You can deploy this model and start earning money today!

FINGU-AI/
FINGU-2.5-instruct-70B

Name: FINGU-2.5-instruct-70B
Rating: 2.5 (5 reviews)

Views

No views yet

FINGU-AI/FINGU-2.5-instruct-70B

Overview

FINGU-AI/FINGU-2.5-instruct-70B is a versatile causal language model designed to excel in various natural language processing (NLP) tasks, including machine translation, text generation, and chat-based applications. The model demonstrates a strong aptitude for reasoning tasks, particularly in the Japanese language, making it a valuable tool for applications requiring logical inference and complex understanding.

Reasoning Capabilities

The model's architecture and training regimen have been optimized to enhance its reasoning abilities. This is particularly evident in tasks involving logical deduction and commonsense reasoning in Japanese. For instance, when evaluated on datasets such as JaQuAD—a Japanese Question Answering Dataset—the model exhibits a nuanced understanding of complex logical structures. :contentReference[oaicite:0]{index=0}

Additionally, FINGU-AI/FINGU-2.5-instruct-70B has been assessed using the JFLD benchmark, which tests a model's ability for deductive reasoning based on formal logic. The model's performance indicates a robust capacity to handle tasks that require understanding and reasoning over formal logical structures.

Example Usage

Installation

Ensure that the required packages are installed:

pip install torch transformers

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Model and Tokenizer
model_id = 'FINGU-AI/FINGU-2.5-instruct-70B'
model = AutoModelForCausalLM.from_pretrained(model_id, attn_implementation="sdpa", torch_dtype=torch.bfloat16, device_map='auto')
tokenizer = AutoTokenizer.from_pretrained(model_id)


# Input Messages for Translation
messages = [
    {"role": "user", "content": """Please reason step by step, and put your final answer within \boxed{}.
     translate korean to Japanese.
     새로운 은행 계좌를 개설하는 절차는 다음과 같습니다:

1. 계좌 개설 목적과 신분 확인을 위한 서류 제출
2. 서류 검토 과정을 거치는 것
3. 고객님의 신원 확인 절차를 진행하는 것
4. 모든 절차가 완료되면 계좌 개설이 가능합니다.

계좌 개설을 원하시는 경우, 신분증과 함께 방문해 주시면 됩니다.
"""}
]

# Tokenize and Generate Response
input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=500,
    do_sample=True,
)

# Decode and Print the Response
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Relevant Datasets

To further evaluate and enhance the reasoning capabilities of FINGU-AI/FINGU-2.5-instruct-70B, the following Japanese reasoning datasets are pertinent:

JaQuAD (Japanese Question Answering Dataset): A human-annotated dataset created for Japanese Machine Reading Comprehension, consisting of 39,696 extractive question-answer pairs on Japanese Wikipedia articles.
📄 ARXIV.ORG
JFLD (Japanese Formal Logic Dataset): A benchmark designed to evaluate deductive reasoning based on formal logic, providing a structured framework to assess logical reasoning capabilities in Japanese.
📄 ACLANTHOLOGY.ORG
JEMHopQA (Japanese Explainable Multi-Hop Question-Answering): A dataset for multi-hop QA in Japanese, including question-answer pairs and supporting evidence in the form of derivation triples, facilitating the development of explainable QA systems.
📄 ACLANTHOLOGY.ORG

These datasets provide diverse challenges that can help in assessing and improving the model's reasoning abilities across different contexts and complexities.

Conclusion

FINGU-AI/FINGU-2.5-instruct-70B stands as a robust and adaptable language model, particularly distinguished by its reasoning capabilities in the Japanese language. Its performance across various reasoning benchmarks underscores its potential for applications that demand advanced logical inference and nuanced understanding in NLP tasks.