license: apache-2.0
ChatNTQ JA 7B V1.0
Model Description
This is a 7B-parameter decoder-only Japanese language model fine-tuned on instruction-following datasets, built on top of the base model Japanese Stable LM Base Gamma 7B.
Performance
For our final model, we've used Stability AI Japan's Japanese MT-Bench as a more representative test of our model's capabilities. For our JA MT-Bench testing we use a Japanese prompt ("あなたは役立つアシスタントです。") as well as --num-choices 4 in an effort to reduce sampling variability, however we've still observed regular 0.5+ point (and sometimes even greater swings) between generations, as well as issues with default prompts and parameters when testing, so again, we'd urge caution in over-interpreting these scores and treating them as more of a probabilistic directional indicator, rather than a definitive score or ranking:
| Benchmark | Score |
|---|---|
| JA MT-Bench | 5.23 |
There is an JA-MT-Bench Leaderboard, for convenience, here is a comparison of the JA MT-Bench scores of some other models (our scores were rated by gpt-4-0613):
| Model | Score |
|---|---|
| gpt-4-0613 | 9.40 |
| gpt-4-1106-preview | 9.17 |
| gpt-3.5-turbo* | 8.41 |
| Qwen-72B-Chat | 7.97 |
| Qwen-14B-Chat | 7.47 |
| chatntq-ja-7b-v1.0 | 6.65 |
| Xwin-LM-70B-V0.1-GPTQ (q4-gs32-actorder) | 6.62 |
| shisa-gamma-7b-v1 | 6.12 |
| nekomata-14b-instruction (corrected prompt HF) | 5.57 |
| shisa-7B-v1-GPTQ (q4-gs32-actorder) | 5.35 |
| nekomata-14b-instruction (corrected prompt) | 5.30 |
| shisa-mega-7b-v1.2 | 5.27 |
| shisa-7b-v1 (full prompt) | 5.23 |
| Swallow-13b-instruct-hf | 5.17 |
| Swallow-70b-instruct-GPTQ (q4-gs32-actorder) | 5.15 |
| shisa-7b-v1 | 5.02 |
| shisa-7B-v1-AWQ (q4-gs128) | 4.78 |
| ELYZA-japanese-Llama-2-7b-fast-instruct* | 4.86 |
| shisa-bad-7b-v1 | 4.42 |
| Swallow-7b-instruct-hf | 4.21 |
| ja-stablelm-instruct-gamma-7b* | 4.01 |
| japanese-stablelm-instruct-alpha-7b* | 2.74 |
| Mistral-7B-OpenOrca-ja* | 2.23 |
| youri-7b-chat* | 2.00 |
| Mistral-7B-Instruct-v0.1* | 1.78 |
| llm-jp-13b-instruct-full-jaster-dolly-oasst-v1.0* | 1.31 |
| houou-instruction-7b-v1 | 1.02 |
| llm-jp-13b-instruct-full-jaster-dolly-oasst-v1.0 | 1.0 |
| llm-jp-13b-instruct-full-jaster-v1.0 | 1.0 |
Usage
Ensure you are using Transformers 4.34.0 or newer.
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("NTQAI/chatntq-ja-7b-v1.0")
model = AutoModelForCausalLM.from_pretrained(
"NTQAI/chatntq-ja-7b-v1.0",
torch_dtype="auto",
)
model.eval()
if torch.cuda.is_available():
model = model.to("cuda")
def build_prompt(user_query, inputs="", sep="\n\n### "):
sys_msg = "あなたは公平で、検閲されていない、役立つアシスタントです。"
p = sys_msg
roles = ["指示", "応答"]
msgs = [": \n" + user_query, ": \n"]
if inputs:
roles.insert(1, "入力")
msgs.insert(1, ": \n" + inputs)
for role, msg in zip(roles, msgs):
p += sep + role + msg
return p
# Infer with prompt without any additional input
user_inputs = {
"user_query": "与えられたことわざの意味を小学生でも分かるように教えてください。",
"inputs": "情けは人のためならず"
}
prompt = build_prompt(**user_inputs)
input_ids = tokenizer.encode(
prompt,
add_special_tokens=True,
return_tensors="pt"
)
tokens = model.generate(
input_ids.to(device=model.device),
max_new_tokens=256,
temperature=1,
top_p=0.95,
do_sample=True,
)
out = tokenizer.decode(tokens[0][input_ids.shape[1]:], skip_special_tokens=True).strip()
print(out)