Stockmark 2 100B Instruct Beta

🚀 Stockmark-2-100B-Instruct-beta

Stockmark-2-100B-Instruct-beta is a 100-billion-parameter large language model. It's built from scratch and focuses on Japanese. It offers high - quality language processing capabilities, especially in Japanese and English, with a certain proportion of code support.

🚀 Quick Start

To quickly start using Stockmark-2-100B-Instruct-beta, you can refer to the following code example:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("stockmark/Stockmark-2-100B-Instruct-beta")
model = AutoModelForCausalLM.from_pretrained(
    "stockmark/Stockmark-2-100B-Instruct-beta", device_map="auto", torch_dtype=torch.bfloat16
)

instruction = "自然言語処理とは？"
input_ids = tokenizer.apply_chat_template(
    [{"role": "user", "content": instruction}], add_generation_prompt=True, return_tensors="pt"
).to(model.device)

with torch.inference_mode():
    tokens = model.generate(
        input_ids,
        max_new_tokens = 512,
        do_sample = True,
        temperature = 0.7,
        top_p = 0.95,
        repetition_penalty = 1.05
    )
    
output = tokenizer.decode(tokens[0], skip_special_tokens=True)
print(output)

✨ Features

• Multilingual Focus: Pre - trained on data with 60% English, 30% Japanese, and 10% code, providing strong multilingual processing capabilities.
• Instruction - Following Ability: After post - training with synthetic Japanese data generated by Qwen2.5 - 32B - Instruct, it has enhanced instruction - following capabilities.
• Under Development: As a beta version, it is continuously being improved, and user feedback will contribute to future refinements.

💻 Usage Examples

Basic Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("stockmark/Stockmark-2-100B-Instruct-beta")
model = AutoModelForCausalLM.from_pretrained(
    "stockmark/Stockmark-2-100B-Instruct-beta", device_map="auto", torch_dtype=torch.bfloat16
)

instruction = "自然言語処理とは？"
input_ids = tokenizer.apply_chat_template(
    [{"role": "user", "content": instruction}], add_generation_prompt=True, return_tensors="pt"
).to(model.device)

with torch.inference_mode():
    tokens = model.generate(
        input_ids,
        max_new_tokens = 512,
        do_sample = True,
        temperature = 0.7,
        top_p = 0.95,
        repetition_penalty = 1.05
    )
    
output = tokenizer.decode(tokens[0], skip_special_tokens=True)
print(output)

📚 Documentation

For more detailed information about Stockmark-2-100B-Instruct-beta, please refer to our blog.

📄 License

This project is licensed under the MIT license.

Developed by

This project is developed by Stockmark Inc.

Author

Takahiro Omi