b update demo deploy scripts dbaranchuk 2022-07-31 03:53:07 +0300
  • 66ce6d9669 make it compatible with low_cpu_mem_usage=True dbaranchuk 2022-07-31 03:52:02 +0300
  • afc1de2627 add script for remote benchmarks dbaranchuk 2022-07-31 01:02:43 +0300
  • ffeceebd4b Fix bug with model duplication in RAM dbaranchuk 2022-07-30 02:50:07 +0300
  • 02619d307d black & isort dbaranchuk 2022-07-29 01:04:03 +0300
  • a621549c06 integrate mixed-8bit model dbaranchuk 2022-07-29 01:01:05 +0300
  • 04a2b6f5e3
    Support various backend dtypes & async serialization (#38) Dmitry Baranchuk 2022-07-28 18:33:58 +0300
  • 1bca7f37ef rm occasional comment support-backend-dtypes dbaranchuk 2022-07-28 18:10:09 +0300
  • 7d5348b10b refactoring dbaranchuk 2022-07-28 17:52:46 +0300
  • 04117d5b44 rm redundant imports dbaranchuk 2022-07-28 17:52:13 +0300
  • 5c17ef3d66 rm redundant imports dbaranchuk 2022-07-28 17:48:55 +0300
  • 648a5d6e1d fix autograd backward & black & isort dbaranchuk 2022-07-28 17:46:31 +0300
  • 0e8b7490f7 revert changes dbaranchuk 2022-07-28 05:59:05 +0300
  • e8f84e7747 revert changes dbaranchuk 2022-07-28 05:57:34 +0300
  • 682c6711df revert changes dbaranchuk 2022-07-28 05:56:30 +0300
  • 3151c1d48d Merge branch 'support-backend-dtypes' of github.com:learning-at-home/bloom-demo into support-backend-dtypes dbaranchuk 2022-07-28 05:52:55 +0300
  • bfeddb5463 revert changes dbaranchuk 2022-07-28 05:52:42 +0300
  • c6c064abc4
    Delete ptune_v2_model.py Dmitry Baranchuk 2022-07-28 05:44:06 +0300
  • acdd95ff0b
    Delete shallow_prompt_tuning_sst2.ipynb Dmitry Baranchuk 2022-07-28 05:43:47 +0300
  • 364ecfbb29
    Delete shallow_prompt_tuning_cola.ipynb Dmitry Baranchuk 2022-07-28 05:43:41 +0300
  • 36e3da4e80
    Delete deep_prompt_tuning_sst2.ipynb Dmitry Baranchuk 2022-07-28 05:43:33 +0300
  • 0e6e4ae192
    Delete deep_prompt_tuning_cola.ipynb Dmitry Baranchuk 2022-07-28 05:43:26 +0300
  • 6898f7bfa8
    Delete requirements-notebook.txt Dmitry Baranchuk 2022-07-28 05:43:10 +0300
  • 2d2ad31e5f update the hivemind version that support bfloat16 dbaranchuk 2022-07-28 04:19:17 +0300
  • c23034a468 async serialization & cast to the scheme dtype dbaranchuk 2022-07-28 04:04:38 +0300
  • beacf25c5c cast incoming tensors to backend dtype dbaranchuk 2022-07-28 04:01:54 +0300
  • e8114fbef2 Merge remote-tracking branch 'origin/main' into support-backend-dtypes dbaranchuk 2022-07-28 03:54:34 +0300
  • d989b94614
    Pack of Inference Changes (#37) Artem Chumachenko 2022-07-27 10:19:45 +0400
  • b7e6a4f12e fixes Artem Chumachenko 2022-07-27 05:58:52 +0000
  • 5b06dc2255 renaming deep-prompt-tuning dbaranchuk 2022-07-27 05:44:50 +0300
  • 1477bdc471 add preliminary notebooks dbaranchuk 2022-07-26 19:10:33 +0300
  • c6e26864da add requirements for notebooks dbaranchuk 2022-07-26 19:06:48 +0300
  • cf4acd007e add p-tuning v2 only for _non-distributed_ case dbaranchuk 2022-07-26 19:03:19 +0300
  • 81422fdfc2 add deep p-tuning only for _non-distributed_ case dbaranchuk 2022-07-26 19:02:39 +0300
  • f005d0a287 WIP mockup justheuristic 2022-07-26 18:52:57 +0300
  • 81ea014bf5 WIP justheuristic 2022-07-26 18:47:34 +0300
  • 9123ce57ac Add tests Artem Chumachenko 2022-07-24 09:32:35 +0000
  • f10b05f912 Return multibatch mode Artem Chumachenko 2022-07-24 06:24:40 +0000
  • 6573076883
    Sequential and parallel forward / backward (#36) Dmitry Baranchuk 2022-07-23 14:32:39 +0300
  • d94008311d add todo & black efficient-forward-backward dbaranchuk 2022-07-23 14:00:51 +0300
  • a2d020afcd rename functions & add descriptions dbaranchuk 2022-07-23 13:57:15 +0300
  • d9d73cb0da isort dbaranchuk 2022-07-23 03:05:11 +0300
  • 3dfa95636e address comments & black & isort dbaranchuk 2022-07-23 03:02:46 +0300
  • c925f4d45b black & isort dbaranchuk 2022-07-23 02:50:15 +0300
  • 44e33f9ea6 Merge branch 'efficient-forward-backward' of github.com:learning-at-home/bloom-demo into efficient-forward-backward dbaranchuk 2022-07-23 02:45:53 +0300
  • 180d91bc7a remove redundant imports dbaranchuk 2022-07-23 02:43:51 +0300
