115 Commits

Author SHA1 Message Date
Evian Schlenz
f96ded3abb Annotate generate_system_prompt 2025-12-31 01:48:47 +01:00
Evian Schlenz
5a3d37c56a Annotate format_example_sharegpt 2025-12-31 01:48:44 +01:00
Evian Schlenz
5ceff59c65 move PileOfTypes to utils 2025-12-31 01:35:36 +01:00
Evian Schlenz
242655af84 annotate random vars and small funcs 2025-12-31 01:21:27 +01:00
Evian Schlenz
c1b5d912d2 Annotate generate_random_parameter and get_random_response 2025-12-31 01:12:22 +01:00
Evian Schlenz
b3df3d5346 correctly type example generation functions 2025-12-31 00:58:22 +01:00
Evian Schlenz
44b296e6df correctly type get_random_state 2025-12-31 00:57:12 +01:00
Evian Schlenz
80669b6522 Rename pile types to not confuse with other type classes 2025-12-30 23:51:27 +01:00
Evian Schlenz
c26ef953c6 Do not use mutable objects as default 2025-12-30 23:47:54 +01:00
Evian Schlenz
b955897c18 make TypedDict types for piles 2025-12-30 18:31:47 +01:00
Alex O'Connell
12ba6d649d clean up readme and add new model 2025-12-21 23:03:44 -05:00
Alex O'Connell
1811a907f7 manually set roles to train on 2025-12-21 22:09:18 -05:00
Alex O'Connell
cf01fd29ae synthesize new data, update training job/configs 2025-12-21 14:14:31 -05:00
Alex O'Connell
4407aefdf5 more synthesizing scenarios + clean up example formatting 2025-12-21 13:31:43 -05:00
copilot-swe-agent[bot]
ecf9586b5a chore: localize new dataset piles
Co-authored-by: acon96 <35843486+acon96@users.noreply.github.com>
2025-12-21 04:32:25 +00:00
copilot-swe-agent[bot]
6e667b17cc fix: address dataset generation review feedback
Co-authored-by: acon96 <35843486+acon96@users.noreply.github.com>
2025-12-21 04:28:33 +00:00
copilot-swe-agent[bot]
eee2a6ed11 feat: add failure and refusal dataset examples
Co-authored-by: acon96 <35843486+acon96@users.noreply.github.com>
2025-12-21 04:26:09 +00:00
Alex O'Connell
29d839eea8 mostly working gemma implementation 2025-12-20 20:29:09 -05:00
Alex O'Connell
ee2d2e7640 lets try functiongemma instead 2025-12-20 10:24:08 -05:00
Alex O'Connell
04e9bc1ee6 Merge branch 'develop' into feature/dataset-new-apis 2025-12-14 20:43:52 -05:00
Alex O'Connell
dac9973cb5 reformat response piles to include confirmation and conclusion 2025-12-07 17:28:38 -05:00
Alex O'Connell
55f254149a start re-working training to use axlotl instead of the custom script 2025-11-30 22:29:08 -05:00
Alex O'Connell
04a5909214 organize training notebook 2025-11-30 16:24:39 -05:00
Alex O'Connell
9f51dd0e94 more training updates 2025-11-30 15:31:58 -05:00
Alex O'Connell
25b6ddfd0c fix dev requirements 2025-11-30 14:10:55 -05:00
Alex O'Connell
61140713d7 refactor dataset generation code + add new synthesis script 2025-11-27 22:19:38 -05:00
Alex O'Connell
753a990a98 add new names 2025-11-26 22:08:33 -05:00
Alex O'Connell
14640bd14b finish implementing alternate dataset generation mode 2025-11-26 22:01:08 -05:00
Alex O'Connell
07507ee5f5 more fixes 2025-11-26 19:09:09 -05:00
Alex O'Connell
a16523f9e5 start re-writing dataset generation to use the new HA Assist API 2025-11-26 19:08:46 -05:00
Alex O'Connell
73f1d82c76 fix bug in data generation script 2025-03-08 22:29:22 -05:00
Alex O'Connell
2712f605a5 fix evaluate + add train notebook 2025-02-10 17:11:44 -05:00
Alex O'Connell
fca80d0504 reorganize requirements files 2025-02-09 22:12:18 -05:00
Alex O'Connell
22c9469f66 try to support SFT for models without system prompts 2024-08-17 22:43:05 -04:00
Witold Gren
2837af8443 Added full translations for all languages during generate data and creating default prompt system (#196) 2024-08-11 22:05:20 +00:00
Witold Gren
cf89f9f478 Added support for Polish language (#193) 2024-08-04 19:19:10 +00:00
Alex O'Connell
179e794283 add cmdline arguments to translate script + add defaults for command r 2024-05-08 20:50:25 -04:00
Alex O'Connell
03c23f1a8c local translation + training update 2024-04-24 19:01:05 -04:00
Ryan Voots
eafb26267b Correct the DeviceType for todo (#121) 2024-04-24 21:37:32 +00:00
Alex O'Connell
adae87addd training fixes, default values + other fixes 2024-04-21 23:40:28 -04:00
Alex O'Connell
bdda97eb45 add date to system prompt + per language and words 2024-04-21 20:48:44 -04:00
Alex O'Connell
3326bd7d6e handle other languages in component 2024-04-21 20:38:46 -04:00
Alex O'Connell
8144267e69 fix templated action quotes + add fsdp config 2024-04-14 18:08:11 -04:00
Alex O'Connell
f5bab7b119 better dpo parameters 2024-04-14 15:42:51 -04:00
Alex O'Connell
85cd5ec036 more DPO example types 2024-04-14 08:02:20 -04:00
Alex O'Connell
ce75bf0d7c very basic DPO data generator is working now 2024-04-13 22:47:07 -04:00
Alex O'Connell
547d2d9989 add proper multi-language support to dataset generator 2024-04-13 22:15:37 -04:00
Alex O'Connell
f1659893d7 start working on dpo for the datasets 2024-03-19 21:31:34 -04:00
Alex O'Connell
b9d394f860 actually fix cover service names 2024-03-16 13:50:04 -04:00
Alex O'Connell
9b71a1860b fix "cover" type system call names 2024-03-06 20:40:25 -05:00