SWE Next SFT Trajectories: 3,693 Software Engineering Training Conversations

SWE_Next_SFT_Trajectories.jsonl is a ShareGPT-style JSONL file containing 3,693 training examples for conversational AI. The dataset, created by TIGER-Lab, was last updated on April 2, 2026. Each example is a JSON object with a messages field containing conversation turns with roles such as system, user, assistant, and tool.

Use Cases

• Fine-tuning language models for software engineering assistance based on the conversational trajectories.
• Training conversational agents to handle tool calls based on the inclusion of the 'tool' role in messages.
• Studying instruction-following patterns in code-related tasks based on the structured user-assistant dialogues.
• Benchmarking model performance on software engineering problem-solving based on the provided training examples.

Strengths

• Contains 3,693 specific training examples for supervised fine-tuning.
• Provides structured ShareGPT-style JSONL format for direct use in training pipelines.
• Includes multi-role conversational data with system, user, assistant, and tool turns.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but the specific content and diversity of the 3,693 examples are not detailed.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

TIGER-Lab

Collection Method

Likely gathered or synthesized for training language models on software engineering tasks.

Time Range

null

Freshness

Last updated 2026-04-02 04:40:56; freshness should be verified.

Geography

null