| { |
| "model_type": "jora_mini_hybrid", |
| "architectures": [ |
| "JoraMiniForCausalLM" |
| ], |
| "version": "v11.2-hybrid", |
| "hidden_size": 512, |
| "num_hidden_layers": 16, |
| "intermediate_size": 1536, |
| "vocab_size": 96000, |
| "max_position_embeddings": 2048, |
| "num_attention_heads": 4, |
| "num_key_value_heads": 1, |
| "head_dim": 128, |
| "rope_head_dim": 128, |
| "kv_lora_rank": 128, |
| "linear_num_heads": 4, |
| "linear_num_kv_heads": 1, |
| "linear_key_head_dim": 128, |
| "linear_value_head_dim": 128, |
| "linear_conv_kernel_dim": 4, |
| "full_attention_interval": 4, |
| "layer_types": [ |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention" |
| ], |
| "num_full_attention_layers": 4, |
| "num_linear_attention_layers": 12, |
| "rope_theta": 100000.0, |
| "partial_rotary_factor": 0.25, |
| "hidden_act": "silu", |
| "rms_norm_eps": 1e-06, |
| "tie_word_embeddings": true, |
| "use_cache": true, |
| "attention_bias": false, |
| "dtype": "bfloat16", |
| "eos_token_id": 1, |
| "bos_token_id": 0, |
| "pad_token_id": 2, |
| "total_steps": 76352, |
| "final_loss": 3.9191064834594727, |
| "tokens_seen": 10007609344, |
| "token_budget": 10000000000 |
| } |
