Skip to content

demo_script.md

Latest commit

 

History

History
52 lines (42 loc) · 2.47 KB

demo_script.md

File metadata and controls

52 lines (42 loc) · 2.47 KB

FineTuneLite Demo Script

Duration: 3-5 minutes Goal: Demonstrate the end-to-end flow of fine-tuning a small LLM on CPU.

Setup (Before Demo)

  1. Ensure LM Studio is running with a model loaded (port 1234).
  2. Ensure Backend is running (python -m uvicorn main:app --reload).
  3. Ensure Frontend is running (npm run dev).
  4. Have a small sample dataset ready (e.g., sample_data.json with 10-20 examples).

Walkthrough

1. Introduction (30s)

  • Say: "This is FineTuneLite, a local studio for fine-tuning LLMs on CPU."
  • Show: Dashboard page. Point out the "System Status" (CPU/RAM usage).

2. Dataset Upload (1 min)

  • Action: Navigate to Datasets.
  • Action: Click "Upload Dataset". Select sample_data.json.
  • Show: The dataset appears in the list with row count and size.
  • Say: "We support JSON and CSV. The system automatically parses and validates the data."

3. Fine-tuning Configuration (1 min)

  • Action: Navigate to Fine-tune.
  • Step 1: Select ibm-granite-4.0-h-tiny as the base model.
  • Step 2: Select the uploaded dataset. Set Epochs=1, Batch Size=1 (for speed).
  • Step 3: Review settings and click "Start Training".
  • Say: "We use LoRA (Low-Rank Adaptation) to make training efficient enough for a laptop CPU."

4. Training Job Monitoring (1 min)

  • Action: You are redirected to Training Jobs.
  • Show: The job status changes to "Running".
  • Show: The loss graph (if implemented) or status updates.
  • Say: "The backend runs the training loop in a background thread, logging loss metrics to SQLite."

5. Chat / Inference (1 min)

  • Action: Navigate to Playground.
  • Action: Select a model from the dropdown (defaults to IBM Granite 4.0 H Tiny).
  • Action: Type "Hello, who are you?".
  • Show: The model responds.
  • Say: "Inference is handled via LM Studio, ensuring fast local responses with any model you have loaded."

6. Teacher/Critic Mode (1 min)

  • Action: Toggle "Teacher/Critic Mode" ON.
  • Action: Select a student model (e.g., smaller model) and teacher model (e.g., Granite).
  • Action: Send a message and click "Ask Teacher for Feedback".
  • Show: The teacher model provides critique and improved answer.
  • Say: "This feature lets you use a stronger model to evaluate and improve responses from smaller models."

Conclusion

  • "FineTuneLite makes custom LLMs accessible to anyone with a standard laptop, preserving privacy and eliminating cloud costs."