Fine-tuning framework: LoRA (Low-Rank Adaptation) with PEFT
Dataset type: Instruction-style chat pairs for Product Management scenarios
📘 Model Overview
Ken Advisor is a fine-tuned version of Microsoft’s Phi-2, trained to serve as an AI Product Management Assistant.
It provides structured, concise, and context-aware guidance across various product management tasks — from ideation and discovery to stakeholder communication and AI integration.
This model was trained using a LoRA adaptation, allowing efficient fine-tuning on consumer GPUs while preserving Phi-2’s original reasoning strength.
💡 Key Features
Responds to PM-related queries naturally and informatively
Understands user-product conversation flows
Produces actionable insights for product strategy, UX, and AI integration
Lightweight (2.7B parameters) and efficient for deployment
🧩 Training Details
Parameter
Value
Base model
microsoft/phi-2
Method
LoRA fine-tuning (PEFT)
Dataset size
~540 samples
Epochs
3
Batch size
1
Learning rate
2e-4
Max sequence length
512
Average training loss
0.2468
Framework
Hugging Face Transformers (v4.38)
🧠 Example Usage
python
1from transformers import pipeline
23pipe = pipeline("text-generation", model="your-username/ken-fine-tuned-phi2")45prompt ="User: How should I prioritize AI features in my product roadmap?\nAssistant:"6result = pipe(prompt, max_new_tokens=150)7print(result[0]["generated_text"])
⚙️ Intended Use
This model is designed for:
Product managers and startup founders
AI-driven project teams
PM education and mentorship use cases
⚠️ Limitations:
Not suitable for legal, financial, or medical advice. Outputs may include hallucinations or inaccuracies — always verify before use.
