58 lines
2.1 KiB
Markdown
58 lines
2.1 KiB
Markdown
|
---
|
||
|
|
base_model:
|
||
|
|
- kagelabs/KageAI-7B-v1.1
|
||
|
|
tags:
|
||
|
|
- text-generation-inference
|
||
|
|
- transformers
|
||
|
|
- unsloth
|
||
|
|
- mistral
|
||
|
|
- galore
|
||
|
|
- tech-specialist
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
datasets:
|
||
|
|
- teknium/OpenHermes-2.5
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
# KageAI-7B-v1.2 (Technical Specialist)
|
||
|
|
|
||
|
|
Developed by **KageLabs**, KageAI-7B-v1.2 is the second iteration of the KageAI series. This version marks a significant shift from general purpose chat to **Specialized Technical Intelligence**.
|
||
|
|
|
||
|
|
This model was trained using **GaLore (Gradient Low-Rank Projection)**, allowing for deeper weight updates and superior technical reasoning compared to standard LoRA fine-tuning.
|
||
|
|
|
||
|
|
## 🚀 Model Details
|
||
|
|
- **Version:** 1.2
|
||
|
|
- **Developed by:** [KageLabs](https://huggingface.co/kagelabs)
|
||
|
|
- **Base Model:** Mistral-7B-v0.3
|
||
|
|
- **Training Method:** GaLore (Rank 128)
|
||
|
|
- **Primary Focus:** Hardware Architecture, Semiconductors, and System Troubleshooting.
|
||
|
|
|
||
|
|
## 🧠 Training Focus & Knowledge Base
|
||
|
|
v1.2 has been optimized to act as a **High-End Tech Guru**. Its internal knowledge base has been refined for:
|
||
|
|
- **Semiconductors:** Deep understanding of 3nm/2nm architectures, FinFET vs. GAA, and EUV lithography.
|
||
|
|
- **Hardware Engineering:** GPU tensor cores, VRAM banking, and micro-architecture bottlenecks.
|
||
|
|
- **PC & Infrastructure:** Thermal management, overclocking logic, and custom workstation builds.
|
||
|
|
- **Instruction Following:** Enforces a **Brevity Rule** for simple queries while providing exhaustive breakdowns for complex engineering problems.
|
||
|
|
|
||
|
|
## 🧪 Version Status & Known Gaps
|
||
|
|
- **Identity:** v1.2 has a strong technical core but may occasionally show identity drift (legacy data influence).
|
||
|
|
- **Brevity:** High success rate in 1-sentence responses, but complex reasoning can sometimes be verbose.
|
||
|
|
- **v1.3 Preview:** The upcoming v1.3 update will patch identity consistency and merge this technical depth with advanced coding logic.
|
||
|
|
|
||
|
|
## 🛠 Usage
|
||
|
|
This model uses the **ChatML** prompt format:
|
||
|
|
```text
|
||
|
|
<|im_start|>system
|
||
|
|
You are KageAI by KageLabs, a High-End Technology and Hardware Specialist.<|im_end|>
|
||
|
|
<|im_start|>user
|
||
|
|
{query}<|im_end|>
|
||
|
|
<|im_start|>assistant
|
||
|
|
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
*Developed with ❤️ by KageLabs using Unsloth.*
|