Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,65 @@
+{{ bos_token }}{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role + '\n' }}
+        {% generation %}
+        {{- content}}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>' }}
+        {% endgeneration%}
+        {{- '\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

config.json

@@ -0,0 +1,499 @@
+{
+  "architectures": [
+    "AfmoeForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_afmoe.AfmoeConfig",
+    "AutoModel": "modeling_afmoe.AfmoeModel",
+    "AutoModelForCausalLM": "modeling_afmoe.AfmoeForCausalLM"
+  },
+  "dtype": "bfloat16",
+  "global_attn_every_n_layers": 4,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 12288,
+  "layer_types": [
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "load_balance_coeff": 5e-05,
+  "max_position_embeddings": 262144,
+  "model_type": "afmoe",
+  "moe_intermediate_size": 3072,
+  "mup_enabled": true,
+  "n_group": 1,
+  "num_attention_heads": 48,
+  "num_dense_layers": 6,
+  "num_expert_groups": 1,
+  "num_experts": 256,
+  "num_experts_per_tok": 4,
+  "num_hidden_layers": 60,
+  "num_key_value_heads": 8,
+  "num_limited_groups": 1,
+  "num_shared_experts": 1,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "pack-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 128,
+          "num_bits": 4,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "group",
+          "symmetric": true,
+          "type": "int",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "model.layers.0.self_attn.q_proj",
+      "model.layers.0.self_attn.k_proj",
+      "model.layers.0.self_attn.v_proj",
+      "model.layers.0.self_attn.o_proj",
+      "model.layers.0.self_attn.gate_proj",
+      "model.layers.1.self_attn.q_proj",
+      "model.layers.1.self_attn.k_proj",
+      "model.layers.1.self_attn.v_proj",
+      "model.layers.1.self_attn.o_proj",
+      "model.layers.1.self_attn.gate_proj",
+      "model.layers.2.self_attn.q_proj",
+      "model.layers.2.self_attn.k_proj",
+      "model.layers.2.self_attn.v_proj",
+      "model.layers.2.self_attn.o_proj",
+      "model.layers.2.self_attn.gate_proj",
+      "model.layers.3.self_attn.q_proj",
+      "model.layers.3.self_attn.k_proj",
+      "model.layers.3.self_attn.v_proj",
+      "model.layers.3.self_attn.o_proj",
+      "model.layers.3.self_attn.gate_proj",
+      "model.layers.4.self_attn.q_proj",
+      "model.layers.4.self_attn.k_proj",
+      "model.layers.4.self_attn.v_proj",
+      "model.layers.4.self_attn.o_proj",
+      "model.layers.4.self_attn.gate_proj",
+      "model.layers.5.self_attn.q_proj",
+      "model.layers.5.self_attn.k_proj",
+      "model.layers.5.self_attn.v_proj",
+      "model.layers.5.self_attn.o_proj",
+      "model.layers.5.self_attn.gate_proj",
+      "model.layers.6.self_attn.q_proj",
+      "model.layers.6.self_attn.k_proj",
+      "model.layers.6.self_attn.v_proj",
+      "model.layers.6.self_attn.o_proj",
+      "model.layers.6.self_attn.gate_proj",
+      "model.layers.6.mlp.router.gate",
+      "model.layers.7.self_attn.q_proj",
+      "model.layers.7.self_attn.k_proj",
+      "model.layers.7.self_attn.v_proj",
+      "model.layers.7.self_attn.o_proj",
+      "model.layers.7.self_attn.gate_proj",
+      "model.layers.7.mlp.router.gate",
+      "model.layers.8.self_attn.q_proj",
+      "model.layers.8.self_attn.k_proj",
+      "model.layers.8.self_attn.v_proj",
+      "model.layers.8.self_attn.o_proj",
+      "model.layers.8.self_attn.gate_proj",
+      "model.layers.8.mlp.router.gate",
+      "model.layers.9.self_attn.q_proj",
+      "model.layers.9.self_attn.k_proj",
+      "model.layers.9.self_attn.v_proj",
+      "model.layers.9.self_attn.o_proj",
+      "model.layers.9.self_attn.gate_proj",
+      "model.layers.9.mlp.router.gate",
+      "model.layers.10.self_attn.q_proj",
+      "model.layers.10.self_attn.k_proj",
+      "model.layers.10.self_attn.v_proj",
+      "model.layers.10.self_attn.o_proj",
+      "model.layers.10.self_attn.gate_proj",
+      "model.layers.10.mlp.router.gate",
+      "model.layers.11.self_attn.q_proj",
+      "model.layers.11.self_attn.k_proj",
+      "model.layers.11.self_attn.v_proj",
+      "model.layers.11.self_attn.o_proj",
+      "model.layers.11.self_attn.gate_proj",
