{ "model_name": "emma-3", "model_type": "gpt", "onnx_file": "emma3.onnx", "architecture": "decoder-only-transformer-gqa", "vocab_size": 32000, "hidden_size": 768, "num_hidden_layers": 16, "num_attention_heads": 12, "num_key_value_heads": 4, "num_kv_heads": 4, "intermediate_size": 2048, "context_length": 2048, "rope_theta": 10000.0, "norm_type": "rmsnorm", "activation": "swiglu", "tokenizer": "sentencepiece", "tied_embeddings": true, "alignment": "SFT", "gpu_memory_limit_gb": 10.0, "onnx_opset": 18 }