bfcl-ms / README.md
khursani8's picture
Upload README.md with huggingface_hub
3a57d73 verified
|
Raw
History Blame Contribute Delete
4.68 kB
metadata
language:
  - ms
  - en
license: apache-2.0
task_categories:
  - text-generation
tags:
  - malay
  - bahasa-malaysia
  - function-calling
  - benchmark
  - bfcl
  - evaluation
size_categories: 1K<n<10K
configs:
  - config_name: simple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_simple.json
  - config_name: multiple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multiple.json
  - config_name: parallel
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_parallel.json
  - config_name: parallel_multiple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_parallel_multiple.json
  - config_name: irrelevance
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_irrelevance.json
  - config_name: java
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_java.json
  - config_name: javascript
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_javascript.json
  - config_name: rest
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_rest.json
  - config_name: sql
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_sql.json
  - config_name: live_simple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_simple.json
  - config_name: live_multiple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_multiple.json
  - config_name: live_parallel
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_parallel.json
  - config_name: live_parallel_multiple
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_parallel_multiple.json
  - config_name: live_relevance
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_relevance.json
  - config_name: live_irrelevance
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_live_irrelevance.json
  - config_name: chatable
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_chatable.json
  - config_name: multi_turn_base
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multi_turn_base.json
  - config_name: multi_turn_composite
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multi_turn_composite.json
  - config_name: multi_turn_long_context
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multi_turn_long_context.json
  - config_name: multi_turn_miss_func
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multi_turn_miss_func.json
  - config_name: multi_turn_miss_param
    data_files:
      - split: test
        path: BFCL_V3/BFCL_v3_multi_turn_miss_param.json

BFCL v3 — Malay (Bahasa Malaysia) Edition

A Malay edition of the Berkeley Function-Calling Leaderboard (BFCL) v3 — the standard benchmark for evaluating LLM function/tool calling.

5,011 test entries across 21 categories (simple, multiple, parallel, parallel_multiple, irrelevance, java, javascript, rest, sql, live_*, chatable, multi_turn_*), with ground-truth answer files unchanged so the official BFCL scoring harness runs unmodified.

What is translated to Malay:

  • User questions / prompts
  • Function descriptions
  • Parameter descriptions

What is unchanged (by design, for scoring compatibility):

  • Function and parameter names, types, enums, required lists
  • Ground truth (possible_answer/) — byte-identical to the original
  • Multi-turn environment state (initial_config), execution paths, ids
  • A small number of entries (23, mostly short live-irrelevance prompts) are kept in English
Coverage Count
Entries with Malay questions 4,988 / 5,011 (99.5%)
Function descriptions translated 8,896 / 8,896
Parameter descriptions translated 32,093 / 32,101

Usage

Evaluate a chat model by pointing the BFCL harness at these entry files in place of the originals: prompts are Malay, expected calls are unchanged, so AST/exe validation and irrelevance detection measure Malay-prompt function calling directly.

Attribution

Derived from gorilla-llm/Berkeley-Function-Calling-Leaderboard. Please cite the original work:

@article{patil2023gorilla,
  title={Gorilla: Large Language Model Connected with Massive APIs},
  author={Patil, Shishir G and Zhang, Tianjun and Wang, Xin and Gonzalez, Joseph E},
  journal={arXiv preprint arXiv:2305.15334},
  year={2023}
}

License

Apache License 2.0, inherited from the original dataset.