Create config.json

config.json

@@ -0,0 +1,59 @@
+{
+  "architectures": [
+    "AdaptiveRiverLM"
+  ],
+  "model_type": "adaptive-river",
+  "auto_map": {
+    "AutoConfig": "configuration_adaptive_river.AdaptiveRiverConfig",
+    "AutoModelForCausalLM": "modeling_adaptive_river.AdaptiveRiverLM"
+  },
+  "vocab_size": 50257,
+  "hidden_size": 1024,
+  "num_hidden_layers": 24,
+  "intermediate_size": 4096,
+  "num_attention_heads": 16,
+  "max_position_embeddings": 2048,
+  "rms_norm_eps": 1e-05,
+  "initializer_range": 0.02,
+  "use_cache": false,
+  "pad_token_id": 50256,
+  "bos_token_id": 50256,
+  "eos_token_id": 50256,
+  "tie_word_embeddings": false,
+  "rope_theta": 10000.0,
+  "rope_scaling": null,
+  "attention_dropout": 0.0,
+  "moe_config": {
+    "num_attention_experts": 6,
+    "num_ffn_experts": 4,
+    "attention_top_k": 6,
+    "ffn_top_k": 1,
+    "gate_temperature": 0.7,
+    "moe_dropout": 0.0,
+    "load_balance_weight": 0.01,
+    "router_z_weight": 0.001,
+    "entropy_weight": 0.0001,
+    "head_entropy_weight": 0.0001
+  },
+  "mamba_config": {
+    "d_state": 16,
+    "d_conv": 4,
+    "expand_factor": 2,
+    "enhanced_expand_factor": 4,
+    "early_mamba_layers": [0, 1],
+    "enhanced_mamba_layers": [22, 23]
+  },
+  "layer_types": {
+    "0": "mamba",
+    "1": "mamba",
+    "2-21": "moe",
+    "22": "mamba_enhanced",
+    "23": "mamba_enhanced"
+  },
+  "budget_ratio_default": 1.0,
+  "checkpoint_attention_threshold": 0.35,
+  "checkpoint_ffn_threshold": 0.35,
+  "rotary_pct": 1.0,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.36.0"
+}