| { |
| "model_id": "trimurti-lm", |
| "model_name": "Trimurti-LM", |
| "model_type": "language_model", |
| "architecture": "GPT2", |
| "framework": "transformers", |
| |
| "languages": { |
| "supported": ["en", "hi", "pa"], |
| "language_tags": ["[EN]", "[HI]", "[PA]"], |
| "description": "Trilingual language model supporting English, Hindi, and Punjabi" |
| }, |
| |
| "model_config": { |
| "vocab_size": 8000, |
| "n_positions": 128, |
| "n_embd": 256, |
| "n_layer": 4, |
| "n_head": 4, |
| "n_inner": 512, |
| "activation_function": "gelu_new", |
| "attn_pdrop": 0.1, |
| "embd_pdrop": 0.1, |
| "resid_pdrop": 0.1, |
| "estimated_parameters": "4.7M" |
| }, |
| |
| "tokenizer": { |
| "type": "sentencepiece", |
| "model_type": "unigram", |
| "vocab_size": 8000, |
| "character_coverage": 0.9995, |
| "byte_fallback": true, |
| "model_path": "final_corpus/multilingual_spm.model" |
| }, |
| |
| "training": { |
| "corpus": "final_corpus/multilingual_corpus_train.txt", |
| "validation": "final_corpus/multilingual_corpus_val.txt", |
| "total_steps": 5000, |
| "batch_size": 2, |
| "gradient_accumulation": 8, |
| "learning_rate": 2e-4, |
| "warmup_steps": 1000, |
| "effective_batch_size": 16 |
| }, |
| |
| "checkpoints": { |
| "path": "checkpoints_tiny", |
| "available_checkpoints": [ |
| "step1000", |
| "step2000", |
| "step3000", |
| "step4000", |
| "step5000", |
| "final" |
| ] |
| }, |
| |
| "evaluation": { |
| "overall_accuracy": 100.0, |
| "english_accuracy": 100.0, |
| "hindi_accuracy": 100.0, |
| "punjabi_accuracy": 100.0, |
| "mixed_accuracy": 100.0, |
| "avg_english_perplexity": 42.29, |
| "avg_hindi_perplexity": 50.56, |
| "avg_punjabi_perplexity": 63.42 |
| }, |
| |
| "entry_points": { |
| "training": "python train_model.py", |
| "testing": "python test_model.py", |
| "evaluation": "python evaluate_model.py", |
| "preprocessing": "python preprocess.py", |
| "web_interface": "python web_interface.py" |
| }, |
| |
| "dependencies": [ |
| "torch", |
| "transformers", |
| "sentencepiece", |
| "tqdm", |
| "gradio", |
| "pandas", |
| "numpy" |
| ], |
| |
| "filter": [ |
| { |
| "bool": { |
| "should": [ |
| { |
| "term": { "path": "model_index.json" } |
| }, |
| { |
| "regexp": { "path": "[^/]*\\.safetensors" } |
| }, |
| { |
| "regexp": { "path": "[^/]*\\.ckpt" } |
| }, |
| { |
| "regexp": { "path": "[^/]*\\.bin" } |
| } |
| ], |
| "minimum_should_match": 1 |
| } |
| } |
| ], |
| |
| "files": [ |
| { |
| "path": "checkpoints_tiny/final/model.safetensors", |
| "description": "Final trained model weights" |
| }, |
| { |
| "path": "checkpoints_tiny/final/config.json", |
| "description": "Model configuration" |
| }, |
| { |
| "path": "checkpoints_tiny/final/generation_config.json", |
| "description": "Generation settings" |
| }, |
| { |
| "path": "final_corpus/multilingual_spm.model", |
| "description": "SentencePiece tokenizer model" |
| }, |
| { |
| "path": "final_corpus/multilingual_spm.vocab", |
| "description": "Tokenizer vocabulary" |
| }, |
| { |
| "path": "train_model.py", |
| "description": "Training script" |
| }, |
| { |
| "path": "test_model.py", |
| "description": "Testing and inference script" |
| }, |
| { |
| "path": "evaluate_model.py", |
| "description": "Evaluation script" |
| }, |
| { |
| "path": "preprocess.py", |
| "description": "Data preprocessing script" |
| }, |
| { |
| "path": "web_interface.py", |
| "description": "Gradio web interface" |
| } |
| ] |
| } |
