Fine-Tune Llama 3.1 8B
Budget: $10 – $30 USD
I have a cleaned, labelled set of WhatsApp conversations and I need a custom checkpoint of Meta’s Llama 3.1 8B (running through Ollama) that can read any new chat line and return structured insights straight away. After training, the model should spot the key words, infer the user’s intent and surface any basic statistics we define, then hand those results back as a simple JSON payload.
The dataset is already pre-processed, so you can jump directly into experimentation. I’m comfortable with PyTorch, LoRA or QLoRA adapters, and the Ollama tooling; choose whichever stack gets us the best balance of accuracy and VRAM use.
What I need from you:
• Full fine-tuned weights (or adapter) ready to be loaded by “ollama run”
We can agree on evaluation metrics together, but the model must outperform the base checkpoint on a held-out slice of my data before sign-off. If you have prior examples of chat-domain tuning, feel free to share them when you reply—otherwise let’s get started.
The dataset is already pre-processed, so you can jump directly into experimentation. I’m comfortable with PyTorch, LoRA or QLoRA adapters, and the Ollama tooling; choose whichever stack gets us the best balance of accuracy and VRAM use.
What I need from you:
• Full fine-tuned weights (or adapter) ready to be loaded by “ollama run”
We can agree on evaluation metrics together, but the model must outperform the base checkpoint on a held-out slice of my data before sign-off. If you have prior examples of chat-domain tuning, feel free to share them when you reply—otherwise let’s get started.