+      "model.layers.11.mlp.router.gate",
+      "model.layers.12.self_attn.q_proj",
+      "model.layers.12.self_attn.k_proj",
+      "model.layers.12.self_attn.v_proj",
+      "model.layers.12.self_attn.o_proj",
+      "model.layers.12.self_attn.gate_proj",
+      "model.layers.12.mlp.router.gate",
+      "model.layers.13.self_attn.q_proj",
+      "model.layers.13.self_attn.k_proj",
+      "model.layers.13.self_attn.v_proj",
+      "model.layers.13.self_attn.o_proj",
+      "model.layers.13.self_attn.gate_proj",
+      "model.layers.13.mlp.router.gate",
+      "model.layers.14.self_attn.q_proj",
+      "model.layers.14.self_attn.k_proj",
+      "model.layers.14.self_attn.v_proj",
+      "model.layers.14.self_attn.o_proj",
+      "model.layers.14.self_attn.gate_proj",
+      "model.layers.14.mlp.router.gate",
+      "model.layers.15.self_attn.q_proj",
+      "model.layers.15.self_attn.k_proj",
+      "model.layers.15.self_attn.v_proj",
+      "model.layers.15.self_attn.o_proj",
+      "model.layers.15.self_attn.gate_proj",
+      "model.layers.15.mlp.router.gate",
+      "model.layers.16.self_attn.q_proj",
+      "model.layers.16.self_attn.k_proj",
+      "model.layers.16.self_attn.v_proj",
+      "model.layers.16.self_attn.o_proj",
+      "model.layers.16.self_attn.gate_proj",
+      "model.layers.16.mlp.router.gate",
+      "model.layers.17.self_attn.q_proj",
+      "model.layers.17.self_attn.k_proj",
+      "model.layers.17.self_attn.v_proj",
+      "model.layers.17.self_attn.o_proj",
+      "model.layers.17.self_attn.gate_proj",
+      "model.layers.17.mlp.router.gate",
+      "model.layers.18.self_attn.q_proj",
+      "model.layers.18.self_attn.k_proj",
+      "model.layers.18.self_attn.v_proj",
+      "model.layers.18.self_attn.o_proj",
+      "model.layers.18.self_attn.gate_proj",
+      "model.layers.18.mlp.router.gate",
+      "model.layers.19.self_attn.q_proj",
+      "model.layers.19.self_attn.k_proj",
+      "model.layers.19.self_attn.v_proj",
+      "model.layers.19.self_attn.o_proj",
+      "model.layers.19.self_attn.gate_proj",
+      "model.layers.19.mlp.router.gate",
+      "model.layers.20.self_attn.q_proj",
+      "model.layers.20.self_attn.k_proj",
+      "model.layers.20.self_attn.v_proj",
+      "model.layers.20.self_attn.o_proj",
+      "model.layers.20.self_attn.gate_proj",
+      "model.layers.20.mlp.router.gate",
+      "model.layers.21.self_attn.q_proj",
+      "model.layers.21.self_attn.k_proj",
+      "model.layers.21.self_attn.v_proj",
+      "model.layers.21.self_attn.o_proj",
+      "model.layers.21.self_attn.gate_proj",
+      "model.layers.21.mlp.router.gate",
+      "model.layers.22.self_attn.q_proj",
+      "model.layers.22.self_attn.k_proj",
+      "model.layers.22.self_attn.v_proj",
+      "model.layers.22.self_attn.o_proj",
+      "model.layers.22.self_attn.gate_proj",
+      "model.layers.22.mlp.router.gate",
+      "model.layers.23.self_attn.q_proj",
+      "model.layers.23.self_attn.k_proj",
+      "model.layers.23.self_attn.v_proj",
+      "model.layers.23.self_attn.o_proj",
+      "model.layers.23.self_attn.gate_proj",
+      "model.layers.23.mlp.router.gate",
+      "model.layers.24.self_attn.q_proj",
+      "model.layers.24.self_attn.k_proj",
+      "model.layers.24.self_attn.v_proj",
+      "model.layers.24.self_attn.o_proj",
+      "model.layers.24.self_attn.gate_proj",
+      "model.layers.24.mlp.router.gate",
+      "model.layers.25.self_attn.q_proj",
+      "model.layers.25.self_attn.k_proj",
+      "model.layers.25.self_attn.v_proj",
+      "model.layers.25.self_attn.o_proj",
+      "model.layers.25.self_attn.gate_proj",
+      "model.layers.25.mlp.router.gate",
+      "model.layers.26.self_attn.q_proj",
+      "model.layers.26.self_attn.k_proj",
+      "model.layers.26.self_attn.v_proj",
+      "model.layers.26.self_attn.o_proj",
+      "model.layers.26.self_attn.gate_proj",
+      "model.layers.26.mlp.router.gate",
+      "model.layers.27.self_attn.q_proj",
+      "model.layers.27.self_attn.k_proj",
+      "model.layers.27.self_attn.v_proj",
+      "model.layers.27.self_attn.o_proj",
+      "model.layers.27.self_attn.gate_proj",
+      "model.layers.27.mlp.router.gate",
+      "model.layers.28.self_attn.q_proj",
+      "model.layers.28.self_attn.k_proj",
+      "model.layers.28.self_attn.v_proj",
