Commit graph

1244 commits

Author SHA1 Message Date
Zafir Stojanovski
695aad4dbc fix(training): Prepend <think> token in format reward (#396)
* prepend think token in format reward

* pre commit + fix some default vals

* add checkpoint config
2025-03-28 09:45:17 +01:00
Tiago Serafim
ed560a80f4 Fix small bit of old code (#386)
Removed the `sat_utils.` that was left there from the original zebra package.

This allows the zebra generator to be used with the full set of available clues.
2025-03-24 09:00:28 +00:00
Zafir Stojanovski
915e36fd53 remove coach and update tests (#393) 2025-03-24 00:10:01 +01:00
Oliver Stanley
9304b23c1b initial verl training codebase (#389)
* fixes for latest verl
* composite dataset training experiment
* use stateful dataloaders to match verl changes
* training readme
* add formatting reward
* length reward impl
* standalone reasoning_gym config section
* curriculum learning, new length reward, more config
2025-03-20 15:04:57 +00:00
Zafir Stojanovski
4c47527130 fix(envs): Add source dataset and index to metadata (#388)
* add source dataset and index to metadata

* fix typo

* fix coach class and its test
2025-03-20 11:12:14 +00:00
Oliver Stanley
c6d01541aa include ranges rather than sampled values in difficulty metadata dicts (#387)
* update difficulty metadata for logic datasets

* update difficulty metadata for graph datasets

* update difficulty metadata for geometry datasets

* update difficulty metadata for games datasets

* update difficulty metadata for cognition datasets

* update difficulty metadata for arithmetic datasets

* update difficulty metadata for arc datasets

* update difficulty metadata for algorithmic datasets

* update difficulty metadata for algebra datasets

* use tuples

* update tests

* update tests
2025-03-20 10:27:03 +01:00
Rich Jones
a774a10843 fix figlet font curr imports (#383) 2025-03-18 23:38:17 +01:00
Zafir Stojanovski
4806a5ac61 add difficulty metadata 2025-03-18 13:51:43 +01:00
Zafir Stojanovski
7404947bc1 figlet font curriculum 2025-03-18 13:51:43 +01:00
joesharratt1229
1da84a0b41 Feat/open instruct example (#381)
* added open-instruct

* fixed hooks

* GRPO

---------

Co-authored-by: Andreas Koepf <andreas.koepf@provisio.com>
2025-03-17 23:20:11 +01:00
Andreas Koepf
eaef88b45b don't pass answer value to eval 2025-03-17 23:13:53 +01:00
Jean Kaddour
2e9877ae07 fix: add score_answer() to number_sorting (#380)
* fix: add score_answer() to number_sorting

* chore: run pre-commit

* fix: use json.loads()

* fix: run isort()
2025-03-17 23:04:13 +01:00
Oliver Stanley
a3d89e338b use StatefulDataLoader in veRL examples (#378) 2025-03-17 07:28:10 +01:00
Andreas Koepf
27c84344ca adapt needle_haystack test to curriculum changes 2025-03-16 17:35:16 +01:00
Andreas Koepf
2ef45201b7 bump version 2025-03-16 15:54:03 +01:00
Andreas Koepf
8a3c29e1ee reduce values in NeedleHaystackCurriculum 2025-03-16 15:52:48 +01:00
Andreas Köpf
ed766028fb Refactor Curriculum Attributes (#335)
* remove min_value from AttributeDefinition
* remove type from AttributeDefinition
* Add CurriculumContext
* add ensure_interval option for RangeAttributes
* docs: Add legend explaining curriculum indicators in dataset gallery
* update GALLERY.md
2025-03-16 15:40:28 +01:00
Andreas Köpf
d344284c80 feat: Add comprehensive visualization script for evaluation results (#376) 2025-03-16 12:26:27 +01:00
Andreas Köpf
677a2af03e Add eval configs, small fixes to eval script & rush-hour score_answer 2025-03-16 09:18:05 +01:00
Andreas Koepf
fa950d0189 add gemma-3-27b & qwq-32b configs 2025-03-15 20:47:51 +01:00
Andreas Koepf (aider)
a829707408 feat: Add fallback to first non-None model answer when best_answer is unset 2025-03-15 16:52:50 +01:00
Andreas Köpf
424ee6751a Eval N completions per prompt (#374)
* feat: Add support for generating multiple completions per prompt
* feat: Track best and mean scores for multiple completions per prompt
* feat: Add checkpoint and resume functionality to evaluation script
2025-03-15 16:39:36 +01:00
Oliver Stanley
1d410cc600 Fix chain sum veRL example for latest veRL (#371)
* fixes for latest verl

