⚠️ WARNING: BROKEN MODEL ⚠️

Model Checkpoint Advisory

🔴 STATUS: BROKEN 🚫 NOT FOR PRODUCTION 🔬 RESEARCH USE ONLY

This model checkpoint is currently broken and is not production-ready. It has not passed the validation, stability, or safety checks required for deployment, and must not be used to serve live traffic or power user-facing features.

⚠️ Known Issues

• Hallucinations — may generate plausible-sounding but false or fabricated information
• Repetitive outputs — may loop, stall, or repeat phrases and content
• Incorrect information — factual accuracy is not reliable
• Unreliable instruction following — may ignore, misread, or partially follow prompts

🔬 Recommended Use

Use this checkpoint only for internal testing, research, and debugging. Do not integrate it into production systems, customer-facing applications, or any environment where correctness or reliability is required.

This notice must remain visible until the checkpoint is fixed, replaced, or formally deprecated.

Nero-800M

A lightweight instruction-tuned language model optimized for efficient local AI.

---

Nero-800M

Nero-800M is a compact open-source language model based on Qwen3.5-0.8B, fine-tuned to improve instruction following, conversational ability, and reasoning behavior while maintaining fast and efficient inference.

Nero-800M is designed for users who want a capable AI assistant that can run locally on limited hardware.

Model Details

Detail	Value
Base Model	unsloth/Qwen3.5-0.8B
Parameters	~800M
Fine-tuning	QLoRA + LoRA SFT
Context Length	2048 tokens
Architecture	Decoder-only Transformer

Training

Nero-800M was trained using parameter-efficient fine-tuning methods:

• Supervised Fine-Tuning (SFT)
• QLoRA
• LoRA refinement

Training focused on:

• Instruction following
• Helpful assistant behavior
• Response formatting
• Reasoning improvements

Intended Use

Nero-800M is suitable for:

• Local AI assistants
• Edge devices
• Lightweight chat applications
• AI experimentation
• Educational projects

Limitations

Due to its compact size, Nero-800M may:

• Have weaker reasoning than larger models
• Make factual mistakes
• Struggle with complex multi-step tasks

For more demanding workloads, larger Nero models are recommended.

Credits

Built with:

• Qwen3.5
• Unsloth
• Open-source AI community