README.md

metadata

library_name: transformers
tags:
  - llm-jp-3-13b
  - llm
  - jp
  - 13b
language:
  - ja
base_model:
  - llm-jp/llm-jp-3-13b
pipeline_tag: question-answering
datasets: i
license: apache-2.0

Model Card for Model ID

Model Details

Uploaded model
Developed by: penguintrainer
License: apache-2.0 cc-by-sa-4.0
Finetuned from model : llm-jp/llm-jp-3-13b

Used ichikara-instruction-003-001-1 for fineturning.

ichikara-instruction: 日本語instructionモデル評価データセット © 2023 Akira Sasaki and Masato Hirakawa and Shintaro Horie and Tomoaki Nakamura (CC BY-SA 4.0 )

Uses

from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    BitsAndBytesConfig,
)
from peft import PeftModel
import torch
from tqdm import tqdm

model_id = "llm-jp/llm-jp-3-13b"
adapter_id = "penguintrainer/llm-jp-3-13b-finetune"

# QLoRA config
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)

# Load model
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto",
    )

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

# combain LoRA。
model = PeftModel.from_pretrained(model, adapter_id)


text = "大規模言語モデルとは何ですか？"

tokenized_input = tokenizer.encode(text, add_special_tokens=False, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        tokenized_input,
        max_new_tokens=100,
        do_sample=True,
        top_p=0.95,
        temperature=0.7,
        repetition_penalty=1.05,
    )[0]
print(tokenizer.decode(output))