| { |
| "architectures": [ |
| "Raon1MiniForCausalLM" |
| ], |
| "model_type": "raon1_mini", |
| "dtype": "bfloat16", |
| "hidden_size": 32, |
| "num_hidden_layers": 8, |
| "num_attention_heads": 1, |
| "head_dim": 32, |
| "vocab_size": 259840, |
| "tie_word_embeddings": false, |
| "max_position_embeddings": 4096, |
| "rms_norm_eps": 1e-05, |
| "rope_parameters": null, |
| "hidden_act": "situ", |
| "activation_situ_beta": 4.0, |
| "activation_situ_linear_beta": 25.0, |
| "intermediate_size": 128, |
| "first_k_dense_replace": 1, |
| "moe_layer_freq": 1, |
| "num_experts": 64, |
| "num_experts_per_token": 6, |
| "moe_intermediate_size": 32, |
| "num_shared_experts": 1, |
| "moe_router_activation_func": "sigmoid", |
| "moe_renormalize": true, |
| "routed_scaling_factor": 1.0, |
| "q_lora_rank": 32, |
| "kv_lora_rank": 32, |
| "qk_nope_head_dim": 32, |
| "qk_rope_head_dim": 32, |
| "v_head_dim": 32, |
| "mla_use_nope": true, |
| "mla_use_output_gate": true, |
| "raon_mhc_streams": 4, |
| "rda_config": { |
| "num_heads": 1, |
| "head_dim": 32, |
| "conv_kernel_size": 4, |
| "full_rank_gate": true, |
| "gate_lower_bound": -5.0, |
| "mla_layers": [ |
| 4, |
| 8 |
| ], |
| "rda_layers": [ |
| 1, |
| 2, |
| 3, |
| 5, |
| 6, |
| 7 |
| ] |
| }, |
| "auto_map": { |
| "AutoConfig": "configuration_raon1_mini.Raon1MiniConfig", |
| "AutoModel": "modeling_raon1_mini.Raon1MiniModel", |
| "AutoModelForCausalLM": "modeling_raon1_mini.Raon1MiniForCausalLM" |
| }, |
| "quantization_config": { |
| "quant_method": "mxfp8", |
| "activation_scheme": "dynamic", |
| "weight_block_size": [ |
| 1, |
| 32 |
| ], |
| "ignored_layers": [ |
| "lm_head", |
| "model.layers.*.self_attn*", |
| "model.layers.*.mlp*", |
| "model.layers.*.moe.gate*", |
| "model.layers.*.moe.shared_experts*", |
| "model.layers.*.moe.experts$" |
| ], |
| "dequantize": false |
| }, |
| "transformers_version": "5.16.1", |
| "rda_decay_rank": 32 |
| } |
|
|
