fix config

config.json

@@ -8,25 +8,26 @@
     "AutoModel": "modeling_openpangu_moe.PanguUltraMoEModel",
     "AutoModelForCausalLM": "modeling_openpangu_moe.PanguUltraMoEForCausalLM"
   },
-  "num_dense_layers": 3,
+  "first_k_dense_replace": 3,
   "hidden_act": "silu",
   "hidden_size": 7680,
   "initializer_range": 0.02,
   "intermediate_size": 18432,
-  "attention_kv_lora_dim": 512,
+  "kv_lora_rank": 512,
   "max_position_embeddings": 131072, 
   "model_type": "pangu_ultra_moe",
   "moe_intermediate_size": 2048,
-  "num_routed_experts": 256,
-  "num_shared_experts": 1,
+  "n_routed_experts": 256,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
   "num_attention_heads": 128,
   "num_experts_per_tok": 8,
   "num_hidden_layers": 61,
   "num_key_value_heads": 128,
-  "num_mtp_layers": 1,
-  "attention_q_lora_dim": 1536,
-  "attention_qk_dim": 128,
-  "attention_qk_rope_dim": 64,
+  "num_nextn_predict_layers": 1,
+  "q_lora_rank": 1536,
+  "qk_nope_head_dim": 128,
+  "qk_rope_head_dim": 64,
   "rms_norm_eps": 1e-05,
   "rope_theta": 25600000,
   "routed_scaling_factor": 2.5,
@@ -35,112 +36,6 @@
   "torch_dtype": "bfloat16",
   "transformers_version": "4.48.2",
   "use_cache": true,
-  "attention_v_dim": 128,
-  "vocab_size": 153600,
-  "quantize": "w8a8_dynamic",
-  "quantization_config": {
-    "config_groups": {
-      "group_0": {
-        "input_activations": {
-          "actorder": null,
-          "block_structure": null,
-          "dynamic": true,
-          "group_size": null,
-          "num_bits": 8,
-          "observer": "memoryless",
-          "observer_kwargs": {},
-          "strategy": "token",
-          "symmetric": true,
-          "type": "int"
-        },
-        "output_activations": null,
-        "targets": [
-          "Linear"
-        ],
-        "weights": {
-          "actorder": null,
-          "block_structure": null,
-          "dynamic": false,
-          "group_size": null,
-          "num_bits": 8,
-          "observer": "minmax",
-          "observer_kwargs": {},
-          "strategy": "channel",
-          "symmetric": true,
-          "type": "int"
-        }
-      }
-    },
-    "format": "int-quantized",
-    "global_compression_ratio": 1.5943962512751308,
-    "ignore": [
-      "model.layers.0.self_attn.kv_b_proj",
-      "model.layers.1.self_attn.kv_b_proj",
-      "model.layers.2.self_attn.kv_b_proj",
-      "model.layers.3.self_attn.kv_b_proj",
-      "model.layers.4.self_attn.kv_b_proj",
-      "model.layers.5.self_attn.kv_b_proj",
-      "model.layers.6.self_attn.kv_b_proj",
-      "model.layers.7.self_attn.kv_b_proj",
-      "model.layers.8.self_attn.kv_b_proj",
-      "model.layers.9.self_attn.kv_b_proj",
-      "model.layers.10.self_attn.kv_b_proj",
-      "model.layers.11.self_attn.kv_b_proj",
-      "model.layers.12.self_attn.kv_b_proj",
-      "model.layers.13.self_attn.kv_b_proj",
-      "model.layers.14.self_attn.kv_b_proj",
-      "model.layers.15.self_attn.kv_b_proj",
-      "model.layers.16.self_attn.kv_b_proj",
-      "model.layers.17.self_attn.kv_b_proj",
-      "model.layers.18.self_attn.kv_b_proj",
-      "model.layers.19.self_attn.kv_b_proj",
-      "model.layers.20.self_attn.kv_b_proj",
-      "model.layers.21.self_attn.kv_b_proj",
-      "model.layers.22.self_attn.kv_b_proj",
-      "model.layers.23.self_attn.kv_b_proj",
-      "model.layers.24.self_attn.kv_b_proj",
-      "model.layers.25.self_attn.kv_b_proj",
-      "model.layers.26.self_attn.kv_b_proj",
-      "model.layers.27.self_attn.kv_b_proj",
-      "model.layers.28.self_attn.kv_b_proj",
-      "model.layers.29.self_attn.kv_b_proj",
-      "model.layers.30.self_attn.kv_b_proj",
-      "model.layers.31.self_attn.kv_b_proj",
-      "model.layers.32.self_attn.kv_b_proj",
-      "model.layers.33.self_attn.kv_b_proj",
-      "model.layers.34.self_attn.kv_b_proj",
-      "model.layers.35.self_attn.kv_b_proj",
-      "model.layers.36.self_attn.kv_b_proj",
-      "model.layers.37.self_attn.kv_b_proj",
-      "model.layers.38.self_attn.kv_b_proj",
-      "model.layers.39.self_attn.kv_b_proj",
-      "model.layers.40.self_attn.kv_b_proj",
-      "model.layers.41.self_attn.kv_b_proj",
-      "model.layers.42.self_attn.kv_b_proj",
-      "model.layers.43.self_attn.kv_b_proj",
-      "model.layers.44.self_attn.kv_b_proj",
-      "model.layers.45.self_attn.kv_b_proj",
-      "model.layers.46.self_attn.kv_b_proj",
-      "model.layers.47.self_attn.kv_b_proj",
-      "model.layers.48.self_attn.kv_b_proj",
-      "model.layers.49.self_attn.kv_b_proj",
-      "model.layers.50.self_attn.kv_b_proj",
-      "model.layers.51.self_attn.kv_b_proj",
-      "model.layers.52.self_attn.kv_b_proj",
-      "model.layers.53.self_attn.kv_b_proj",
-      "model.layers.54.self_attn.kv_b_proj",
-      "model.layers.55.self_attn.kv_b_proj",
-      "model.layers.56.self_attn.kv_b_proj",
-      "model.layers.57.self_attn.kv_b_proj",
-      "model.layers.58.self_attn.kv_b_proj",
-      "model.layers.59.self_attn.kv_b_proj",
-      "model.layers.60.self_attn.kv_b_proj",
-      "lm_head",
-      "model.layers.61.self_attn.kv_b_proj",
-      "model.layers.61.shared_head.head"
-    ],
-    "kv_cache_scheme": null,
-    "quant_method": "compressed-tensors",
-    "quantization_status": "compressed"
-  }
-}
+  "v_head_dim": 128,
+  "vocab_size": 153600
+}