|
|
--- |
|
|
license: mit |
|
|
datasets: |
|
|
- vitalune/business-assistant-ai-tools |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- mistralai/Mistral-7B-v0.1 |
|
|
tags: |
|
|
- business |
|
|
- startup |
|
|
- ai_tools |
|
|
--- |
|
|
|
|
|
# Ovarra-v1 — Business Planning Assistant (Fine-tuned Mistral 7B) |
|
|
|
|
|
**Ovarra-v1** is a fine-tuned version of [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1), designed to help users plan, clarify, and launch AI-powered business ideas. |
|
|
|
|
|
This model acts like a **startup co-pilot**, guiding users from rough ideas to concrete action steps — with AI tools, strategy, and marketing guidance embedded in every response. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📌 Model Details |
|
|
|
|
|
| Attribute | Value | |
|
|
|------------------|-----------------------------------------| |
|
|
| Base model | `mistralai/Mistral-7B-v0.1` | |
|
|
| Finetuned on | ~1500 GPT-4-generated prompt-response pairs, validated by human reviews of the best AI tools for each sector. | |
|
|
| Format | Instruction-tuned with `[INST] ... [/INST]` | |
|
|
| Finetuning method| QLoRA (4-bit), Flash Attention 2 | |
|
|
| Trainer | Axolotl + RunPod (A40 GPU) | |
|
|
|
|
|
--- |
|
|
|
|
|
## Training Objective |
|
|
|
|
|
The model was trained to: |
|
|
- **Ask clarifying questions** when input is vague |
|
|
- **Plan structured roadmaps** (4–6 steps) |
|
|
- **Recommend relevant AI tools** for each stage |
|
|
- **Support follow-up questions** with contextual awareness |
|
|
|
|
|
Training categories included: |
|
|
- AI tool integration |
|
|
- Product planning |
|
|
- MVP prototyping |
|
|
- Marketing strategy |
|
|
- Startup operations |
|
|
|
|
|
--- |
|
|
|
|
|
## Example Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained("vitalune/ovarra-v1", device_map="auto") |
|
|
tokenizer = AutoTokenizer.from_pretrained("vitalune/ovarra-v1") |
|
|
|
|
|
prompt = "[INST] I want to launch an AI copywriting SaaS. Help me plan it. [/INST]" |
|
|
inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
|
|
|
output = model.generate(**inputs, max_new_tokens=512) |
|
|
print(tokenizer.decode(output[0], skip_special_tokens=True)) |
|
|
``` |
|
|
|
|
|
## Credits |
|
|
|
|
|
- Fine-tuned and built by @vitalune |
|
|
- Based on prompt logic and data generation co-created using GPT-4 API |
|
|
- Hosted and accelerated via RunPod |
