Elizezen's picture
Upload 12 files
67716f9 verified
metadata
language:
  - ja
tags:
  - causal-lm
  - not-for-all-audiences
  - nsfw
pipeline_tag: text-generation

Hameln Japanese Mistral 7B

drawing

Model Description

This is a 7B-parameter decoder-only Japanese language model fine-tuned on novel datasets, built on top of the base model Japanese Stable LM Base Gamma 7B. Japanese Stable LM Instruct Gamma 7B

Usage

Ensure you are using Transformers 4.34.0 or newer.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Elizezen/Hameln-japanese-mistral-7B")
model = AutoModelForCausalLM.from_pretrained(
  "Elizezen/Hameln-japanese-mistral-7B",
  torch_dtype="auto",
)
model.eval()

if torch.cuda.is_available():
    model = model.to("cuda")

input_ids = tokenizer.encode(
    "ใ‚€ใ‹ใ—ใ‚€ใ‹ใ—ใ€ใ‚ใ‚‹ใจใ“ใ‚ใซใ€ใŠใ˜ใ„ใ•ใ‚“ใจใŠใฐใ‚ใ•ใ‚“ใŒไฝใ‚“ใงใ„ใพใ—ใŸใ€‚ ใŠใ˜ใ„ใ•ใ‚“ใฏๅฑฑใธๆŸดๅˆˆใ‚Šใซใ€",
    add_special_tokens=True, 
    return_tensors="pt"
)

tokens = model.generate(
    input_ids.to(device=model.device),
    max_new_tokens=512,
    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)

"""
output example:
ใ‚€ใ‹ใ—ใ‚€ใ‹ใ—ใ€ใ‚ใ‚‹ใจใ“ใ‚ใซใ€ใŠใ˜ใ„ใ•ใ‚“ใจใŠใฐใ‚ใ•ใ‚“ใŒไฝใ‚“ใงใ„ใพใ—ใŸใ€‚ ใŠใ˜ใ„ใ•ใ‚“ใฏๅฑฑใธๆŸดๅˆˆใ‚Šใซใ€ใŠใฐใ‚ใ•ใ‚“ใฏ็”ฐใ‚“ใผใฎ็จฒใฎๆ‰‹ไผใ„ใ‚’ใ™ใ‚‹ใชใฉใ€ไบŒไบบใงๅŠ›ใ‚’ๅˆใ‚ใ›ใฆๆฅฝใ—ใๆšฎใ‚‰ใ—ใฆใ„ใพใ—ใŸใ€‚
ใ‚ใ‚‹ๆ—ฅใฎใ“ใจใ€ใใฎๅœฐๆ–นไธ€ๅธฏใซๅคงใใชๅฐ้ขจใŒใ‚„ใฃใฆๆฅใพใ—ใŸใ€‚ๅผท้ขจใซ้ฃ›ใฐใ•ใ‚ŒใŸๆœจใ‚„ใ€ๅฎถๅฑ‹ใชใฉใŒๆฌกใ€…ใจๅ€’ใ‚Œใ‚‹ไธญใ€ๅนธใ„ใซใ‚‚ใŠใ˜ใ„ใ•ใ‚“ใจใŠใฐใ‚ใ•ใ‚“ใฎไฝใ‚“ใงใ„ใŸๆ‘ใฏ็„กไบ‹ใงใ—ใŸใ€‚
ใ—ใ‹ใ—ใ€่ฟ‘้šฃใฎๅฐใ•ใชๆ‘ใงใฏ่ขซๅฎณใŒๅ‡บใฆใ„ใพใ—ใŸใ€‚ๅฎถๅฑ‹ใฏๅ…จๅฃŠใ€่พฒไฝœ็‰ฉใฏ่’ใ‚‰ใ•ใ‚Œใ€ไฝ•ใ‚ˆใ‚Šๅคšใใฎๅ‘ฝใŒๅคฑใ‚ใ‚Œใฆใ„ใพใ—ใŸใ€‚
ใ€Œๅฏๅ“€ๆƒณใซโ€ฆโ€ฆใ€
ใŠใฐใ‚ใ•ใ‚“ใฏๅฟƒใ‚’็—›ใ‚ใ€็ฅžๆง˜ใซ็ฅˆใ‚Šใ‚’ๆงใ’็ถšใ‘ใพใ—ใŸใ€‚
ใ€ŒๅคฉไธŠใฎ็ฅžๆง˜๏ผใฉใ†ใ‹ใ€็ง้”ไบบ้–“ใ‚’ๅฎˆใฃใฆไธ‹ใ•ใ„๏ผใ€
ใŠใฐใ‚ใ•ใ‚“ใฎ็ฅˆใ‚ŠใŒ้€šใ˜ใŸใฎใ‹ใ€ๅฐ้ขจใฏๆ€ฅ้€Ÿใซๅ‹ขๅŠ›ใ‚’่ฝใจใ—ใ€่ขซๅฎณใฏๆœ€ๅฐ้™ใฎๅ†…ใซๆฒปใพใ‚Šใพใ—ใŸใ€‚
"""

Datasets

  • less than 1GB of web novels(non-PG)
  • 70GB of web novels(PG)

Intended Use

The primary purpose of this language model is to assist in generating novels. While it can handle various prompts, it may not excel in providing instruction-based responses. Note that the model's responses are not censored, and occasionally sensitive content may be generated.