| { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm", | |
| "class_name": "Qwen3MoeCausalLM", | |
| "config": { | |
| "backbone": { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_backbone", | |
| "class_name": "Qwen3MoeBackbone", | |
| "config": { | |
| "name": "qwen3_moe_backbone", | |
| "trainable": true, | |
| "dtype": { | |
| "module": "keras", | |
| "class_name": "DTypePolicy", | |
| "config": { | |
| "name": "float32" | |
| }, | |
| "registered_name": null | |
| }, | |
| "vocabulary_size": 151936, | |
| "num_layers": 48, | |
| "num_query_heads": 32, | |
| "head_dim": 128, | |
| "hidden_dim": 2048, | |
| "intermediate_dim": 6144, | |
| "moe_intermediate_dim": 768, | |
| "rope_max_wavelength": 10000000, | |
| "num_key_value_heads": 4, | |
| "rope_scaling_factor": 1.0, | |
| "layer_norm_epsilon": 1e-06, | |
| "dropout": 0, | |
| "tie_word_embeddings": false, | |
| "sliding_window_size": null, | |
| "num_experts": 128, | |
| "top_k": 8, | |
| "norm_top_k_prob": true, | |
| "decoder_sparse_step": 1, | |
| "mlp_only_layers": [], | |
| "router_aux_loss_coefficient": 0.001 | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeBackbone" | |
| }, | |
| "preprocessor": { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm_preprocessor", | |
| "class_name": "Qwen3MoeCausalLMPreprocessor", | |
| "config": { | |
| "name": "qwen3_moe_causal_lm_preprocessor_2", | |
| "trainable": true, | |
| "dtype": { | |
| "module": "keras", | |
| "class_name": "DTypePolicy", | |
| "config": { | |
| "name": "float32" | |
| }, | |
| "registered_name": null | |
| }, | |
| "tokenizer": { | |
| "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer", | |
| "class_name": "Qwen3MoeTokenizer", | |
| "config": { | |
| "name": "qwen3_moe_tokenizer", | |
| "trainable": true, | |
| "dtype": { | |
| "module": "keras", | |
| "class_name": "DTypePolicy", | |
| "config": { | |
| "name": "int32" | |
| }, | |
| "registered_name": null | |
| }, | |
| "config_file": "tokenizer.json", | |
| "sequence_length": null, | |
| "add_prefix_space": false, | |
| "unsplittable_tokens": [ | |
| "<tool_response>", | |
| "</think>", | |
| "<|box_end|>", | |
| "</tool_response>", | |
| "<|fim_suffix|>", | |
| "<|quad_start|>", | |
| "<tool_call>", | |
| "<think>", | |
| "<|vision_pad|>", | |
| "<|file_sep|>", | |
| "<|box_start|>", | |
| "<|im_start|>", | |
| "<|endoftext|>", | |
| "</tool_call>", | |
| "<|object_ref_start|>", | |
| "<|repo_name|>", | |
| "<|vision_start|>", | |
| "<|vision_end|>", | |
| "<|object_ref_end|>", | |
| "<|quad_end|>", | |
| "<|fim_middle|>", | |
| "<|video_pad|>", | |
| "<|im_end|>", | |
| "<|image_pad|>", | |
| "<|fim_prefix|>", | |
| "<|fim_pad|>" | |
| ] | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeTokenizer" | |
| }, | |
| "config_file": "preprocessor.json", | |
| "sequence_length": 1024, | |
| "add_start_token": true, | |
| "add_end_token": true | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeCausalLMPreprocessor" | |
| }, | |
| "name": "qwen3_moe_causal_lm" | |
| }, | |
| "registered_name": "keras_hub>Qwen3MoeCausalLM" | |
| } |