GrepSeek: Cold-Start Supervised Fine-Tuning Data for Direct Corpus Interaction Agents

GrepSeek is a dataset of 10,000 trajectories for cold-start supervised fine-tuning of a Direct Corpus Interaction search agent. The agent answers questions by issuing Unix shell commands like grep and rg directly against a raw text corpus, coupling retrieval and reasoning. The dataset was created by alireza7 and uploaded to Hugging Face on 2026-05-29.

Use Cases

• Training a search agent's initial policy based on described command trajectories.
• Studying the coupling of retrieval and reasoning in a single policy as described.
• Benchmarking Direct Corpus Interaction agents against traditional retrieval methods.
• Fine-tuning language models to generate shell commands for information seeking.

Strengths

• Contains 10,000 supervised fine-tuning trajectories for agent initialization.
• Designed for a specific, novel agent architecture (Direct Corpus Interaction).
• Associated with a research paper, suggesting a defined methodology.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update date is in the future (2026-05-29).

Provenance

Source

Hugging Face user alireza7.

Collection Method

Likely generated for the GrepSeek research project, involving trajectories of shell commands issued against a text corpus.

Freshness

Last updated 2026-05-29 01:50:25.