diff --git a/gold-causal/best/config.json b/gold-causal/best/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/best/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/best/generation_config.json b/gold-causal/best/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/best/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/last/config.json b/gold-causal/last/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/last/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/last/generation_config.json b/gold-causal/last/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/last/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/step_001180/config.json b/gold-causal/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/step_001180/generation_config.json b/gold-causal/step_001180/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/step_001180/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/step_002360/config.json b/gold-causal/step_002360/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/step_002360/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/step_002360/generation_config.json b/gold-causal/step_002360/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/step_002360/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/step_003540/config.json b/gold-causal/step_003540/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/step_003540/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/step_003540/generation_config.json b/gold-causal/step_003540/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/step_003540/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/step_004720/config.json b/gold-causal/step_004720/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/step_004720/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/step_004720/generation_config.json b/gold-causal/step_004720/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/step_004720/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-causal/step_005900/config.json b/gold-causal/step_005900/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-causal/step_005900/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-causal/step_005900/generation_config.json b/gold-causal/step_005900/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-causal/step_005900/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/best/config.json b/gold-constant/best/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/best/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/best/generation_config.json b/gold-constant/best/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/best/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/last/config.json b/gold-constant/last/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/last/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/last/generation_config.json b/gold-constant/last/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/last/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/step_001180/config.json b/gold-constant/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/step_001180/generation_config.json b/gold-constant/step_001180/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/step_001180/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/step_002360/config.json b/gold-constant/step_002360/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/step_002360/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/step_002360/generation_config.json b/gold-constant/step_002360/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/step_002360/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/step_003540/config.json b/gold-constant/step_003540/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/step_003540/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/step_003540/generation_config.json b/gold-constant/step_003540/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/step_003540/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/step_004720/config.json b/gold-constant/step_004720/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/step_004720/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/step_004720/generation_config.json b/gold-constant/step_004720/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/step_004720/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-constant/step_005900/config.json b/gold-constant/step_005900/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-constant/step_005900/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-constant/step_005900/generation_config.json b/gold-constant/step_005900/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-constant/step_005900/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/best/config.json b/gold-cosine/best/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/best/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/best/generation_config.json b/gold-cosine/best/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/best/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/last/config.json b/gold-cosine/last/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/last/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/last/generation_config.json b/gold-cosine/last/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/last/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/step_001180/config.json b/gold-cosine/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/step_001180/generation_config.json b/gold-cosine/step_001180/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/step_001180/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/step_002360/config.json b/gold-cosine/step_002360/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/step_002360/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/step_002360/generation_config.json b/gold-cosine/step_002360/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/step_002360/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/step_003540/config.json b/gold-cosine/step_003540/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/step_003540/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/step_003540/generation_config.json b/gold-cosine/step_003540/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/step_003540/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/step_004720/config.json b/gold-cosine/step_004720/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/step_004720/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/step_004720/generation_config.json b/gold-cosine/step_004720/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/step_004720/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-cosine/step_005900/config.json b/gold-cosine/step_005900/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-cosine/step_005900/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-cosine/step_005900/generation_config.json b/gold-cosine/step_005900/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-cosine/step_005900/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/best/config.json