Commit Graph

  • e2d664ddf0 Update README.md master Matthew Hayes 2023-06-30 11:36:16 -07:00
  • f9dc122f84 Update README.md Matthew Hayes 2023-06-30 11:35:51 -07:00
  • b337d53747 Merge pull request #200 from edurdevic/update-readme-DBR-13 Sean Owen 2023-06-23 20:01:22 -07:00
  • b1e8ca091d Update README.md changed quotes for DBR Erni Durdevic 2023-06-22 16:14:09 +02:00
  • aead199517 Update README.md to 13.x ML or later Erni Durdevic 2023-06-22 16:12:42 +02:00
  • 1b3d9b2371 Update README.md with instructions for DBR 13 Erni Durdevic 2023-06-21 22:39:41 +02:00
  • aaa0ecb5a5 Merge pull request #193 from srowen/DBR13 Sean Owen 2023-06-08 07:46:47 -05:00
  • a357706573 Merge pull request #192 from srowen/bf16flag Sean Owen 2023-06-06 19:12:34 -05:00
  • 10d0d65644 Remove DBR 12 notes in favor of 13; remove test loading data as it doesn't respect flag anyway Sean Owen 2023-06-06 17:55:30 -05:00
  • 6609c85664 Set bf16 flags corretly for a10/a100 Sean Owen 2023-06-06 17:49:24 -05:00
  • fd1a733a77 Merge pull request #189 from holdenk/fix-requirements Sean Owen 2023-06-05 14:02:32 -05:00
  • 364a221943 Drop duplicate torch from requirements_dev.txt, installing with "pip install -r requirements_dev.txt" results in ERROR: Double requirement given: torch==1.13.1 (from -r dolly/requirements_dev.txt (line 4)) (already in torch<2,>=1.13.1 (from -r dolly/requirements.txt (line 7)), name='torch') with pip version pip 20.0.2 Holden Karau 2023-06-02 16:37:29 -07:00
  • a33d774e85 Merge pull request #182 from tnixon/a10_v100_config Sean Owen 2023-06-02 12:14:42 -05:00
  • cd67192d0c use batch_size of 4 for A10 GPUs tnixon 2023-05-31 20:05:25 +00:00
  • 8ca36a1c8b restoring # MAGIC comments in front of ! command cells Tristan Nixon 2023-05-26 14:42:15 -07:00
  • 9ca66873ea updated README instructions for training on alternate GPUs Tristan Nixon 2023-05-26 14:37:22 -07:00
  • ae379b0362 set torch_dtype as string for different fp architectures tnixon 2023-05-26 21:24:27 +00:00
  • bd1a3a9edf Ubuntu 2004 libs for DBR 12.x (2204 for 13.x only) tnixon 2023-05-26 20:19:29 +00:00
  • 5b1fa64ae4 The bf16 setting throws errors when enabled = True, but fine with "auto" tnixon 2023-05-26 20:18:47 +00:00
  • 31c3761f15 Merge branch 'databrickslabs:master' into a10_v100_config Tristan Nixon 2023-05-26 09:55:30 -07:00
  • 5021d941d9 [Fix] Move attention mask to the model device type (#180) Baiqing Lyu 2023-05-26 09:40:58 -07:00
  • 29377312a3 include option for A10 gpus in gpu_family widget Tristan Nixon 2023-05-26 09:17:28 -07:00
  • 221dee88fd updating ubuntu release version Tristan Nixon 2023-05-26 09:15:55 -07:00
  • 42d31a5cf4 Updates for A10 configuration Tristan Nixon 2023-05-26 09:04:45 -07:00
  • c29d650d12 Following up on PR review comments Tristan Nixon 2023-05-25 20:57:34 -07:00
  • 75db1eb09e working on several PR review comments tnixon 2023-05-26 03:30:15 +00:00
  • 6d14b0af96 update README to reference gpu_family select widget tnixon 2023-05-24 22:45:26 +00:00
  • 1cfeb58c6d set pipeline args for v100 GPUs tnixon 2023-05-24 22:44:34 +00:00
  • 963cf0528f removed a10 gpu configs from this branch tnixon 2023-05-23 20:51:58 +00:00
  • 1ea257f563 setting memory & fp options tnixon 2023-05-18 05:30:06 +00:00
  • e021eb451d configuration for V100 GPUs tnixon 2023-05-18 05:29:30 +00:00
  • 81585297c7 configuration as specified in README for A10 GPUs tnixon 2023-05-17 23:37:48 +00:00
  • e88abca686 notebook param (widget) to select gpu family tnixon 2023-05-17 22:53:56 +00:00
  • cda4eeb481 add trainer args for supporting fp16 mode tnixon 2023-05-17 22:53:33 +00:00
  • ec8cb63a1f cloning config file for v100 gpu family tnixon 2023-05-17 22:52:58 +00:00
  • f8ef041263 cloning config files for each GPU family tnixon 2023-05-17 22:52:11 +00:00
  • 3725600f2c Merge pull request #164 from xuanyuanking/add_dolly_as_input_model Sean Owen 2023-05-16 13:29:57 -05:00
  • a28d031e33 fix Yuanjian Li 2023-05-16 10:47:58 -07:00
  • 6238b936b4 revert Yuanjian Li 2023-05-15 16:31:54 -07:00
  • eb8f146dfc Merge pull request #162 from SamiKalliomaki/fix-docs Sean Owen 2023-05-15 09:24:11 -05:00
