Climate-Education-QA-Chatbot / model_architecture.json
Climi's picture
upload everything about the climate education chatbot
74b76c0 verified
{
"model_architecture": {
"name": "T5-small",
"type": "Encoder-Decoder Transformer",
"purpose": "Text-to-Text Generation for Climate Education QA",
"framework": "TensorFlow",
"pretrained_source": "Hugging Face Transformers",
"parameters": 60506624,
"size_mb": 230.8
},
"transformer_config": {
"vocab_size": 32128,
"d_model": 512,
"num_heads": 8,
"num_layers": 6,
"d_ff": 2048,
"d_kv": 64,
"n_positions": 512,
"dropout_rate": 0.1
},
"architecture_details": {
"encoder": {
"layers": 6,
"attention_heads": 8,
"hidden_size": 512,
"feed_forward_size": 2048,
"key_value_size": 64
},
"decoder": {
"layers": 6,
"attention_heads": 8,
"hidden_size": 512,
"feed_forward_size": 2048
}
},
"task_adaptation": {
"input_format": "question: [QUESTION]",
"output_format": "Natural language answer",
"max_input_length": 128,
"max_output_length": 100,
"tokenizer": "SentencePiece",
"vocabulary_size": 32128
},
"training_configuration": {
"fine_tuning_type": "Full model fine-tuning",
"dataset_size": {
"training": 60,
"validation": 13,
"test": 13
},
"domain": "Climate Education",
"optimization": "AdamW with learning rate scheduling"
},
"computational_requirements": {
"memory_estimate": "~230.8 MB",
"inference_time": "~17 seconds per generation",
"training_time": "~27 minutes for 20 epochs",
"hardware_used": "Google Colab (GPU/CPU)"
},
"model_comparison": {
"chosen_model": "T5-small (60M parameters)",
"alternatives_considered": [
"T5-base (220M) - larger but slower",
"BERT - extractive QA only",
"GPT-2 - decoder-only architecture"
],
"selection_rationale": [
"Generative QA capability",
"Manageable size for fine-tuning",
"Good performance on text-to-text tasks",
"Established architecture for educational applications"
]
},
"documentation_created": "2025-06-17T01:56:25.610039"
}