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>
This commit is contained in:
joesharratt1229 2025-04-16 07:04:52 +01:00 committed by GitHub
parent 224532f12a
commit d0ef136d5b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
21 changed files with 1331 additions and 48 deletions

View file

@ -6,6 +6,7 @@ from reasoning_gym.arithmetic.basic_arithmetic import (
BasicArithmeticDatasetConfig,
eval_floordiv,
)
from reasoning_gym.coaching.base_curriculum import DefaultCurriculumContext, RangeAttributeMode
def test_arithmetic_dataset_config_validation():
@ -103,7 +104,7 @@ def test_basic_arithmetic_curriculum():
"""Test the BasicArithmeticCurriculum functionality"""
curriculum = BasicArithmeticCurriculum()
base_value = {"size": 150, "seed": 1}
base_value = {"size": 150, "seed": 1, "min_terms": 2, "max_terms": 2, "min_digits": 1, "max_digits": 1}
base_cfg: BasicArithmeticDatasetConfig = curriculum.generate_configuration(base_value)
assert base_cfg.seed == 1
@ -115,7 +116,7 @@ def test_basic_arithmetic_curriculum():
curriculum.increment_attr_level("num_terms")
curriculum.increment_attr_level("num_digits")
increased_cfg = curriculum.generate_configuration(base_value)
assert increased_cfg.min_terms == 2 and increased_cfg.max_terms == 5
assert increased_cfg.min_terms == 2 and increased_cfg.max_terms == 3
assert increased_cfg.min_digits == 1 and increased_cfg.max_digits == 2
# Test decrementing attribute level for num_terms
@ -128,7 +129,7 @@ def test_basic_arithmetic_curriculum():
curriculum.increment_attr_level("num_terms")
curriculum.increment_attr_level("num_terms")
higher_level_cfg = curriculum.generate_configuration(base_value)
assert higher_level_cfg.min_terms == 2 and higher_level_cfg.max_terms == 10
assert higher_level_cfg.min_terms == 2 and higher_level_cfg.max_terms == 4
assert higher_level_cfg.min_digits == 1 and higher_level_cfg.max_digits == 2
# Test boundary conditions - trying to decrement below level 0
@ -144,5 +145,26 @@ def test_basic_arithmetic_curriculum():
curriculum.increment_attr_level("num_terms")
curriculum.increment_attr_level("num_digits")
upper_bound_cfg = curriculum.generate_configuration(base_value)
assert upper_bound_cfg.min_terms == 2 and upper_bound_cfg.max_terms == 15
assert upper_bound_cfg.min_digits == 1 and upper_bound_cfg.max_digits == 10
assert upper_bound_cfg.min_terms == 2 and upper_bound_cfg.max_terms == 6
assert upper_bound_cfg.min_digits == 1 and upper_bound_cfg.max_digits == 4
def test_basic_arithmetic_curriculum_upper_bound():
curriculum = BasicArithmeticCurriculum()
base_value = {"size": 150, "seed": 1, "min_terms": 2, "max_terms": 2, "min_digits": 1, "max_digits": 1}
base_cfg: BasicArithmeticDatasetConfig = curriculum.generate_configuration(
base_value, context=DefaultCurriculumContext(mode=RangeAttributeMode.UPPER_BOUND)
)
assert base_cfg.seed == 1
assert base_cfg.size == 150
assert base_cfg.min_terms == 2 and base_cfg.max_terms == 2
assert base_cfg.min_digits == 1 and base_cfg.max_digits == 1
# Test incrementing attribute levels
curriculum.increment_attr_level("num_terms")
curriculum.increment_attr_level("num_digits")
increased_cfg = curriculum.generate_configuration(base_value)
assert increased_cfg.min_terms == 2 and increased_cfg.max_terms == 3
assert increased_cfg.min_digits == 1 and increased_cfg.max_digits == 2