b/gold-linear/best/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/best/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/best/generation_config.json b/gold-linear/best/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/best/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/last/config.json b/gold-linear/last/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/last/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/last/generation_config.json b/gold-linear/last/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/last/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/step_001180/config.json b/gold-linear/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/step_001180/generation_config.json b/gold-linear/step_001180/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/step_001180/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/step_002360/config.json b/gold-linear/step_002360/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/step_002360/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/step_002360/generation_config.json b/gold-linear/step_002360/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/step_002360/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/step_003540/config.json b/gold-linear/step_003540/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/step_003540/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/step_003540/generation_config.json b/gold-linear/step_003540/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/step_003540/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/step_004720/config.json b/gold-linear/step_004720/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/step_004720/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/step_004720/generation_config.json b/gold-linear/step_004720/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/step_004720/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-linear/step_005900/config.json b/gold-linear/step_005900/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-linear/step_005900/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-linear/step_005900/generation_config.json b/gold-linear/step_005900/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-linear/step_005900/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-masked/step_001180/config.json b/gold-masked/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-masked/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/best/config.json b/gold-shift/best/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/best/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/best/generation_config.json b/gold-shift/best/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/best/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/last/config.json b/gold-shift/last/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/last/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/last/generation_config.json b/gold-shift/last/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/last/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/step_001180/config.json b/gold-shift/step_001180/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/step_001180/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/step_001180/generation_config.json b/gold-shift/step_001180/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/step_001180/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/step_001180/model.safetensors b/gold-shift/step_001180/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d04b2a4f74d15f9875f70ba7a021a5406e63be55
--- /dev/null
+++ b/gold-shift/step_001180/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8734cd215ec6c1ad1dbbcda0b8f7a6b6811f58b86f2bf4424afdc9e0bec8956f
+size 442333936
diff --git a/gold-shift/step_001180/trainer_state.pt b/gold-shift/step_001180/trainer_state.pt
new file mode 100644
index 0000000000000000000000000000000000000000..fc8305f0c90085cc41dc7aeba1d34a7ca4419258
--- /dev/null
+++ b/gold-shift/step_001180/trainer_state.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a774add1046222e8460d718c1101aa8901b57efff6dcabe924e3e9a52d1a71bb
+size 884221979
diff --git a/gold-shift/step_002360/config.json b/gold-shift/step_002360/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/step_002360/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/step_002360/generation_config.json b/gold-shift/step_002360/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/step_002360/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/step_002360/trainer_state.pt b/gold-shift/step_002360/trainer_state.pt
new file mode 100644
index 0000000000000000000000000000000000000000..1825a1bed758f615cb1993a68030f2f51854db37
--- /dev/null
+++ b/gold-shift/step_002360/trainer_state.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d66ceb7b97a6d826168ac95768ba4838e5205a17a26f9fe3799c740e643a032d
+size 884221979
diff --git a/gold-shift/step_003540/config.json b/gold-shift/step_003540/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/step_003540/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/step_003540/generation_config.json b/gold-shift/step_003540/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/step_003540/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/step_004720/config.json b/gold-shift/step_004720/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/step_004720/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/step_004720/generation_config.json b/gold-shift/step_004720/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/step_004720/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}
diff --git a/gold-shift/step_005900/config.json b/gold-shift/step_005900/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5800c6cb97e51281d294ee4e187a222051889d3
--- /dev/null
+++ b/gold-shift/step_005900/config.json
@@ -0,0 +1,33 @@
+{
+  "architectures": [
+    "FlexQwenForCausalLM"
+  ],
+  "cls_token_id": 1,
+  "dropout_rate": 0.0,
+  "dtype": "float32",
+  "embedding_dim": 768,
+  "head_dim": 64,
+  "hidden_dim": 1536,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "flexqwen",
+  "moe_hidden_dim": 512,
+  "moe_num_experts": 0,
+  "moe_num_experts_per_token": -1,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_kv_groups": 4,
+  "pad_token_id": 3,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 10000,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.3.0",
+  "vocab_size": 64000,
+  "auto_map": {
+    "AutoConfig": "flexqwen.FlexQwenConfig",
+    "AutoModel": "flexqwen.FlexQwen",
+    "AutoModelForCausalLM": "flexqwen.FlexQwenForCausalLM",
+    "AutoModelForSequenceClassification": "flexqwen.FlexQwenForSequenceClassification"
+  }
+}
\ No newline at end of file
diff --git a/gold-shift/step_005900/generation_config.json b/gold-shift/step_005900/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdde72fc7644fe801e3a54754b0594ec498f2120
--- /dev/null
+++ b/gold-shift/step_005900/generation_config.json
@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 3,
+  "transformers_version": "5.3.0"
+}