+      "model.layers.28.self_attn.o_proj",
+      "model.layers.28.self_attn.gate_proj",
+      "model.layers.28.mlp.router.gate",
+      "model.layers.29.self_attn.q_proj",
+      "model.layers.29.self_attn.k_proj",
+      "model.layers.29.self_attn.v_proj",
+      "model.layers.29.self_attn.o_proj",
+      "model.layers.29.self_attn.gate_proj",
+      "model.layers.29.mlp.router.gate",
+      "model.layers.30.self_attn.q_proj",
+      "model.layers.30.self_attn.k_proj",
+      "model.layers.30.self_attn.v_proj",
+      "model.layers.30.self_attn.o_proj",
+      "model.layers.30.self_attn.gate_proj",
+      "model.layers.30.mlp.router.gate",
+      "model.layers.31.self_attn.q_proj",
+      "model.layers.31.self_attn.k_proj",
+      "model.layers.31.self_attn.v_proj",
+      "model.layers.31.self_attn.o_proj",
+      "model.layers.31.self_attn.gate_proj",
+      "model.layers.31.mlp.router.gate",
+      "model.layers.32.self_attn.q_proj",
+      "model.layers.32.self_attn.k_proj",
+      "model.layers.32.self_attn.v_proj",
+      "model.layers.32.self_attn.o_proj",
+      "model.layers.32.self_attn.gate_proj",
+      "model.layers.32.mlp.router.gate",
+      "model.layers.33.self_attn.q_proj",
+      "model.layers.33.self_attn.k_proj",
+      "model.layers.33.self_attn.v_proj",
+      "model.layers.33.self_attn.o_proj",
+      "model.layers.33.self_attn.gate_proj",
+      "model.layers.33.mlp.router.gate",
+      "model.layers.34.self_attn.q_proj",
+      "model.layers.34.self_attn.k_proj",
+      "model.layers.34.self_attn.v_proj",
+      "model.layers.34.self_attn.o_proj",
+      "model.layers.34.self_attn.gate_proj",
+      "model.layers.34.mlp.router.gate",
+      "model.layers.35.self_attn.q_proj",
+      "model.layers.35.self_attn.k_proj",
+      "model.layers.35.self_attn.v_proj",
+      "model.layers.35.self_attn.o_proj",
+      "model.layers.35.self_attn.gate_proj",
+      "model.layers.35.mlp.router.gate",
+      "model.layers.36.self_attn.q_proj",
+      "model.layers.36.self_attn.k_proj",
+      "model.layers.36.self_attn.v_proj",
+      "model.layers.36.self_attn.o_proj",
+      "model.layers.36.self_attn.gate_proj",
+      "model.layers.36.mlp.router.gate",
+      "model.layers.37.self_attn.q_proj",
+      "model.layers.37.self_attn.k_proj",
+      "model.layers.37.self_attn.v_proj",
+      "model.layers.37.self_attn.o_proj",
+      "model.layers.37.self_attn.gate_proj",
+      "model.layers.37.mlp.router.gate",
+      "model.layers.38.self_attn.q_proj",
+      "model.layers.38.self_attn.k_proj",
+      "model.layers.38.self_attn.v_proj",
+      "model.layers.38.self_attn.o_proj",
+      "model.layers.38.self_attn.gate_proj",
+      "model.layers.38.mlp.router.gate",
+      "model.layers.39.self_attn.q_proj",
+      "model.layers.39.self_attn.k_proj",
+      "model.layers.39.self_attn.v_proj",
+      "model.layers.39.self_attn.o_proj",
+      "model.layers.39.self_attn.gate_proj",
+      "model.layers.39.mlp.router.gate",
+      "model.layers.40.self_attn.q_proj",
+      "model.layers.40.self_attn.k_proj",
+      "model.layers.40.self_attn.v_proj",
+      "model.layers.40.self_attn.o_proj",
+      "model.layers.40.self_attn.gate_proj",
+      "model.layers.40.mlp.router.gate",
+      "model.layers.41.self_attn.q_proj",
+      "model.layers.41.self_attn.k_proj",
+      "model.layers.41.self_attn.v_proj",
+      "model.layers.41.self_attn.o_proj",
+      "model.layers.41.self_attn.gate_proj",
+      "model.layers.41.mlp.router.gate",
+      "model.layers.42.self_attn.q_proj",
+      "model.layers.42.self_attn.k_proj",
+      "model.layers.42.self_attn.v_proj",
+      "model.layers.42.self_attn.o_proj",
+      "model.layers.42.self_attn.gate_proj",
+      "model.layers.42.mlp.router.gate",
+      "model.layers.43.self_attn.q_proj",
+      "model.layers.43.self_attn.k_proj",
+      "model.layers.43.self_attn.v_proj",
+      "model.layers.43.self_attn.o_proj",
+      "model.layers.43.self_attn.gate_proj",
+      "model.layers.43.mlp.router.gate",
+      "model.layers.44.self_attn.q_proj",
+      "model.layers.44.self_attn.k_proj",
+      "model.layers.44.self_attn.v_proj",
+      "model.layers.44.self_attn.o_proj",
+      "model.layers.44.self_attn.gate_proj",
+      "model.layers.44.mlp.router.gate",
