DeepSeek V4 Pro Agent: Raw Traces for Supervised Fine-Tuning

TeichAI's dataset contains raw agent trace files generated using the teich tool. The dataset includes 4,006 JSONL files where all assistant responses were generated by the deepseek/deepseek-v4-pro model. It was last updated on May 12, 2026.

Use Cases

• Fine-tuning language models for agentic behavior based on the provided raw trace files.
• Studying the response patterns and reasoning traces of the DeepSeek V4 Pro model.
• Preparing training data for tool-use or multi-step reasoning agents using the embedded tools schema.

Strengths

• Contains 4,006 distinct JSONL trace files.
• All assistant responses are generated by a single, specified model (deepseek/deepseek-v4-pro).
• Includes a configured tools schema snapshot for training-ready tools.

Limitations

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

Provenance

Source

TeichAI

Collection Method

Generated using the teich tool by TeichAI.

Freshness

Last updated 2026-05-12 04:27:13; freshness should be verified.