| # Synthetic sample data | |
| These JSONL files are synthetic examples for documenting the expected schema and running short functional checks without distributing the paper evaluation data. | |
| ## Files | |
| - sample_qa2.jsonl: small synthetic benchmark example. | |
| - sample_qa2_32k.jsonl: optional synthetic long-context example. | |
| ## JSONL schema | |
| Each line is one JSON object with the same fields as the full evaluation files: | |
| ```json | |
| { | |
| "index": 0, | |
| "input": "prompt text passed to the tokenizer/model", | |
| "outputs": ["acceptable answer"], | |
| "length": 1234, | |
| "length_w_model_temp": 1234, | |
| "answer_prefix": " Answer:" | |
| } | |
| ``` | |
| The benchmark script tokenizes `input` directly. `length` and `length_w_model_temp` are retained for schema compatibility, but are not used to choose prompt length. | |
| For the included `sample_qa2.jsonl`, use a small truncation target such as `--target-prompt-token-length 512`. The sample is not intended to reproduce paper numbers. | |
| For checking the 32k truncation path, use `sample_qa2_32k.jsonl`. For quick functional checks, prefer `sample_qa2.jsonl` with `--target-prompt-token-length 512`. | |