{ "model_class": "KeyeARModel", "qwen_config": { "model_class": "UnifiedQwen3Model", "vocab_size": 151936, "embed_dim": 4096, "num_layers": 36, "tie_word_embeddings": false, "hidden_act": "silu", "num_heads": 32, "num_kv_heads": 8, "head_dim": 128, "attn_dropout": 0.0, "attention_function": "flash_attention_2", "q_proj_bias": false, "k_proj_bias": false, "v_proj_bias": false, "attention_bias": false, "intermediate_dim": 12288, "max_seq_len": 40960, "rope_base": 1000000.0, "rope_impl": "llama", "rope_theta": 10000.0, "rope_scaling": null, "use_sliding_window": false, "sliding_window": null, "norm_eps": 1e-06, "rms_norm_eps": 1e-06, "q_norm": true, "k_norm": true, "eos_token_id": 151645, "use_multimodal_rope": false, "mrope_section": null, "pre_embedding_size": null, "pre_embedding_tokens": null, "image_token_id": 151655, "pad_token_id": null, "q_eos_token": 151681, "vision_start_token_id": 151652, "vision_end_token_id": 151653, "output_last_hidden_states_only": false, "token_decoder_with_teacher_forcing": true, "codebook_size": 65536, "n_q_tokens": 8, "skip_output_layer": false }, "tokenizer_config": { "model_class": "KeyeImageTokenizer", "vision_config": { "model_class": "KeyeVL1_5VisionModel", "image_size": 384, "patch_size": 14, "num_channels": 3, "hidden_size": 1152, "num_hidden_layers": 27, "num_attention_heads": 16, "intermediate_size": 4304, "hidden_act": "gelu_pytorch_tanh", "max_seq_len": 4096, "layer_norm_eps": 1e-06, "attention_dropout": 0.0, "has_learnable_position_embedding": true, "use_qk_norm": false, "qk_norm_eps": 1e-06, "rope_theta": 10000.0, "attention_function": "flash_attention_2", "output_attentions": false, "output_hidden_states": false }, "codebook_size": 65536, "embedding_dim": 128, "init_embedding_dim": 4096, "llm_hidden_size": 4096, "n_q_tokens": 8, "split_dim": false, "split_voc": 8, "add_voc_reducer": false, "vq_sampling_mode": "argmin", "vq_temperature": 1.0, "vq_temperature_decay": 0.999, "vq_min_temperature": 0.1, "pre_llm_align": false, "output_dim": 1024, "fusion_type": "sum" }, "token_decoder_config": { "model_class": "UnifiedTokenDecoder", "vocab_size": 65536, "max_pos_length": 65537, "max_length": 9, "d_model": 512, "eos_token": 151681, "nhead": 4, "num_layers": 1, "dim_feedforward": 2048, "use_gradient_checkpointing": true, "input_dim": 4096, "reduce": true, "attention_function": "eager" }, "__class__": "KeyeARConfig" }