+      "model.layers.45.self_attn.q_proj",
+      "model.layers.45.self_attn.k_proj",
+      "model.layers.45.self_attn.v_proj",
+      "model.layers.45.self_attn.o_proj",
+      "model.layers.45.self_attn.gate_proj",
+      "model.layers.45.mlp.router.gate",
+      "model.layers.46.self_attn.q_proj",
+      "model.layers.46.self_attn.k_proj",
+      "model.layers.46.self_attn.v_proj",
+      "model.layers.46.self_attn.o_proj",
+      "model.layers.46.self_attn.gate_proj",
+      "model.layers.46.mlp.router.gate",
+      "model.layers.47.self_attn.q_proj",
+      "model.layers.47.self_attn.k_proj",
+      "model.layers.47.self_attn.v_proj",
+      "model.layers.47.self_attn.o_proj",
+      "model.layers.47.self_attn.gate_proj",
+      "model.layers.47.mlp.router.gate",
+      "model.layers.48.self_attn.q_proj",
+      "model.layers.48.self_attn.k_proj",
+      "model.layers.48.self_attn.v_proj",
+      "model.layers.48.self_attn.o_proj",
+      "model.layers.48.self_attn.gate_proj",
+      "model.layers.48.mlp.router.gate",
+      "model.layers.49.self_attn.q_proj",
+      "model.layers.49.self_attn.k_proj",
+      "model.layers.49.self_attn.v_proj",
+      "model.layers.49.self_attn.o_proj",
+      "model.layers.49.self_attn.gate_proj",
+      "model.layers.49.mlp.router.gate",
+      "model.layers.50.self_attn.q_proj",
+      "model.layers.50.self_attn.k_proj",
+      "model.layers.50.self_attn.v_proj",
+      "model.layers.50.self_attn.o_proj",
+      "model.layers.50.self_attn.gate_proj",
+      "model.layers.50.mlp.router.gate",
+      "model.layers.51.self_attn.q_proj",
+      "model.layers.51.self_attn.k_proj",
+      "model.layers.51.self_attn.v_proj",
+      "model.layers.51.self_attn.o_proj",
+      "model.layers.51.self_attn.gate_proj",
+      "model.layers.51.mlp.router.gate",
+      "model.layers.52.self_attn.q_proj",
+      "model.layers.52.self_attn.k_proj",
+      "model.layers.52.self_attn.v_proj",
+      "model.layers.52.self_attn.o_proj",
+      "model.layers.52.self_attn.gate_proj",
+      "model.layers.52.mlp.router.gate",
+      "model.layers.53.self_attn.q_proj",
+      "model.layers.53.self_attn.k_proj",
+      "model.layers.53.self_attn.v_proj",
+      "model.layers.53.self_attn.o_proj",
+      "model.layers.53.self_attn.gate_proj",
+      "model.layers.53.mlp.router.gate",
+      "model.layers.54.self_attn.q_proj",
+      "model.layers.54.self_attn.k_proj",
+      "model.layers.54.self_attn.v_proj",
+      "model.layers.54.self_attn.o_proj",
+      "model.layers.54.self_attn.gate_proj",
+      "model.layers.54.mlp.router.gate",
+      "model.layers.55.self_attn.q_proj",
+      "model.layers.55.self_attn.k_proj",
+      "model.layers.55.self_attn.v_proj",
+      "model.layers.55.self_attn.o_proj",
+      "model.layers.55.self_attn.gate_proj",
+      "model.layers.55.mlp.router.gate",
+      "model.layers.56.self_attn.q_proj",
+      "model.layers.56.self_attn.k_proj",
+      "model.layers.56.self_attn.v_proj",
+      "model.layers.56.self_attn.o_proj",
+      "model.layers.56.self_attn.gate_proj",
+      "model.layers.56.mlp.router.gate",
+      "model.layers.57.self_attn.q_proj",
+      "model.layers.57.self_attn.k_proj",
+      "model.layers.57.self_attn.v_proj",
+      "model.layers.57.self_attn.o_proj",
+      "model.layers.57.self_attn.gate_proj",
+      "model.layers.57.mlp.router.gate",
+      "model.layers.58.self_attn.q_proj",
+      "model.layers.58.self_attn.k_proj",
+      "model.layers.58.self_attn.v_proj",
+      "model.layers.58.self_attn.o_proj",
+      "model.layers.58.self_attn.gate_proj",
+      "model.layers.58.mlp.router.gate",
+      "model.layers.59.self_attn.q_proj",
+      "model.layers.59.self_attn.k_proj",
+      "model.layers.59.self_attn.v_proj",
+      "model.layers.59.self_attn.o_proj",
+      "model.layers.59.self_attn.gate_proj",
+      "model.layers.59.mlp.router.gate",
+      "lm_head"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.13.1.a20260205"
+  },
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000,
+  "route_norm": true,
+  "route_scale": 2.448,
+  "score_func": "sigmoid",
+  "sliding_window": 4096,
+  "tie_word_embeddings": false,
+  "topk_group": 1,
+  "transformers_version": "4.57.6",
+  "use_cache": true,
+  "use_grouped_mm": true,
+  "vocab_size": 200192
+}

