Datasets:

PNYX
/

T-Eval_pnyx

license: apache-2.0
viewer: true
task_categories:
  - text-generation
  - question-answering
language:
  - en
tags:
  - code
pretty_name: PNYX-T-Eval
size_categories:
  - 1K<n<10K
extra_gated_prompt: >-
  You agree to NOT reveal examples from this dataset in plain text or images
  online, to reduce the risk of leakage into foundation model training corpora.
extra_gated_fields:
  I accept these terms: checkbox
configs:
  - config_name: instruct_v2
    data_files:
      - split: test
        path: data/instruct_v2/test.parquet
  - config_name: instruct_tool_native_v2
    data_files:
      - split: test
        path: data/instruct_tool_native_v2/test.parquet
  - config_name: plan_json_v2
    data_files:
      - split: test
        path: data/plan_json_v2/test.parquet
  - config_name: plan_tool_native_v2
    data_files:
      - split: test
        path: data/plan_tool_native_v2/test.parquet
  - config_name: review_str_v2
    data_files:
      - split: test
        path: data/review_str_v2/test.parquet
  - config_name: reason_retrieve_understand_json_v2
    data_files:
      - split: test
        path: data/reason_retrieve_understand_json_v2/test.parquet
  - config_name: reason_retrieve_understand_tool_native_v2
    data_files:
      - split: test
        path: data/reason_retrieve_understand_tool_native_v2/test.parquet

PNYX - T-Eval

This is a splitted and formated version of the T-Eval dataset, refer to it for further information on the original dataset construction. Also, a new format for the tests was created, the tool_native format, which uses the OpenAI API function calling standard. This version is designed to be executed with lm-eval package.

This dataset does not include all the original fields. Some are modified and some are missing:

id: String ID of the sample, the original sample key.
ground_truth: Ground truth (json format) of the problem. Same as the original T-Eval dataset.
origin_prompt: A chat-style prompt (json format) for the problem. Same as the original T-Eval dataset.
tools: A tool list in OpenAI tool format (only used in tool_native variations). Extracted from the original T-Eval dataset system prompt.
response_format: A string indicating the expected response format. Obtained from the original dataset meta_data field.
evaluation_data: Evaluation data (json format) requiered to construct the evaluation steps, derived from the original dataset meta_data field.

To reproduce the dataset construction, you can use the provided script:

cd ..
git clone https://huggingface.co/datasets/lovesnowbest/T-Eval
cd T-Eval_pnyx
python generate_dataset.py --source-dataset ../T-Eval/data

(the output will be this readme metadata)

Missing Samples

This dataset only contain the english versions and (when available), only the json format versions of the samples.