---
license: cc-by-4.0
task_categories:
- question-answering
tags:
- code
size_categories:
- n<1K

dataset_info:
  - config_name: stack_effect
    features:
      - name: code
        dtype: string
      - name: stack_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string
      - name: flag_states
        struct:
          - name: FLAG_0
            dtype: bool
          - name: FLAG_1
            dtype: bool
          - name: FLAG_2
            dtype: bool
          - name: FLAG_3
            dtype: bool
          - name: FLAG_4
            dtype: bool
      - name: return_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string

  - config_name: flag_reach
    features:
      - name: code
        dtype: string
      - name: stack_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string
      - name: flag_states
        struct:
          - name: FLAG_0
            dtype: bool
          - name: FLAG_1
            dtype: bool
          - name: FLAG_2
            dtype: bool
          - name: FLAG_3
            dtype: bool
          - name: FLAG_4
            dtype: bool
      - name: return_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string

  - config_name: result_estimation
    features:
      - name: code
        dtype: string
      - name: stack_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string
      - name: flag_states
        struct:
          - name: FLAG_0
            dtype: bool
          - name: FLAG_1
            dtype: bool
          - name: FLAG_2
            dtype: bool
          - name: FLAG_3
            dtype: bool
          - name: FLAG_4
            dtype: bool
      - name: return_values
        list:
          struct:
            - name: type
              dtype: string
            - name: value
              dtype: string

configs:
- config_name: stack_effect
  default: true
  data_files:
    - split: eval
      path: dataset_stack/data.jsonl

- config_name: flag_reach
  data_files:
    - split: eval
      path: dataset_flags/data.jsonl

- config_name: result_estimation
  data_files:
    - split: eval
      path: dataset_results/data.jsonl
---
### Exemplary baseline datasets for testing LLM runtime reasoning on WebAssembly, automatically generated with WasmWeaver.

All entries share common metadata; only the field required by the task (`stack_values`, `flag_states`, or `return_values`) is populated.

1. dataset\_stack – 10 000 samples<br>
   Predict the runtime stack at each `;;INSPECT` marker.

2. dataset\_flags – 10 000 samples<br>
   Identify which flags (`;;FLAG_1 … ;;FLAG_N`) are reachable.

3. dataset\_result – 1 000 samples<br>
   Predict the program’s final return values.