configuration_afmoe.py

@@ -0,0 +1,133 @@
+# coding=utf-8
+# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+from transformers.configuration_utils import layer_type_validation
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+class AfmoeConfig(PretrainedConfig):
+    """
+    n_group (`int`, *optional*, defaults to 1):
+            Number of groups for routed experts.
+    topk_group (`int`, *optional*, defaults to 1):
+        Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
+    """
+    model_type = "afmoe"
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        num_hidden_layers: int = 32,
+        vocab_size: int = 200192,
+        hidden_size: int = 2048,
+        intermediate_size: int = 6144,
+        moe_intermediate_size=1408,
+        num_dense_layers=1,
+        num_attention_heads=16,
+        num_key_value_heads=None,
+        head_dim=128,
+        hidden_act="silu",
+        max_position_embeddings=16384,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        num_experts=64,
+        num_experts_per_tok=6,
+        num_shared_experts=2,
+        num_expert_groups=1,
+        num_limited_groups=1,
+        score_func="sigmoid",
+        route_norm=True,
+        route_scale=1.0,
+        global_attn_every_n_layers=4,
+        sliding_window=1024,
+        mup_enabled=False,
+        layer_types=None,
+        attention_dropout: float = 0.0,
+        n_group: int = 1,
+        topk_group: int = 1,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_dense_layers = num_dense_layers
+        self.num_attention_heads = num_attention_heads
+        self.head_dim = head_dim
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+
+
+        # MoE specific
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_experts_per_tok = num_experts_per_tok
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.num_experts = num_experts
+        self.num_shared_experts = num_shared_experts
+        self.num_expert_groups = num_expert_groups
+        self.num_limited_groups = num_limited_groups
+        self.score_func = score_func
+        self.route_norm = route_norm
+        self.route_scale = route_scale
+
+
+        # Attention specific
+        self.attention_dropout = attention_dropout
+        self.global_attn_every_n_layers = global_attn_every_n_layers
+        self.sliding_window = sliding_window
+        self.layer_types = layer_types
+        if self.layer_types is None:
+            self.layer_types = [
+                "sliding_attention" if bool((i + 1) % global_attn_every_n_layers) else "full_attention" for i in range(self.num_hidden_layers)
+            ]
+        layer_type_validation(self.layer_types)
+
+        # muP specific
+        self.mup_enabled = mup_enabled
+
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+
+
+        # Validate rope configs
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self)
+
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["AfmoeConfig"]

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 3,
+  "pad_token_id": 12,
+  "transformers_version": "4.57.6"
+}

model-00001-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:693a701a7007e8ad7341b9914b046765e83bd80f540f96c908746f5813321c0c
+size 4998314816

model-00002-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fecf9dbf12d82551ae335bfddf3206ca36270e6b922364b9c1b46362daa9bc60
+size 4998729232

model-00003-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de053a01d5876fa1d944cbbaf25846543acb9699868ec53589c3b5291c96b697
+size 4999630856

model-00004-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40959bba5afb11710c4f40e2c16e0ab14393dc27d90a69cfe63c56ee81359dff
+size 4998730128

model-00005-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92e4a10d1786b12a8ab2fbe05a66cf88234002169a2257f19090c873757a8eec
+size 4998732240

model-00006-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801ff854fcd03a97609f2c2a724b5556ebcb8cbf2dc5833a37b87138d366a946
+size 4999634128

model-00007-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9a5fdaf87026a7dc6d7e4a58a0a3a69035c054563c4edd0775384bf9645f289
+size 4998731568

model-00008-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b6c3c744cc56a2eef15f17e746294329b90c88492b7bf78e38a30531a395242
+size 4998732240

model-00009-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35d070c7a5bbe3afd3c082738c1e4e5ef5751620165995bffd56d0b1a710f69b
+size 4998732240

model-00010-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df441508111e55bab83cf3dba5ce22cbd024c4410df730b63340fad7eb1170eb
+size 4999633760

model-00011-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35ea938eacc58159454a79b75ec0cfc61bfc6ffb1ab951e4156259833853a5ee
+size 4998731936

model-00012-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7126093dad24a8d6f563c4635abedf8f6bafa7f9936ce3998d266bf395ac4197
+size 4998732240

model-00013-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea65d8470b2cdb61696772cefec5719fb80232cb8f9a70e722c03ff901201961
+size 4999634080

model-00014-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e147904632210fb188d8655611137a3d433e6b34d6777f3ec12f06a2e63635a
+size 4998731608

model-00015-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dc84acd99c3c823e04960b59449f23574c3d34c2fd4e6815b3e6c9d416b456f
+size 4998732248

model-00016-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c74252abfcb076f3279dee574789390f63223c467298766b6e99303813c778df
+size 4990666848

