---
license: mit
tags:
- t1_synthesize_knowledge
- base_evaluation
- countdown
---

# t1-base-eval-hosted_vllm-qwen-qwen3-1-7b-5arg

Base evaluation of hosted_vllm/Qwen/Qwen3-1.7B on 5-arg countdown tasks. 100 problems, 8 samples each.

## Dataset Info

- **Rows**: 100
- **Columns**: 9

## Columns

| Column | Type | Description |
|--------|------|-------------|
| question | Value('string') | The countdown problem prompt text |
| metadata | Value('string') | JSON with target number, available numbers, and solution expression |
| task_source | Value('string') | Task type identifier (always 'countdown') |
| formatted_prompt | List({'content': Value('string'), 'role': Value('string')}) | Chat-formatted prompt sent to the model |
| response | List(Value('string')) | List of model response strings (one per sample) |
| response__extractions | List(List(Value('string'))) | Extracted answers from <answer> tags per sample |
| response__extractions_metadata | List(List({'end_index': Value('int64'), 'end_phrase': Value('string'), 'start_index': Value('int64'), 'start_phrase': Value('string')})) | Extraction position metadata |
| eval_correct | List(List(Value('bool'))) | List of booleans indicating correctness per sample |
| eval_correct__metadata | List(List({'answer_block': Value('string'), 'error': Value('string'), 'final_answer': Value('float64'), 'is_correct': Value('bool')})) | Detailed evaluation metadata per sample |


## Generation Parameters

```json
{
  "script_name": "01_base_eval.py",
  "model": "hosted_vllm/Qwen/Qwen3-1.7B",
  "hyperparameters": {
    "temperature": 0.7,
    "top_p": 0.95,
    "max_tokens": 16384,
    "enable_thinking": false
  },
  "input_datasets": [],
  "description": "Base evaluation of hosted_vllm/Qwen/Qwen3-1.7B on 5-arg countdown tasks. 100 problems, 8 samples each.",
  "custom_metadata": {
    "experiment_name": "t1_synthesize_knowledge_improvement",
    "stage": "base_evaluation",
    "num_examples": 100,
    "num_args": 5,
    "samples": 8,
    "seed": 42,
    "metrics": {
      "pass@1": 0.28,
      "pass@2": 0.36,
      "pass@3": 0.4,
      "pass@4": 0.47,
      "pass@5": 0.5,
      "pass@6": 0.51,
      "pass@7": 0.56,
      "pass@8": 0.58,
      "total_problems": 100,
      "samples_per_problem": 8
    }
  }
}
```

## Experiment Documentation

For complete experiment details, see [https://github.com/reasoning-degeneration/experiments/t1_synthesize_knowledge_improvement](https://github.com/reasoning-degeneration/experiments/t1_synthesize_knowledge_improvement)

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("reasoning-degeneration-dev/t1-base-eval-hosted_vllm-qwen-qwen3-1-7b-5arg", split="train")
print(f"Loaded {len(dataset)} rows")
```

---

*This dataset is tracked in [reasoning-degeneration-dev/PROJECT-MANIFEST](https://huggingface.co/datasets/reasoning-degeneration-dev/PROJECT-MANIFEST)*