Zafir Stojanovski
0cda6b1205
qwen math training code ( #435 )
...
* qwen math training code
* pre-commit
2025-05-16 13:19:19 +02:00
Oliver Stanley
47303211b3
update gallery ( #434 )
2025-05-15 22:41:43 +02:00
Zafir Stojanovski
6d9e14729a
fix rubiks cube prompt ( #433 )
2025-05-15 21:31:21 +01:00
Oliver Stanley
85f3c6dd02
updated inter-domain generalisation eval configs ( #432 )
...
* tweak eval configs
* add eval configs
* add eval config
2025-05-15 09:08:16 +02:00
Zafir Stojanovski
4cab1c3e6d
comparison plots ( #431 )
2025-05-12 15:41:58 +02:00
Oliver Stanley
3df26e0fb2
fix decimal arithmetic curriculum to respect constraints ( #430 )
...
* fix decimal arithmetic curriculum to respect constraints
* update test accordingly
2025-05-06 19:23:19 +01:00
rasdani
dca8117e7e
Feat: Add script to save datasets on HuggingFace ( #416 )
...
* feat: add script to save datasets on HuggingFace
* fix
* refactor
* fix formatting
---------
Co-authored-by: Oliver Stanley <olivergestanley@gmail.com >
2025-04-28 18:04:53 +01:00
joesharratt1229
73e3cb33a4
Added games training and evaluation configuration ( #426 )
...
* added games
* Update eval_games_composite.yaml
* Delete training/evaluations/eval_qwen_3b.yaml
* Add files via upload
* Delete training/evaluations/eval_algebraic_composite.yaml
* Delete training/evaluations/eval_algorithmic_composite.yaml
* Delete training/evaluations/eval_arithmetic_composite.yaml
* Delete training/evaluations/eval_cognition_composite.yaml
* Delete training/evaluations/eval_games_composite.yaml
2025-04-26 19:45:32 +01:00
Oliver Stanley
10863ea12b
inter-domain generalisation evaluation configs ( #424 )
...
* add inter-domain generalisation eval config for algebra
* add algorithmic eval cfg
* vllm infer
* add arithmetic eval cfg
* add geometry eval cfg
* add arc cfg
* add games eval cfg
* add cognition eval cfg
* add graphs eval cfg
2025-04-22 17:32:35 +01:00
Zafir Stojanovski
98e976642d
gemini flash and o3 mini configs ( #425 )
2025-04-20 21:24:52 +02:00
rasdani
72e45e9401
Feat: expose score_answer function without needing to instantiate a dataset ( #422 )
...
* feat: get `score_answer` for a given dataset
* fix: `self` error
* add test
2025-04-18 10:36:44 +02:00
Adefioye
169d8c3aec
Evals: Sonnet-3.7-eval partial results ( #423 )
...
* Add results for sonnet and config
* Make some cleanup
2025-04-18 10:31:40 +02:00
joesharratt1229
d0ef136d5b
Feat/intragen experiments ( #414 )
...
* added curriculum
* readapted readme
* corrected small errors
* Delete eval/eval/r1/algorithmic/word_sorting.json
* removed redundant argument
* added spell
* removed duplicated fit
* changed config
* added composite changes
* added composite changes
* updated yaml
* added spell backward
* updated read me
* added qwen2.5
* added
* Add files via upload
* updated missing trainer func
* updated curr
* updated spell back
* updated correctness score func
* updated configs
* added local evals
* added updates
* updated datasets
* added fsdp to hf utility
* added algorithmic qwen 3b yaml
* updated read me
* updated configs
* added preappend token
* updated with thinking token
* updated test score board
* resolved comments
* added evaluation scripts
* removed results from pr
* added config
* added partial reward scoring
* added evaluation composites
* added training configs
* added games eval
* added rubriks cube
* resolved merge cinflicts
* added games config
* added latest eval configs
* updated strucutre
* Delete training/evaluations/eval_graphs_composite.yaml
---------
Co-authored-by: joesharratt1229 <joesharrat1229@gmail.com >
2025-04-16 08:04:52 +02:00
Oliver Stanley
224532f12a
first inter-domain generalisation experiments ( #412 )
...
* tweak len reward
* first inter-generalisation experiment config
* update inter algorithmic config
* default to empty config
* fix typo
* change config to match experiment script
* long prompt fixes
* algorithmic training config tweaks
* imports
* update algorithmic training cfgs
* first logic composite config
* fix dset name
* tweaks
* fix syllogisms dataset
* rm temp print
* initial algebra config
* algebra cfg tweaks
* add gc
* add initial games cfg
* rename games cfg
* fix dset name
* fix sokoban metadata
* remove boxnet
* games cfg tweak
2025-04-14 21:06:40 +01:00
Oliver Stanley
ff5407f766
fix boxnet scoring function ( #420 )
...
* avoid error print on invalid move format in tower of hanoi
* avoid boxnet error on hallucinated starting posittions
2025-04-14 20:44:24 +01:00
Zafir Stojanovski
71787c6a0e
fix(env): Remove custom score function in bf ( #418 )
...