model-00017-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c51bcd53e5386578486ffb2de89fcbd399d0f81bcf36c601710183a1e7bb9960
+size 4997966232

model-00018-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e60debb067b7ea39b4b215e3c7f6174aba96bbbf8601450031a1db66266b6111
+size 4998731976

model-00019-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8dd482db83aa143e23e3b2b767d7f4dcc27290c1caef360843704c2d23edb388
+size 4998732240

model-00020-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:987efff9321824fae81eb59376469c5ccd12b46d84ae5ce1370657cae2be354d
+size 4999634040

model-00021-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ecd92aba56224baf14cec2f813b68ccedb55f565fe5a6b39ed1d67c46e4f345
+size 4998731656

model-00022-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95cedd1d652544aeb1f15e35426aa8991cb9ddce295e1370523d2b949ab7d44
+size 4998732240

model-00023-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d85320860244112f7728e3511b7ee1282a50ff155f96205d6459e5fd7339ab75
+size 4999634416

model-00024-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:930c786845bf08b66f9c23e14f5c554a76aa34009c1910c16b6cd07fa65c2b47
+size 4998731496

model-00025-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd59cf5371aa47f5c37038c763120c7a183ca9230db1f6d8fbb77db857f03973
+size 4998732024

model-00026-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2ea762ae4584844dd0375bc7d266028973da65f53af6cabf1bdb3ab60b0a036
+size 4998732240

model-00027-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c07cddea6255daf250d813485d25c7f633ffcefd3f68d2b605fd8768c2985401
+size 4999633992

model-00028-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9277029f22df1007a05e2db4318a5287b24ad8d4176d4f3a799bb943ff93627
+size 4998731704

model-00029-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b7588f284bd6a49eb10bc064f5e3d3aad9d431a372ab89364ea5738af8ccba5
+size 4998732248

model-00030-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d755bf79a36a34a4c224a6feb0c4a7f7f8613c7313011fdc8a813cdbc2bdbab6
+size 4999634360

model-00031-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b205db46a0b147621b586e8ca62551250da4badb3d86ace6fda5aa6d4274b164
+size 4998731504

model-00032-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68a00944a880628bfd8cb1d699908ec0507b5e46e917e15bc834e4864b78bc4d
+size 4998732072

model-00033-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fb54c2131ede6db0934cfd452a22407e598632950909f464edb91427fb14651
+size 4998732248

model-00034-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:942a6d5f32ee5f6810bde270856b7df7ced21913f3a0f027788d67f8573ecdb6
+size 4999633944

model-00035-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d80c093d413f50b9b0c1a315af6749ae9ea19382db6ac40b1b8ed83421bc107
+size 4998731752

model-00036-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63f599cd93b3629d261d9e3fe7af966da37e1785e6c7e88ff535be127fbfc3a8
+size 4998732240

model-00037-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d65e88ba24358e195f2df190e78e2e8d6506f03684094926a0b2d1e9a7490c27
+size 4999634264

model-00038-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2f111aef160e913926f6503302631f8fbeb2cba41885d4ced5b8577a983da3a
+size 4998731552

model-00039-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2146a723260dcef461de0bb8bf6087a57da9b1661624e1d14040330f61395310
+size 4998732120

model-00040-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f31628df84bd2b701ed6a5a6e1a3b6ffc7cfc82343b8f3e83ecd3b9a817985fa
+size 4998732240

model-00041-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6967c177261c2180738f56e6b82603254d5e6cee360c91a156c768f5f43f5960
+size 4999633896

model-00042-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:459878198b0d05c3a0048194fa378bd2bbcc7d6eb5f596327fee408f0dee22bb
+size 4998731800

model-00043-of-00043.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:328cd5141cf89ab93c824dc499e7e235b01da0336c3c95f86a7230d5d2762068
+size 3108429960

model.safetensors.index.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46a188bac954c4aa299e6749c6d41ffa4788235cdedb25ea57388795ace26db6
+size 12206302

