LLM basado en LLaMA Ajustado al Dominio de Patología

Primera Versión de un LLM ajustado para responder preguntas de Patología

Uploaded model

• Developed by: jjsprockel
• License: apache-2.0
• Finetuned from model : unsloth/llama-3-8b-bnb-4bit

Código para descarga: El siguiente es el código sugerido para descargar el modelo usando Unslot:

import torch
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "jjsprockel/Patologia_lora_model1",
    max_seq_length = 2048, # Choose any! Llama 3 is up to 8k
    dtype = None,
    load_in_4bit = True,
    )

FastLanguageModel.for_inference(model)

alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

Código para la inferencia:

El siguiente codigo demuestra como se puede llevar a cabo la inferencia.

instruction = input("Ingresa la pregunta que tengas de Patología: ")

inputs = tokenizer(
[
    alpaca_prompt.format(
        instruction, # instruction
        "", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 2048)

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.