* add balance_batch cofg

* 1 -> 2 gpu

* tweaks

* also add raw ids to server script
2025-03-14 20:15:54 +01:00
Adefioye
7bab2ecef8 Add jugs curriculum (#369) 2025-03-14 18:04:33 +01:00
Andreas Koepf
615acb9d84 bump version 2025-03-14 16:21:34 +01:00
Andreas Koepf
e4063f1f85 update gallery 2025-03-14 16:18:43 +01:00
Zafir Stojanovski
843a277248 circuit logic curriculum (#368) 2025-03-14 16:17:00 +01:00
Rich Jones
3e0254afea Add curricula status to gallery (#367)
* Add curricula status to gallery

* lint
2025-03-14 16:16:26 +01:00
Rich Jones
4006200132 GoL-Halt Curricula (#366)
* GoL-Halt Curricula

* trivial
2025-03-14 16:15:45 +01:00
Zafir Stojanovski
75dfd8ffed feat(env): Propositional Logic Curriculum (#365)
* propositional logic curriculum

* lint

* difficulty meta
2025-03-14 16:12:39 +01:00
Zafir Stojanovski
1b6b13566a time intervals curriculum (#363) 2025-03-14 16:11:55 +01:00
Oliver Stanley
8ed8f0b696 add rush hour curriculum (#362)
* add rush hour curriculum

* add to __init__.py
2025-03-14 16:11:23 +01:00
Oliver Stanley
0fdb8f77dd add word ladder curriculum (#361)
* add word ladder curriculum

* add to __init__.py
2025-03-14 16:10:52 +01:00
Oliver Stanley
fe0cf2291f add sudoku curriculum (#360)
* sudoku curriculum

* add test

* add to __init__.py
2025-03-14 16:09:49 +01:00
joesharratt1229
4df3a3b92b added family relatinship curriculum (#359) 2025-03-14 16:08:49 +01:00
joesharratt1229
9df5aabbcc added boxnet generator and curr (#364) 2025-03-14 08:48:57 +01:00
Adefioye
5e732de191 Simple equations curriculum (#340)
* Add curriculum to simple_equations and unit tests
* Add difficulty to metadata
2025-03-13 21:17:14 +01:00
joesharratt1229
28b9c63f4b added rearc curr (#358) 2025-03-13 21:14:03 +01:00
Zafir Stojanovski
5159307e27 feat(env): Rubiks Cube Curriculum (#357) 2025-03-13 21:12:32 +01:00
joesharratt1229
3975fafae5 added aiw curric (#356)
added metadata
2025-03-13 21:10:52 +01:00
Zafir Stojanovski
9d1692d2d1 feat(env): Needle in Haystack Curriculum (#355)
* needle haystack curriculum
2025-03-13 21:09:45 +01:00
Adefioye
02e0e3b0c5 Add gol curriculum (#354)
* Add gol curriculum

* Add difficulty

* Make levels of grid size of x and y be valid
2025-03-13 21:09:09 +01:00
Zafir Stojanovski
e8791d9f11 feat(env): Advanced Geometry Curriculum (#350)
* advanced geometry curriculum
* metadata

---------

Co-authored-by: Andreas Köpf <andreas.koepf@xamla.com>
2025-03-13 21:07:49 +01:00
Zafir Stojanovski
4e9d800540 fraction simplification curriculum (#349) 2025-03-13 21:05:50 +01:00
Zafir Stojanovski
322f878488 lcm curriculum (#348) 2025-03-13 21:05:06 +01:00
Zafir Stojanovski
61da151f0d color cube rotation curriculum (#347) 2025-03-13 21:04:34 +01:00
Adefioye
5e0ccf3499 Cryptarithm curriculum (#346)
* Add curriculum for cryptarithm
* Add difficulty to metadata
2025-03-13 21:03:57 +01:00
Adefioye
454250a4ea Add curriculum to ab dataset (#345)
* Add curriculum to ab dataset

* Add difficulty to metadata
2025-03-13 21:03:02 +01:00
Zafir Stojanovski
4f45c8d655 feat(env): Prime Factorization (#344)
* prime factorization

* lint
2025-03-13 21:02:20 +01:00
Zafir Stojanovski
36a7ca1150 maze curriculum (#343) 2025-03-13 21:01:48 +01:00