modeling_afmoe.py

@@ -0,0 +1,680 @@
+from typing import Callable, Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import (
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel, ALL_ATTENTION_FUNCTIONS
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.masking_utils import (
+    create_causal_mask,
+    create_sliding_window_causal_mask,
+)
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.integrations import use_kernel_forward_from_hub
+
+
+try:
+    from .configuration_afmoe import AfmoeConfig
+except:
+    from configuration_afmoe import AfmoeConfig
+
+class AfmoeRotaryEmbedding(nn.Module):
+
+    def __init__(self, config: AfmoeConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    def _dynamic_frequency_update(self, position_ids, device):
+        """
+        dynamic RoPE layers should recompute `inv_freq` in the following situations:
+        1 - growing beyond the cached sequence length (allow scaling)
+        2 - the current sequence length is in the original scale (avoid losing precision with small sequences)
+        """
+        seq_len = torch.max(position_ids) + 1
+        if seq_len > self.max_seq_len_cached:  # growth
+            inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len)
+            self.register_buffer("inv_freq", inv_freq, persistent=False)  # TODO joao: may break with compilation
+            self.max_seq_len_cached = seq_len
+
+        if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len:  # reset
+            # This .to() is needed if the model has been moved to a device after being initialized (because
+            # the buffer is automatically moved, but not the original copy)
+            self.original_inv_freq = self.original_inv_freq.to(device)
+            self.register_buffer("inv_freq", self.original_inv_freq, persistent=False)
+            self.max_seq_len_cached = self.original_max_seq_len
+
+    @torch.no_grad()
+    def forward(self, x, position_ids):
+        if "dynamic" in self.rope_type:
+            self._dynamic_frequency_update(position_ids, device=x.device)
+
+        # Core RoPE block
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+        # Force float32 (see https://github.com/huggingface/transformers/pull/29285)
+        device_type = x.device.type
+        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.float().to(x.device) @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos()
+            sin = emb.sin()
+
+        # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention
+        cos = cos * self.attention_scaling
+        sin = sin * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(
+        batch, num_key_value_heads, n_rep, slen, head_dim
+    )
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+@use_kernel_forward_from_hub("RMSNorm")
+class AfmoeRMSNorm(nn.Module):
+    def __init__(self, hidden_size: int, eps: float):
+        """
+        AfmoeRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs,
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(
+        query.dtype
+    )
+    attn_weights = nn.functional.dropout(
+        attn_weights, p=dropout, training=module.training
+    )
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+class AfmoeMLP(nn.Module):
+    def __init__(self, config, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = intermediate_size or config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class AfmoeTokenChoiceRouter(nn.Module):
+    """Token-choice top-K router for MoE routing."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.num_experts = config.num_experts
+        self.score_func = config.score_func
+        self.route_norm = config.route_norm
+        self.route_scale = config.route_scale
+        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)     
+
+    def forward(self, hidden_states, expert_bias: torch.Tensor | None):
+        _, _, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+
+        scores = self.gate(hidden_states)
+
+        # Apply scoring function in float32 for stability
+        if self.score_func == "sigmoid":
+            scores = torch.sigmoid(scores.to(torch.float32))
+        else:
+            scores = F.softmax(scores.to(torch.float32), dim=-1)
+
+        if expert_bias is not None:
+            _, selected_experts = torch.topk(scores + expert_bias, k=self.top_k, dim=1)
+            top_scores = scores.gather(dim=1, index=selected_experts)
+        else:
+            top_scores, selected_experts = torch.topk(scores, k=self.top_k, dim=1)
+
+        # Normalize weights if using sigmoid
+        if self.score_func == "sigmoid" and self.route_norm:
+            denominator = top_scores.sum(dim=-1, keepdim=True) + 1e-20
+            top_scores = top_scores / denominator
+
+        top_scores = top_scores * self.route_scale
+        return top_scores, selected_experts
+
+class AfmoeMoE(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.router = AfmoeTokenChoiceRouter(config)
+
+        self.shared_experts = None
+        if config.num_shared_experts > 0:
+            self.shared_experts = AfmoeMLP(
+                config, config.moe_intermediate_size * config.num_shared_experts
+            )
+        self.experts = nn.ModuleList(
+            [AfmoeMLP(
+                config, intermediate_size=config.moe_intermediate_size
+            ) for _ in range(config.num_experts)]
+        )
+        self.expert_bias = nn.Parameter(torch.zeros(config.num_experts, dtype=torch.float32), requires_grad=False)
+        
+
+    def forward(self, hidden_states):
+        batch_size, seq_len, hidden_dim = hidden_states.shape
+        hidden_states_flat = hidden_states.view(-1, hidden_dim)
+
+        # Get routing decisions
+        top_scores, selected_experts = self.router(hidden_states, self.expert_bias)
+
+        # Process through shared experts
+        if self.shared_experts is not None:
+            shared_output = self.shared_experts(hidden_states_flat)
+        else:
+            shared_output = torch.zeros_like(hidden_states_flat)
+
+        # Reorder tokens by expert for efficient processing
+        token_indices_sorted = torch.argsort(selected_experts.view(-1), stable=True)
+        top_scores_sorted = top_scores.view(-1)[token_indices_sorted]
+        token_to_expert = selected_experts.view(-1)[token_indices_sorted]
+        token_indices_sorted = token_indices_sorted // self.config.num_experts_per_tok
+
+        # Gather input tokens
+        token_indices_expanded = token_indices_sorted.unsqueeze(-1).expand(
+            -1, hidden_dim
+        )
+        routed_input = torch.gather(
+            hidden_states_flat, dim=0, index=token_indices_expanded
+        )
+
+        routed_output = torch.zeros_like(routed_input)
+        for expert_id in range(self.config.num_experts):
