CoolFace
Modelpublic

kunaliitkgp09/working-unified-multi-model-pt

sourceHugging Facemitupdated 1y agoView on Hugging Face
0likes
Model Card

Working Unified Multi-Model (.pt)

A complete unified PyTorch model that delegates to specialized child models for different AI tasks.

๐Ÿš€ Features

  • โ€”Single .pt file containing all capabilities
  • โ€”True model delegation to specialized child models
  • โ€”Unified reasoning and routing
  • โ€”Production-ready deployment

๐Ÿ“ฆ Model Components

  • โ€”Base Reasoning Model: distilgpt2 (~300MB)
  • โ€”Image Captioning Model: BLIP (~990MB)
  • โ€”Text-to-Image Model: Stable Diffusion v1.5
  • โ€”Task Classifiers: Routing and confidence scoring
  • โ€”Embeddings: Task type embeddings

๐ŸŽฏ Capabilities

  1. 1.Text Processing: Q&A, summarization, text generation
  2. 2.Image Captioning: Describe images using BLIP model
  3. 3.Text-to-Image: Generate images using Stable Diffusion
  4. 4.Reasoning: Step-by-step reasoning tasks

๐Ÿ“Š Model Size

  • โ€”File Size: 1.26 GB
  • โ€”Total Parameters: ~1.2B parameters
  • โ€”Architecture: Unified PyTorch model

๐Ÿ”ง Usage

python
import torch
from working_complete_unified_model_pt import WorkingUnifiedMultiModelPT

# Load the model
model = WorkingUnifiedMultiModelPT.load_model("working_unified_multi_model.pt")

# Process different types of requests
result = model.process("What is machine learning?")
print(f"Task: {result['task_type']}")
print(f"Output: {result['output']}")

result = model.process("Generate an image of a peaceful forest")
print(f"Task: {result['task_type']}")
print(f"Output: {result['output']}")

๐Ÿ—๏ธ Architecture

The model uses a unified architecture where:

  1. 1.Parent LLM (distilgpt2) analyzes requests and routes to appropriate child models
  2. 2.Child Models handle specialized tasks:
  3. 3.BLIP for image captioning
  4. 4.Stable Diffusion for text-to-image generation
  5. 5.Base model for text processing and reasoning

๐ŸŽ‰ Key Innovations

  • โ€”Single .pt file for all capabilities
  • โ€”True delegation to specialized models
  • โ€”Unified interface like DeepSeek
  • โ€”Portable across environments
  • โ€”Production-ready deployment

๐Ÿ“„ License

MIT License

๐Ÿค Contributing

This model demonstrates the future of AI - unified, portable, and intelligent models that can handle multiple tasks through intelligent delegation.