Fine-Tuning Examples

1. Llama Fine-Tuning

This example demonstrates how to fine-tune Llama 2 on a Q&A dataset.

---

1.1 Problem Configuration

• Name: Llama Finetuning
• Problem Type: text_causal_classification_modeling
• Model Source: HF
• Model Name: meta-llama/Llama-2-7b-hf
• Secrets Blueprint: HF Meta (requires a token with access to meta-llama/Llama-2-7b-hf)

---

1.2 Dataset

• Train Data: finetune_train.csv
• Validation Data: finetune_validation.csv
• Answer Column: output
• Prompt Column: instruction
• Validation Size: 0.01

---

1.3 Output Storage

• Model Storage: HF
• Model Name: llama-finetuned
• Store Only LoRA Adapters: true
• Secrets Blueprint: Write Token (token with write access to HF)

---

1.4 Run Configuration

• Run Title: Run 01
• Resources:
  • Accelerator: A10G
  • GPU Count: 1
  • Memory: 64
• Tracking:
  • Experiment Name: Llama Finetuning
  • API Key: Generate from User > API Keys
  • Tracking Mode: after_epoch

---

Next Steps

• Fine-Tuning UI – Create your own fine-tuning problems and runs.
• Inference UI – Test your newly fine-tuned model in real-time.
• Benchmarks UI – Compare performance against built-in or custom benchmarks.