Approval Bureau Merged

🚀 Approval Bureau Merged Model

This model is the result of merging the base model DeepSeek-R1-Distill-Qwen-32B with the approval_bureau_model_lora adapter, suitable for document review, text generation, and content approval scenarios.

🚀 Quick Start

Loading the Model with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Set the model path
model_path = "distill/approval-bureau-merged"  # Or use a local path

# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float16,  # Use half-precision to save memory
    device_map="auto"           # Automatically allocate to available devices
)

# Example of text generation
input_text = "Please review the following document content:"
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
outputs = model.generate(
    inputs.input_ids,
    max_length=500,
    temperature=0.7,
    top_p=0.9
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Streaming Output Example

from transformers import TextIteratorStreamer
from threading import Thread

# Initialize streaming output
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)

# Create generation parameters
generation_kwargs = {
    "input_ids": inputs.input_ids,
    "max_length": 500,
    "temperature": 0.7,
    "top_p": 0.9,
    "streamer": streamer
}

# Run generation in a background thread
thread = Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()

# Stream output results
for text in streamer:
    print(text, end="", flush=True)

⚠️ Important Note

• This model requires a large amount of GPU memory (at least 24GB is recommended).
• For environments with limited memory, it is recommended to use 4-bit or 8-bit quantization.
• The model output is for reference only. For critical decisions, please combine with manual review.

✨ Features

• Based on the powerful DeepSeek-R1-Distill-Qwen-32B architecture.
• Integrates a LoRA fine-tuning layer specific to the approval bureau.
• Suitable for scenarios related to document review, text generation, and content approval.
• Retains the inference ability of the base model while enhancing specific domain functions.

📦 Model Information

📄 License

This model is licensed under the GPL-3.0 license. Please ensure you understand the relevant license terms before use.

Acknowledgments

Thanks to the DeepSeek and Qwen teams for providing the base model, and to all developers who contributed to the model training and tuning.