ask-offline

Steps

1. Download Ollama client [for Windows, size of EXE 700MB]

2. In command prompt, issue: ollama pull qwen2.5:1.5b or ollama pull llama-3.2:1b

3. Fine-tune:

from peft import LoraConfig, get_peft_model
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "gemma-2-2b"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

config = LoraConfig(r=8, lora_alpha=16, lora_dropout=0.1)
model = get_peft_model(model, config)