metadata
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, 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
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