| { |
| "_name_or_path": "pico-7b", |
| "architectures": [ |
| "PiCoForCausalLM" |
| ], |
| "model_type": "pico", |
| "torch_dtype": "bfloat16", |
| "transformers_version": "4.46.0", |
| "auto_map": { |
| "AutoConfig": "pico.configuration.PiCoConfig", |
| "AutoModelForCausalLM": "pico.modeling.PiCoForCausalLM" |
| }, |
|
|
| "vocab_size": 152064, |
| "hidden_size": 2048, |
| "intermediate_size": 5888, |
| "num_hidden_layers": 30, |
| "num_attention_heads": 16, |
| "num_key_value_heads": 4, |
| "head_dim": 128, |
|
|
| "hidden_act": "silu", |
| "max_position_embeddings": 131072, |
| "rms_norm_eps": 1e-06, |
| "rope_theta": 10000.0, |
| "rope_scaling": null, |
| "tie_word_embeddings": false, |
| "use_cache": false, |
| "sliding_window": 4096, |
| "attention_dropout": 0.0, |
| "attention_bias": false, |
| "mlp_bias": false, |
| "initializer_range": 0.02, |
| "pretraining_tp": 1, |
|
|
| "num_experts": 8, |
| "num_experts_per_tok": 2, |
| "num_shared_experts": 1, |
| "moe_layers": [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28], |
| "router_aux_loss_coef": 0.001, |
| "router_jitter_noise": 0.01, |
|
|
| "use_hierarchical_routing": true, |
| "num_routing_levels": 2, |
| "expert_capacity_factor": 1.25, |
|
|
| "use_expert_compression": true, |
| "expert_compression_ratio": 0.5, |
|
|
| "use_meta_router": true, |
| "meta_router_hidden_size": 256, |
| "meta_router_learning_rate": 0.01, |
|
|
| "use_shared_knowledge": true, |
| "shared_knowledge_size": 2048, |
|
|
| "use_dynamic_expert_pruning": true, |
| "expert_pruning_threshold": 0.01, |
|
|
| "pad_token_id": 0, |
| "bos_token_id": 1, |
| "eos_token_id": 2 |
| } |
|
|
