README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: string
    - name: desc
      dtype: string
    - name: testbench
      list:
        - name: content
          dtype: string
        - name: name
          dtype: string
    - name: verified
      dtype: string
  splits:
    - name: test
      num_bytes: 267825
      num_examples: 50
  download_size: 113144
  dataset_size: 267825
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*

RTLLM V2.0

Huggingface Datasets format of RTLLM 2.0.

I'm not the author of this work; please refer to original repository to get more info.

Columns:

• desc: Design description;
• testbench: All testbench-related files, including testbench.v, makefile, etc.;
  • name: File name;
  • content: File content;
• verified: Verified implementation; The verified_ prefix is removed.

Example usage:

dataset = datasets.load_dataset("observerw/RTLLM", split="test")
for item in dataset:
    with tempfile.TemporaryDirectory() as tempdir:
        generation = LLM.generate(item["desc"])
        (Path(tempdir) / "impl.v").write_text(generation)
        for tb in item["testbench"]: # write all testbench-related files into temp dir
            (Path(tempdir) / tb["name"]).write_text(tb["content"])

        subprocess.run(["make"], cwd=tempdir)