{ "architectures": [ "TestAIModelForCausalLM" ], "model_type": "testai", "auto_map": { "AutoConfig": "configuration_test.TestAIConfig", "AutoModelForCausalLM": "modelling_test.TestAIModelForCausalLM" }, "quantization_config": { "quant_method": "fp8", "activation_scheme": "dynamic", "weight_block_size": [128, 128] }, "vocab_size": 151936, "hidden_size": 4096, "intermediate_size": 11008, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 8, "hidden_act": "silu", "max_position_embeddings": 8192, "initializer_range": 0.02, "rms_norm_eps": 1e-6, "use_cache": true, "pad_token_id": 151643, "bos_token_id": 151643, "eos_token_id": 151643, "tie_word_embeddings": false, "rope_theta": 1000000.0, "rope_scaling": null, "attention_dropout": 0.0, "torch_dtype": "float8_e4m3fn", "transformers_version": "4.39.0", "model_name": "TestAI-4B-Base", "model_version": "1.0.0" }