NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-19 12:57:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: correct dataset name to bigcode/humanevalpack

Browse source
This commit is contained in:
RUFFY-369 2026-03-24 15:32:25 +05:30
parent 590e8a1ef2
commit 5c2afa8ea7
2 changed files with 6 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

4
environments/community/code_debug_env/README.md
View file

@ -4,7 +4,7 @@ An Atropos RL environment for training LLMs to debug and fix buggy Python code.
## Overview
This environment uses the [HumanEvalFix](https://huggingface.co/datasets/bigcode/humanevalfix-python) dataset, which contains 164 buggy Python functions with associated test suites. The model receives a buggy function and must output the corrected version inside `\boxed{}`. Scoring is done by executing the fixed code against the original test cases.
This environment uses the [HumanEvalPack](https://huggingface.co/datasets/bigcode/humanevalpack) dataset (Python subset, HumanEvalFix task), which contains 164 buggy Python functions with associated test suites. The model receives a buggy function and must output the corrected version inside `\boxed{}`. Scoring is done by executing the fixed code against the original test cases.
## Architecture
@ -69,7 +69,7 @@ python code_debug_env.py evaluate \
## Dataset
- **Source**: [bigcode/humanevalfix-python](https://huggingface.co/datasets/bigcode/humanevalfix-python)
- **Source**: [bigcode/humanevalpack](https://huggingface.co/datasets/bigcode/humanevalpack) (Python subset)
- **License**: Apache 2.0
- **Size**: 164 problems
- **Split**: 80% train / 20% test

8
environments/community/code_debug_env/code_debug_env.py
View file

@ -2,7 +2,7 @@
Code Debug Environment for Atropos
Trains LLMs to debug and fix buggy Python functions.
Uses the HumanEvalFix dataset with execution-based verification
Uses the HumanEvalPack dataset (HumanEvalFix subset) with execution-based verification
against ground-truth test cases.
Environment pattern follows sql_query_env for consistency.
@ -162,11 +162,11 @@ class CodeDebugEnv(BaseEnv):
await super().wandb_log(wandb_metrics)
async def setup(self):
"""Load the HumanEvalFix dataset and prepare train/test splits."""
"""Load the HumanEvalPack dataset (HumanEvalFix) and prepare train/test splits."""
from datasets import load_dataset
print("Loading HumanEvalFix dataset...")
dataset = load_dataset("bigcode/humanevalfix-python", split="test")
print("Loading HumanEvalPack (python) dataset...")
dataset = load_dataset("bigcode/humanevalpack", "python", split="test")
all_items: List[CodeDebugItem] = []
for row in dataset: