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

some cleanup for final merge

Browse source
This commit is contained in:
teknium1 2025-05-16 19:24:50 -07:00
parent daa6f0ff18
commit 287bbcd356
2 changed files with 58 additions and 39 deletions
Download patch file Download diff file Expand all files Collapse all files

6
environments/README.md
View file

@ -93,7 +93,9 @@ You are a deep thinking AI, you may use extremely long chains of thought to deep
### Instruction Following Environment (`instruction_following_algorithm_environment.py`)
Environment for training models to follow natural language instructions and constraints, based on the `allenai/RLVR-IFeval` dataset and environment.
*This environment has unique dependencies: `datasets` (from Hugging Face) and `langdetect`.*
**Dependencies:**
- `datasets` (Hugging Face)
- `langdetect`
**Input Format:**
- Each item from the processed `allenai/RLVR-IFeval` dataset contains:
@ -120,7 +122,7 @@ You are a deep thinking AI, you may use extremely long chains of thought to deep
- `dataset_config_name`: Optional name for a specific configuration or subset of the dataset.
- `test_set_ratio`: Defines the proportion of the dataset reserved for testing (defaults to 5%).
- **Verifier-Based Scoring:** Utilizes a comprehensive map of verifier functions (`IF_FUNCTIONS_MAP`) to evaluate whether the model's
- **Verifier-Based Scoring:** Utilizes a comprehensive map of verifier functions (`IF_FUNCTIONS_MAP`) to evaluate whether the model's
output adheres to diverse and specific constraints defined in the input instructions (e.g., keyword presence, response length, JSON format, etc.).
- **Specialized Dataset Processing:** The `setup` method is specifically designed to parse the `allenai/RLVR-IFeval` dataset, extracting user instructions, the corresponding verifier function name, and its arguments.