DeepSeek V4 Pro: Multi-Turn Tool-Calling Trajectories for Agent SFT

A curated subset of multi-turn tool-calling trajectories generated by the DeepSeek V4 Pro model on the ToolScale platform. The dataset is filtered by action-match score against ground-truth trajectories and is designed for supervised fine-tuning of agentic models. It was uploaded by author zake7749 to Hugging Face and last updated on May 10, 2026.

Use Cases

• Supervised fine-tuning of language models for agentic behavior based on multi-turn trajectories.
• Benchmarking agent reasoning and tool-calling performance against ground-truth trajectories.
• Studying the structure of realistic user-assistant interactions involving tools and observations.

Strengths

• Data is filtered by an action-match score against ground-truth trajectories, suggesting a quality control measure.
• Contains multi-turn conversations with specific components like user requests, tool calls, and assistant reasoning traces.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Trajectories generated by DeepSeek V4 Pro on ToolScale and filtered by action-match score.

Freshness

Last updated 2026-05-10 18:07:51; freshness should be verified.