How to use ・ 使い方
We recommend on running with at least 4 A100 cards A100の4枚の環境がおすすめです
Huggingface
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
tokenizer = AutoTokenizer.from_pretrained("lightblue/ao-karasu-72B")
model = AutoModelForCausalLM.from_pretrained("lightblue/ao-karasu-72B", device_map="auto")
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
messages = [{"role": "system", "content": "あなたはAIアシスタントです。"}]
messages.append({"role": "user", "content": "イギリスの首相は誰ですか?"})
prompt = tokenizer.apply_chat_template(conversation=messages, add_generation_prompt=True, tokenize=False)
pipe(prompt, max_new_tokens=100, do_sample=False, temperature=0.0, return_full_text=False)
vLLM
from vllm import LLM, SamplingParams
sampling_params = SamplingParams(temperature=0.0, max_tokens=100)
llm = LLM(model="lightblue/aokarasu-72B", tensor_parallel_size=4)
messages = [{"role": "system", "content": "あなたはAIアシスタントです。"}]
messages.append({"role": "user", "content": "イギリスの首相は誰ですか?"})
prompt = llm.llm_engine.tokenizer.tokenizer.apply_chat_template(conversation=messages, add_generation_prompt=True, tokenize=False)
prompts = [prompt]
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
Training details 学習詳細
Training data 学習データ
Roughly 20 million characters samples from a dataset of more than 1.1 billion characters, which was made up of:
~450 million characters from Wikipedia-based QA (same as Qarasu)
~200 million characters from technical blogs (new)
~200 million characters from Japanese QA site answers (new)
~100 million characters from LLM generated prompts and responses (same as Qarasu)
~70 million characters from news articles (new)
Training schedule
Training for ~1 day on a A100 (80GB) GPU
- Downloads last month
- 36
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.