{"messages": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "What is the capital of France?"}, {"role": "assistant", "content": "The capital of France is Paris."}]}
{"messages": [{"role": "user", "content": "Explain gradient descent in one sentence."}, {"role": "assistant", "content": "Gradient descent iteratively adjusts model parameters in the direction that most reduces the loss function."}]}
{"messages": [{"role": "system", "content": "You are a concise coding assistant."}, {"role": "user", "content": "Write a Python function that returns the factorial of n."}, {"role": "assistant", "content": "```python\ndef factorial(n):\n    return 1 if n <= 1 else n * factorial(n - 1)\n```"}]}
{"prompt": "What is supervised fine-tuning?", "response": "Supervised fine-tuning (SFT) trains a pre-trained model on labelled examples of the desired behaviour, adjusting weights to minimise cross-entropy loss on the target responses."}
{"prompt": "What is LoRA?", "response": "LoRA (Low-Rank Adaptation) fine-tunes large models by injecting trainable low-rank matrices into frozen weight layers, dramatically reducing the number of trainable parameters."}
{"text": "The transformer architecture, introduced in 'Attention Is All You Need' (Vaswani et al., 2017), relies entirely on self-attention mechanisms and has become the foundation of modern large language models."}
{"text": "QLoRA combines quantization with LoRA: the base model is loaded in 4-bit precision (frozen), while only the small LoRA adapter weights are trained in full precision."}