szurkemarha-mistral / README.md
boapps's picture
Update README.md
5eac576 verified
|
raw
history blame
2.01 kB
metadata
license: apache-2.0
datasets:
  - boapps/alpaca-hu
  - mlabonne/alpagasus
language:
  - hu
library_name: transformers
pipeline_tag: text-generation

szürkemarha-mistral v1

Ez az első (teszt) verziója egy magyar nyelvű instrukciókövető modellnek.

Használat

Ebben a repoban van egy app.py script, ami egy gradio felületet csinál a kényelmesebb használathoz.

Vagy kódból valahogy így:

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, GenerationConfig

tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")

BASE_MODEL = "mistralai/Mistral-7B-v0.1"
LORA_WEIGHTS = "boapps/szurkemarha-mistral"

device = "cuda"

try:
    if torch.backends.mps.is_available():
        device = "mps"
except:
    pass

nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.bfloat16
)

model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, quantization_config=nf4_config)

model = PeftModel.from_pretrained(
    model, LORA_WEIGHTS, torch_dtype=torch.float16, force_download=True
)

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:
Melyik megyében található az alábbi város?

### Input:
Pécs

### Response:"""
inputs = tokenizer(prompt, return_tensors="pt")
input_ids = inputs["input_ids"].to(device)
generation_config = GenerationConfig(
    temperature=0.1,
    top_p=0.75,
    top_k=40,
    num_beams=4,
)
with torch.no_grad():
    generation_output = model.generate(
        input_ids=input_ids,
        generation_config=generation_config,
        return_dict_in_generate=True,
        output_scores=True,
        max_new_tokens=256,
    )
s = generation_output.sequences[0]
output = tokenizer.decode(s)
print(output.split("### Response:")[1].strip())