Trim Gpt 5.4 Mini Separate Neighbors Only 4Nbr

TRIM Agent Reasoning Messages is a Hugging Face export of supervised fine-tuning data for language models. The dataset contains 11,464 records from the 'train' split, generated by the OpenAI GPT-5.4-mini model via OpenRouter. It was uploaded by Kiria-Nozan and last updated on April 28, 2026.

Use Cases

• Fine-tuning language models for agent reasoning based on the described SFT data.
• Training models on biochemistry prediction tasks like AMES or Bioavailability based on the task list.
• Studying the 'local_neighbor_only' SFT mode's effect on model performance.
• Benchmarking model outputs against reasoning traces from the GPT-5.4-mini model.

Strengths

• Contains 11,464 records for model training.
• Focuses on agent reasoning across multiple specific tasks, including several biochemistry predictions.
• Data is sourced from a specific model version (OpenAI GPT-5.4-mini) via a known provider (OpenRouter).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Only the 'train' split is present, lacking validation or test splits for standard evaluation.

Provenance

Source

OpenRouter, using the OpenAI GPT-5.4-mini model.

Collection Method

Supervised fine-tuning data generated in 'local_neighbor_only' mode.

Time Range

null

Freshness

Last updated 2026-04-28 00:50:03; freshness should be verified.

Geography

null