Dorman11/datasetdb / README.md
Dorman11's picture
|
download
raw
1.49 kB
metadata
license: apache-2.0
pretty_name: NeMo Gym Conversational Tool-Use Assets

NeMo Gym Conversational Tool-Use Assets

This dataset repository stores prompt and reference assets for NeMo Gym's conversational tool-use generation pipeline. It is an asset bundle for Gym components, not a training or evaluation dataset.

Contents

  • conversational_tool_use_domain_generation/prompts: the domain-generation prompt.
  • conversational_tool_use_domain_generation/prompt_history: historical domain-generation prompt revisions.
  • conversational_tool_use_policy_tool_generation/prompts: policy, tool, refinement, and judge prompts.
  • conversational_tool_use_policy_tool_generation/prompt_history: historical policy/tool-generation prompt revisions.
  • conversational_tool_use_policy_tool_generation/references/golden_policies: golden policy Markdown and paired tool JSONL references.
  • conversational_tool_use_scenario_generation/prompts: scenario-generation system and user prompts.

The rollout simulation server and policy agent keep their prompt templates in their Python implementations. Those templates are not duplicated in this dataset. JSON schemas, runnable examples, package documentation, and Python requirements remain in Git.

Run the conversational tool-use simulation resource server's preparation command once per checkout to download and checksum-validate the runtime assets:

python -m resources_servers.conversational_tool_use_simulation.prepare

Xet Storage Details

Size:
1.49 kB
·
Xet hash:
fbccab2e71c4755f4435dd9888a5b3fb503e307df8900ad03071eeb9afd4426b

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.