  • ba9f3aec16
    Delete async_forward_backward.py Dmitry Baranchuk 2022-07-23 02:40:44 +0300
  • d1abb2f169 address the comments dbaranchuk 2022-07-23 02:39:17 +0300
  • ed86b36911 address the comments dbaranchuk 2022-07-23 02:34:27 +0300
  • b60eedc8ad Merge remote-tracking branch 'origin/main' into efficient-forward-backward dbaranchuk 2022-07-23 02:17:46 +0300
  • f0cffbf67e
    Miscellaneous fixes to automatic tests (#35) justheuristic 2022-07-22 22:38:40 +0300
  • abd3da6f3a
    Update .github/workflows/run-tests.yaml fix-branch-name Dmitry Baranchuk 2022-07-22 22:26:53 +0300
  • 84532ee5f9 revert debug changes justheuristic 2022-07-22 22:21:47 +0300
  • dc277b34da minimize peak CPU memory usage justheuristic 2022-07-22 22:15:57 +0300
  • 3cc4e0b9f0 update server deployment scripts dbaranchuk 2022-07-22 21:53:49 +0300
  • f976191551 efficient forward & backward dbaranchuk 2022-07-22 21:50:13 +0300
  • 996104abeb [WIP] disable conversion justheuristic 2022-07-22 21:49:24 +0300
  • 3f048e5016 undo half justheuristic 2022-07-22 21:45:02 +0300
  • 05fed964a3 remove torch_dtype on converted model justheuristic 2022-07-22 21:42:16 +0300
  • dfe57ebf9a set dtype='auto' justheuristic 2022-07-22 21:30:21 +0300
  • aee295f0db
    de-flappify: give servers more time to download the model justheuristic 2022-07-22 21:02:31 +0300
  • 6583763ed9
    remove debugprint justheuristic 2022-07-22 20:37:18 +0300
  • 375f5adc8f
    [WIP] justheuristic 2022-07-22 20:32:58 +0300
  • 39dd3d990e
    wait for the first server to initialize DHT justheuristic 2022-07-22 20:25:11 +0300
  • c322865dc3
    Fix branch name in CI justheuristic 2022-07-22 20:20:14 +0300
  • 5a4ac2fa71 change tag name justheuristic 2022-07-22 20:17:53 +0300
  • e1454e4448 make tests pass for now justheuristic 2022-07-22 05:19:17 +0300
  • a87be914f3 make tests pass for now justheuristic 2022-07-22 05:14:00 +0300
  • 32dc9649ca routing strategy interface justheuristic 2022-07-22 05:00:23 +0300
  • 6af1572a43 update imports justheuristic 2022-07-22 04:50:17 +0300
  • f31f3eb3af update imports justheuristic 2022-07-22 04:49:57 +0300
  • ab732886e3 update imports justheuristic 2022-07-22 04:49:40 +0300
  • 3f4cbdd1d0 List -> Sequence justheuristic 2022-07-22 04:49:30 +0300
  • ab2fc03f30 decouple tests justheuristic 2022-07-22 04:46:33 +0300
  • c7eda494bf WIP justheuristic 2022-07-22 02:28:13 +0300
  • 7de3acf909
    Fix is_subsequence (#32) Dmitry Baranchuk 2022-07-19 22:37:19 +0300
  • 1f6c02903e
    fix is_subsequence dbaranchuk-patch-1 Dmitry Baranchuk 2022-07-19 22:26:26 +0300
  • f0c7383181
    Implement RemoteSequential slicing and extra repr, add tests (#30) justheuristic 2022-07-19 04:28:04 +0300
  • b02d09dd31 use conftest to clean up child processes get_sequence justheuristic 2022-07-19 03:58:58 +0300
  • 57a3a065a9 run each test in a forked process justheuristic 2022-07-19 03:50:49 +0300
  • 31163a95f8 black-isort justheuristic 2022-07-19 03:40:23 +0300
  • 25a4b23640 unify environment variable names in all tests justheuristic 2022-07-19 03:40:07 +0300
  • b737a60be4 run all ./tests automatically (instead of calling each one manually) justheuristic 2022-07-19 03:36:58 +0300
  • 52238e8e5c black-isort justheuristic 2022-07-19 03:21:40 +0300
  • c597f4d520 make a test for remote sequential with grads justheuristic 2022-07-19 03:21:25 +0300
  • dd4e792937 print minimalistic remote server info Tim Dettmers 2022-07-19 02:54:49 +0300
  • ec20fa6ec5 black/isort justheuristic 2022-07-19 02:49:58 +0300
  • 1b314f64db refactor: extract both inference sessions into a dedicated file (rationale: otherwise, their original files become hard to read) justheuristic 2022-07-19 02:48:47 +0300
  • 93bc211ec6 rm unused imports justheuristic 2022-07-19 02:41:39 +0300
  • d99ca3f383 prepare for merge justheuristic 2022-07-19 02:33:19 +0300
  • d7742cfca7 fix undefined variable justheuristic 2022-07-19 02:32:27 +0300
  • 653a85e6a2 fix decorator; avoid loading model and ref_model at the same time justheuristic 2022-07-19 02:31:32 +0300