  • dc053397c8 add dolly as input model yuanjian.li@databricks.com 2023-05-14 22:49:11 +00:00
  • 24ced1a0e3 Fix documentation for max_new_tokens. Sami Kalliomäki 2023-05-14 11:25:00 +01:00
  • fc22734003 Merge pull request #144 from srowen/DatasetHF Sean Owen 2023-05-03 16:09:24 -05:00
  • cad7b3fefe Update to Trainer.train to Allow Override Dataset (#142) rmosleydb 2023-05-03 15:47:26 -05:00
  • a9f11db96a Note that dataset should be used from Hugging Face now. Sean Owen 2023-05-03 15:08:30 -05:00
  • 5f9bfba300 Merge pull request #130 from srowen/deepspeed083 Sean Owen 2023-04-25 11:28:21 -05:00
  • 8212ae2501 Drop back to deepspeed 0.8.3 because of issues with 0.9.x Sean Owen 2023-04-25 09:30:28 -05:00
  • 34c0bd289c Merge pull request #123 from srowen/Dataset15k Sean Owen 2023-04-21 18:31:02 -05:00
  • 3fd1286c3a Merge pull request #122 from srowen/UpdateReqs Sean Owen 2023-04-21 18:30:33 -05:00
  • 255c14970b Reference HF dataset by default, now that it's live Sean Owen 2023-04-21 14:11:50 -05:00
  • 6a0c06e93d Update reqs to match DBR 13; add torch Sean Owen 2023-04-21 13:56:48 -05:00
  • 56e7c0f40f Merge pull request #113 from srowen/FixPythia28 Sean Owen 2023-04-20 12:26:57 -05:00
  • 54e3f78946 Fix reference to pythia-2.8b Sean Owen 2023-04-20 12:24:22 -05:00
  • d69a93b90e Merge pull request #106 from srowen/UpdateOtherTraining Sean Owen 2023-04-19 14:02:46 -05:00
  • 3e8c659f76 Improve batch size guidance for other instance training Sean Owen 2023-04-19 08:43:04 -05:00
  • 0eadcb7b06 load_model_tokenizer_for_generate should raise ValueError (#99) Matthew Hayes 2023-04-17 12:04:23 -07:00
  • dcef926686 Suggest bfloat16 and add generation notes for A10, V100 (#95) Sean Owen 2023-04-16 23:58:00 -05:00
  • 3ecb8d9556 Update notes for training on A10/V100 (#94) Sean Owen 2023-04-16 22:25:47 -05:00
  • 662d181ce7 Add NVIDIA library installs for DBR 13.0 (#93) Sean Owen 2023-04-16 21:52:00 -05:00
  • 905e58a1ef Dolly V2 Updates (#88) Matthew Hayes 2023-04-15 16:49:21 -07:00
  • 3ea242cb55 fix typo in trainer.py (#81) Ikko Eltociear Ashimine 2023-04-15 04:34:59 +09:00
  • 7bdfa8557c Fix new line in language / version (#70) Reynold Xin 2023-04-14 04:29:00 +08:00
  • 5d3d5cf84b Merge pull request #71 from databrickslabs/data-update-apr13 Mike Conover 2023-04-13 11:22:00 -06:00
  • 2305eb7f2f Removing record from dataset. Mike Conover 2023-04-13 11:14:33 -06:00
  • a1b02ec5b2 Update README (#61) Matthew Hayes 2023-04-12 13:00:38 -07:00
  • 6724e9e207 Update README (#59) Matthew Hayes 2023-04-12 09:56:14 -07:00
  • d000e30309 Databricks Dolly 15k (#56) Matthew Hayes 2023-04-11 23:28:41 -07:00
  • e950ab3375 Improve tokenization to work with other tokenizers (#40) Matthew Hayes 2023-03-30 16:36:26 -07:00
  • 8398e3a236 Add special tokens for keys in prompt (#39) Matthew Hayes 2023-03-30 13:42:25 -07:00
  • e8c517535f Update README Matthew Hayes 2023-03-30 10:01:43 -07:00
  • e876704bf4 Update README Matthew Hayes 2023-03-30 09:09:23 -07:00
  • 05753f2376 Increase eval_steps to 100 (#31) Sean Owen 2023-03-28 16:52:54 -05:00
  • 50126d1c59 Add note on A10, V100 training (#30) Sean Owen 2023-03-28 11:24:55 -05:00
  • af91832c66 Add response generation for examples from blog post (#27) Matthew Hayes 2023-03-27 22:44:13 -07:00
  • f0de3ce986 Add note about selecting A100 instances (#22) Sean Owen 2023-03-28 00:03:28 -05:00
  • 596ab07022 Clarify that train_dolly NB is train_dolly.py (#23) Sean Owen 2023-03-27 23:59:03 -05:00
  • d29393a658 Merge pull request #20 from databrickslabs/mike-conover-db-patch-1 Mike Conover 2023-03-27 09:34:28 -06:00
  • 6441251e46 Updating README Mike Conover 2023-03-27 09:30:06 -06:00
  • ea46513792 Remove empty responses, append end tokens to training dataset Matt Hayes 2023-03-25 07:56:45 -07:00
  • 03bf3852da Update README.md Matt Hayes 2023-03-24 15:47:42 -07:00
  • 72c3557af8 Trainer config fixes Matt Hayes 2023-03-24 14:31:44 -07:00
  • 57cdfcd7da Move directory, add test instructions Matt Hayes 2023-03-24 12:17:57 -07:00
  • f9fc005bed Hello Dolly Matt Hayes 2023-03-24 10:38:23 -07:00
  • 4b060638ce Initial commit Serge Smertin 2023-03-24 17:15:56 +01:00