AmirhoseinGH/mhlc-training-qwen3vl-qwen3_vl_4b_instruct_hard_mixed_sources_120k

Multi Head Latent Control Training Data - Qwen3-VL 4B Instruct hard Mixed Sources 120k

Dataset Description

This repository contains verified training data for the Multi Head Latent Control paper release. It is part of the Multi Head Latent Control training data Hugging Face collection.

Paper

https://arxiv.org/abs/2607.14277

Code

https://github.com/Amirhosein-gh98/Multi-Head-Latent-Control

Dataset Summary

Data Files

The release data is stored under:

data/

Metadata files are stored under:

• —metadata/selection_manifest.json
• —metadata/verification_stats.json
• —metadata/generation_stats.json

Data Fields

The examples are stored in the same schema as the local release artifacts. For parquet shards, inspect the schema with:

from datasets import load_dataset

ds = load_dataset("AmirhoseinGH/mhlc-training-qwen3vl-qwen3_vl_4b_instruct_hard_mixed_sources_120k", data_dir="data", split="train")
print(ds)
print(ds.features)

Intended Use

These files are intended for reproducing and extending the Multi Head Latent Control training pipeline. The raw, unverified folders are excluded from this release by default.

Notes

• —

Citation

@misc{ghasemabadi2026multiheadlatentcontrolunified,
  title={Multi-Head Latent Control: A Unified Interface for LLM Agent Decision Making},
  author={Amirhosein Ghasemabadi and Ruichen Chen and Bahador Rashidi and Di Niu},
  year={2026},
  eprint={2607.14277},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2607.14277}
}