Model Card for Korean STT Data error correction Feedback Model

This model provides feedback on Korean STT Data error correction, utilizing the Gemma2 architecture fine-tuned specifically for this task. It evaluates and generates constructive feedback to enhance the quality of STT Data.

Model Details

Model Description

This model is based on the Gemma2 architecture and has been fine-tuned on a dataset of Korean STT Data error correction. It aims to provide useful feedback for improving STT Data by evaluating various aspects of the text and suggesting enhancements.

Developed by: Kray, Yunyoung
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Model type: Text Generation
Language(s) (NLP): Korean
License: [More Information Needed]
Finetuned from model [optional]: Gemma2

Model Sources [optional]

Repository: [More Information Needed]
Paper [optional]: [More Information Needed]
Demo [optional]: [More Information Needed]

Uses

Direct Use

The model can be used directly to generate feedback on Korean STT Data error correction, helping enhance the understanding of counseling

Downstream Use [optional]

When integrated into applications or services aimed at improving Contact Center or Chatbots, the model can enhance user-generated content through feedback and suggestions.

Out-of-Scope Use

The model may not perform well on non-Korean text. It is not designed for tasks outside the scope of personal statement improvement.

Bias, Risks, and Limitations

Recommendations

Users should be aware that the model's feedback is based on patterns learned from training data, which might not cover all possible STT Data scenarios. It is recommended to use the feedback as a guide rather than an absolute measure.

How to Get Started with the Model

To get started with the model, you can load it using the Hugging Face transformers library and use it to generate feedback for Korean STT Data.


import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Load the pre-trained model and tokenizer for STT correction
model = AutoModelForCausalLM.from_pretrained("stt-error-correction-model")
tokenizer = AutoTokenizer.from_pretrained("stt-error-correction-tokenizer")

# Create a text generation pipeline using the fine-tuned model and tokenizer
pipe_finetuned = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=512)

# Define a placeholder for the STT error text
stt_error_text = "안뇽하새요"

# Construct a list of messages for STT correction
messages = [
    {
        "role": "user",
        "content": (
            f"STT 오류가 포함된 텍스트를 올바르게 수정해주세요.\n"
            f"STT 오류 텍스트: {stt_error_text}\n"
        )
    }
]

# Prepare the input prompt using the tokenizer
prompt = pipe_finetuned.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# Generate correction by passing the formatted prompt to the pipeline
outputs = pipe_finetuned(
    prompt,
    do_sample=True,          # Enable sampling to generate diverse outputs
    temperature=0.2,         # Control randomness in text generation (lower value makes the output more focused)
    top_k=50,                # Limit the sampling pool to the top 50 tokens
    top_p=0.95,              # Use nucleus sampling to focus on the top 95% of probability mass
    add_special_tokens=True  # Include special tokens as per the model's requirements
)

# Print the generated correction
print(outputs[0]["generated_text"][len(prompt):])