* remove custom score function in bf
* pre commit
2025-04-14 11:26:18 +02:00
Zafir Stojanovski
290bfc4fdd
(evals): Medium configs ( #415 )
...
* updated medium configs
* fix problematic curriculum values / small issues causing exceptions to be raised
* optimus alpha config
* all configs so far
* fix tests
2025-04-14 08:25:31 +02:00
Andreas Köpf
cd1a9ea58b
add llama-4-maverick eval yaml config ( #413 )
2025-04-06 21:10:00 +01:00
Zafir Stojanovski
dced3bfc45
fix(curriculum): Make boundaries in curriculum more sensible ( #407 )
...
* init
* fix tests
* unify codeio
* filtered for libraries not present in reasoning-gym
* fix more bounds
* puzzle24
* knight swap curriculum
* fix number sorting
* fix attributes
* add validation of config in creation of dataset
* dry run for instantiating and validating the datasets
* remove unused imports
* fix curriculum tests to reference newly updated attribute names
2025-04-04 20:24:14 +02:00
Andreas Köpf
7853263650
bump dev version to 0.1.19
2025-04-03 22:19:07 +01:00
Zafir Stojanovski
dafdee621e
fix(env): Unify CodeIO datasets ( #405 )
...
* unify codeio
* filtered for libraries not present in reasoning-gym
2025-04-02 22:40:03 +02:00
Adefioye
5b653b346c
Data collisions notebooks and data ( #406 )
...
* Add collisions data
* Fix logic issues in basic_arithmetic and gsm_symbolic data
2025-04-02 09:36:09 +02:00
joesharratt1229
43c739cb3e
Feat/curr adj ( #394 )
2025-04-02 06:39:14 +01:00
Roman Machacek
2c52f33c3a
CodeIO HQ Dataset ( #382 )
...
* ADD: CodeIO high quality dataset
Based on the dataset for CodeI/O. Annotated using Qwen-Coder and filtered based on the various metrics resulting in high quality filtered dataset, where approx 50% of the original data is kept.
* ADD: Compressed version
* Delete pure json version
2025-04-01 22:34:33 +02:00
Oliver Stanley
415bcb5ace
use constant for dataset name in arc agi ( #404 )
2025-04-01 22:06:31 +02:00
Zafir Stojanovski
ec485cce24
Add banner in README.md ( #399 )
2025-04-01 21:29:41 +02:00
Adefioye
e3af2dd2bd
Add curriculum for arc_agi ( #372 )
...
* Add curriculum for arc_agi
* Resolve conflicts
* Remove code smell
* Remove unwanted code
2025-04-01 21:17:52 +02:00
Zafir Stojanovski
50846c3534
fix(env): ARC 1D curriculum ( #402 )
...
* Add arc_1d curriculum
* Add difficulty to metadata
* use range attribute instead of scalar
---------
Co-authored-by: abdulhakeem <abdulhakeemadefioye@gmail.com >
Co-authored-by: Oliver Stanley <olivergestanley@gmail.com >
2025-04-01 13:01:15 +02:00
vncntt
cd85c2d632
add knights knaves curriculum ( #401 )
...
* add knights knaves curriculum
* add metadata + width constraints
2025-04-01 12:20:58 +02:00
Oliver Stanley
ea10a0f932
update task count in readme ( #400 )
...
* update task count in readme
* fix link
2025-04-01 10:51:36 +02:00
Zafir Stojanovski
8c45571a48
visualize heatmap sorted by overall performance ( #397 )
2025-04-01 00:08:39 +02:00
Zafir Stojanovski
c6663cdb81
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
7ae2942c34
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
2a970390e8
remove coach and update tests ( #393 )
2025-03-24 00:10:01 +01:00
Oliver Stanley
eb69916c1b
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
ce0a6c4878
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
7475a20700
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
b69c35818a
fix figlet font curr imports ( #383 )
2025-03-18 23:38:17 +01:00
Zafir Stojanovski
814da6e08a
add difficulty metadata
2025-03-18 13:51:43 +01:00
Zafir Stojanovski
29bf78293f
figlet font curriculum
2025-03-18 13:51:43 +01:00
joesharratt1229
9234aa77bf
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
1511c5e301
don't pass answer value to eval
2025-03-17 23:13:53 +01:00
Jean Kaddour
d6aad5a329
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
1c6f2d01ee
use StatefulDataLoader in veRL examples ( #378 )
2025-03-17 07:28:10 +01:00
Andreas Koepf
c760da6b1b
adapt needle_haystack test to curriculum changes
2025-03-16 17:35:16 +01:00
Andreas Koepf
f25820a0c7
bump version
2025-03-16 15:54:03 +01:00
Andreas Koepf
08b422f673
reduce values in NeedleHaystackCurriculum
2025-03-16 15:52:48 +01:00
Andreas Köpf
d2c895f1d3
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
4e7d9296ee
feat: Add comprehensive visualization script for evaluation results ( #376 )
2025-03-16 12:26:27 +01:00
Andreas Köpf
d6f399b8e4
Add eval configs, small fixes to eval script & rush-hour score_answer
2025-03-16 09:18:05 +01:00
