43 lines
1.4 KiB
Markdown
43 lines
1.4 KiB
Markdown
|
|
||
|
|
---
|
||
|
|
base_model: mistralai/Mistral-7B-Instruct-v0.3
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- tool-calling
|
||
|
|
- custom-finetune
|
||
|
|
- axim-alignment
|
||
|
|
datasets:
|
||
|
|
- GRRNMAKER/axim-alignment-data
|
||
|
|
---
|
||
|
|
|
||
|
|
# Model Card for GRRNMAKE/Magnus
|
||
|
|
|
||
|
|
Magnus is a fine-tuned iteration of the Mistral-7B-Instruct-v0.3 model, specifically optimized for rigorous **tool calling**, precise formatting adherence, and extreme conciseness.
|
||
|
|
|
||
|
|
## Training Details
|
||
|
|
- **Base Model**: `mistralai/Mistral-7B-Instruct-v0.3`
|
||
|
|
- **Dataset**: `GRRNMAKER/axim-alignment-data` (Axim Corrective Benchmark)
|
||
|
|
- **Training Method**: 4-bit QLoRA natively merged into the base architecture.
|
||
|
|
- **Hardware**: Lambda Instance (NVIDIA GPU)
|
||
|
|
|
||
|
|
## Evaluation Results
|
||
|
|
This model was evaluated against the rigorous Axim Corrective Benchmark targeting tool-calling performance and formatting fidelity.
|
||
|
|
|
||
|
|
| Metric | Score | Note |
|
||
|
|
|---|---|---|
|
||
|
|
| **Concise Response Accuracy** | 98.200 | Axim Alignment Dataset |
|
||
|
|
| **Formatting Adherence (ROUGE-L)** | Verified | Axim Alignment Dataset |
|
||
|
|
| **Instruction Adherence** | High | |
|
||
|
|
| **Hallucination Rate** | Low | |
|
||
|
|
| **Conciseness Score** | 94.00 | Local Verification Subset |
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
The model natively supports the standard Mistral v0.3 tool-calling chat template.
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained("GRRNMAKE/Magnus")
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained("GRRNMAKE/Magnus")
|
||
|
|
```
|