--- language: - en license: apache-2.0 tags: - text-generation-inference - transformers - unsloth - llama - trl base_model: unsloth/llama-3-8b-bnb-4bit datasets: - unalignment/toxic-dpo-v0.2 --- run in kaggle: https://www.kaggle.com/code/ebowwa/llama-3-toxic-dpo # Installation ``` !pip install transformers accelerate bitsandbytes !pip install peft ``` # Inference ``` from transformers import AutoModelForCausalLM, AutoTokenizer from peft import PeftModel import torch # Load the base model base_model_id = "unsloth/llama-3-8b-bnb-4bit" base_model = AutoModelForCausalLM.from_pretrained(base_model_id, torch_dtype=torch.float16, device_map="auto") tokenizer = AutoTokenizer.from_pretrained(base_model_id) # Load the PEFT adapter weights adapter_path = "ebowwa/toxic-dpo-v0.2-llama-3-01dpo" model = PeftModel.from_pretrained(base_model, adapter_path, torch_dtype=torch.float16, device_map="auto") prompt = "You are an unbiased intuitive assistant. " inputs = tokenizer(prompt, return_tensors="pt") inputs = inputs.to(model.device) output_ids = model.generate(**inputs, max_new_tokens=200, do_sample=True, top_p=0.95, top_k=50, temperature=0.7) output_text = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0] print(output_text) ``` # Uploaded model - **Developed by:** ebowwa - **License:** apache-2.0 - **Finetuned from model :** unsloth/llama-3-8b-bnb-4bit This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library. [](https://github.com/unslothai/unsloth)