| { | |
| "model_type": "tinystate", | |
| "architectures": ["TinyStateForCausalLM"], | |
| "model_name": "TinyState-19B-A9B", | |
| "hidden_size": 6144, | |
| "num_hidden_layers": 48, | |
| "num_attention_heads": 48, | |
| "num_key_value_heads": 8, | |
| "intermediate_size": 16384, | |
| "hidden_act": "silu", | |
| "max_position_embeddings": 32768, | |
| "initializer_range": 0.02, | |
| "rms_norm_eps": 1e-6, | |
| "use_cache": true, | |
| "pad_token_id": 151643, | |
| "bos_token_id": 151643, | |
| "eos_token_id": 151645, | |
| "tie_word_embeddings": false, | |
| "rope_theta": 1000000.0, | |
| "rope_scaling": null, | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "vocab_size": 151936, | |
| "model_params": "19B", | |
| "distillation_ratio": 0.12, | |
| "target_params": 19000000000, | |
| "teacher_model": "Qwen/Qwen3-235B-A22B", | |
| "chunk_count": 8, | |
| "safetensors_format": true, | |
| "version": "A9B", | |
| "torch_dtype": "float16", | |
| "attn_implementation": "flash_attention_2", | |
| "quantization_config": { | |
| "load_in_4bit": true, | |
| "bnb_4bit_quant_type": "nf4", | |
| "bnb_4bit_compute_dtype": "float16" | |
| }, | |
| "moe_active": true, | |
| "num_experts": 8, | |
| "num_experts_per_tok": 2 | |
| } |