Mistral-7b-v2-hv2

Model Card for mistral-7b-hlx2

Model Summary

mistral-7b-hlx2 is a supervised fine-tuned (SFT) version of Mistral-7B-Instruct-v0.2, trained on cleaned, de-identified WhatsApp conversation logs (dataset: hailsbop/Haleyv2).
The model is optimized for interpersonal dialogue, conversational tone, and expressive communication.

Table of Contents

• Model Details
• Uses
• Bias, Risks, and Limitations
• Training Details
• Evaluation
• Societal Impact Assessment
• Model Examination
• Environmental Impact
• Technical Specifications
• Citation
• Glossary
• More Information
• Model Card Authors
• Model Card Contact
• How to Get Started

Model Details

Model Description

mistral-7b-hlx2 is a decoder-only transformer model fine-tuned via SFT using WhatsApp-derived conversational data.

• Developed by: Haley
• License: Apache-2.0
• Fine-tuned From: mistralai/Mistral-7B-Instruct-v0.2
• Dataset: hailsbop/Haleyv2

Uses

Direct Use

Conversational agents, expressive dialogue, creative writing.

Downstream Use

Persona-aligned chatbots, stylistic generation.

Out-of-Scope

Safety-critical use, factual decision-making, impersonation.

Bias, Risks, and Limitations

This model reflects interpersonal tone, subjectivity, and informal WhatsApp communication patterns. Not suited for factual accuracy.

Training Details

Training Data

Dataset: hailsbop/Haleyv2 — de-identified WhatsApp logs.

Preprocessing

PII removed, paired into instruction–response sets.

Training Procedure

Full SFT, mixed precision, AdamW optimizer.

Evaluation

Qualitative testing only — no benchmarks.

Environmental Impact

Training CO₂ impact not recorded.

Technical Specifications

• 7B parameter decoder LM
• FlashAttention-like optimizations
• Causal LM objective

Citation

@misc{mistral7bhlx2,
  title={mistral-7b-hlx2},
  author={Haley},
  year={2025},
  howpublished={https://huggingface.co/hailsbop/mistral-7b-hlx2}
}

Model Card Authors

Haley