Datasets:

AdamLucek
/

Qwen3-4B-Instruct-2507-PII-RL-pii-masking-eval

dataset_info:
  features:
    - name: example_id
      dtype: int64
    - name: prompt
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: completion
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: task
      dtype: string
    - name: reward
      dtype: float64
    - name: generation_ms
      dtype: float64
    - name: scoring_ms
      dtype: float64
    - name: total_ms
      dtype: float64
    - name: info
      struct:
        - name: pii_count
          dtype: int64
    - name: answer
      dtype: string
    - name: exact_match_reward
      dtype: float64
    - name: pii_count_reward
      dtype: float64
    - name: format_reward_func
      dtype: float64
  splits:
    - name: train
      num_bytes: 176352
      num_examples: 150
  download_size: 26186
  dataset_size: 176352
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Evaluation Results: Qwen3-4B-Instruct-2507-PII-RL on PII Masking

This dataset contains evaluation results for the RL-trained model AdamLucek/Qwen3-4B-Instruct-2507-PII-RL on the adamlucek/pii-masking environment from Prime Intellect's Environment Hub. The model was fine-tuned using reinforcement learning to mask personally identifiable information (PII) in text.

Evaluation Configuration

Environment: adamlucek/pii-masking
Model: AdamLucek/Qwen3-4B-Instruct-2507-PII-RL
Base Model: Qwen/Qwen3-4B-Instruct-2507
Examples Evaluated: 50
Rollouts per Example: 3
Total Samples: 150

Performance Summary

Overall Metrics

Metric	Mean	Std Dev	Min	Max
Total Reward	0.883	0.729	0.100	1.600
Exact Match	0.500	0.502	0.000	1.000
PII Count Accuracy	0.567	0.497	0.000	1.000
Format Compliance	1.000	0.000	1.000	1.000

Reward Breakdown by Rollout

Rollout	Mean Reward	Std Dev	Range
Rollout 1	0.880	0.737	[0.100, 1.600]
Rollout 2	0.890	0.729	[0.100, 1.600]
Rollout 3	0.880	0.737	[0.100, 1.600]