+            mask = token_to_expert == expert_id
+            if mask.any():
+                expert_input = routed_input[mask]
+                expert_out = self.experts[expert_id](expert_input)
+                routed_output[mask] = expert_out
+          
+        routed_output = (
+            routed_output.to(torch.float32) * top_scores_sorted.unsqueeze(-1)
+        ).to(hidden_states.dtype)
+
+        # Scatter back to original positions
+        output = shared_output.scatter_add(
+            dim=0, index=token_indices_expanded, src=routed_output
+        )
+
+        return output.view(batch_size, seq_len, hidden_dim)
+
+
+class AfmoeAttention(nn.Module):
+    """Multi-headed attention with local/global pattern and gating."""
+
+    def __init__(self, config: AfmoeConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_heads = config.num_attention_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_local_attention = config.layer_types[layer_idx] == "sliding_attention"
+        self.sliding_window = config.sliding_window if self.is_local_attention else None
+
+        self.q_proj = nn.Linear(
+            config.hidden_size, self.num_heads * self.head_dim, bias=False
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False
+        )
+        self.v_proj = nn.Linear(
+            config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False
+        )
+        self.o_proj = nn.Linear(
+            self.num_heads * self.head_dim, config.hidden_size, bias=False
+        )
+
+        self.q_norm = AfmoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+        self.k_norm = AfmoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+
+        self.gate_proj = nn.Linear(
+            config.hidden_size, self.num_heads * self.head_dim, bias=False
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:   
+
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape)
+        key_states = self.k_proj(hidden_states).view(hidden_shape)
+        value_states = self.v_proj(hidden_states).view(hidden_shape)
+        gate_states = self.gate_proj(hidden_states)
+
+        query_states = self.q_norm(query_states)
+        key_states = self.k_norm(key_states)
+        
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        if self.is_local_attention:
+            cos, sin = position_embeddings
+            query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_value is not None:
+            cache_kwargs = {"cache_position": cache_position}
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[
+                self.config._attn_implementation
+            ]
+
+        output, _ = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask=attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            sliding_window=self.sliding_window,
+            **kwargs,
+        )
+
+        output = output.view(*input_shape, -1).contiguous()
+        output = output * F.sigmoid(gate_states)
+        return self.o_proj(output)
+
+
+class AfmoeDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: AfmoeConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.layer_idx = layer_idx
+
+        self.self_attn = AfmoeAttention(config=config, layer_idx=layer_idx)
+        self.attention_type = config.layer_types[layer_idx]
+
+        # Dual normalization for attention
+        self.input_layernorm = AfmoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = AfmoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        # Dual normalization for FFN
+        self.pre_mlp_layernorm = AfmoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_mlp_layernorm = AfmoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        # MoE or dense FFN
+        self.moe_enabled = layer_idx >= config.num_dense_layers
+        if self.moe_enabled:
+            self.mlp = AfmoeMoE(config)
+        else:
+            self.mlp = AfmoeMLP(config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.FloatTensor:
+        residual = hidden_states
+
+        # Self Attention with dual normalization
+        hidden_states = self.input_layernorm(hidden_states)
+        hidden_states = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = residual + hidden_states
+
+        # FFN with dual normalization
+        residual = hidden_states
+        hidden_states = self.pre_mlp_layernorm(hidden_states)
+
+        if self.moe_enabled:
+            hidden_states = self.mlp(hidden_states)
+        else:
+            hidden_states = self.mlp(hidden_states)
+
+        hidden_states = self.post_mlp_layernorm(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+class AfmoePreTrainedModel(PreTrainedModel):
+    config_class = AfmoeConfig
+    base_model_prefix = "model"
+    _no_split_modules = ["AfmoeDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _keep_in_fp32_modules = [
+        "input_layernorm",
+        "post_attention_layernorm",
+        "pre_mlp_layernorm",
+        "post_mlp_layernorm",
+        "q_norm",
+        "k_norm",
+        "norm",
+    ]
+    _supports_sdpa = True
+    _supports_attention_backend = True
+    supports_gradient_checkpointing = True
+
+
+class AfmoeModel(AfmoePreTrainedModel):
+    _no_split_modules = ["AfmoeDecoderLayer"]
+
+    def __init__(self, config: AfmoeConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(
+            config.vocab_size, config.hidden_size, self.padding_idx
+        )
+        self.layers = nn.ModuleList(
+            [
+                AfmoeDecoderLayer(config, layer_idx)
+                for layer_idx in range(config.num_hidden_layers)
+            ]
+        )
+        self.norm = AfmoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = AfmoeRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError(
+                "You must specify exactly one of input_ids or inputs_embeds"
+            )
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if cache_position is None:
+            past_seen_tokens = (
+                past_key_values.get_seq_length() if past_key_values is not None else 0
+            )
+            cache_position = torch.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+                device=inputs_embeds.device,
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        # It may already have been prepared by e.g. `generate`
+        if not isinstance(causal_mask_mapping := attention_mask, dict):
+            mask_kwargs = {
+                "config": self.config,
+                "input_embeds": inputs_embeds,
+                "attention_mask": attention_mask,
+                "cache_position": cache_position,
+                "past_key_values": past_key_values,
+            }
+            causal_mask_mapping = {
+                "full_attention": create_causal_mask(**mask_kwargs),
+                "sliding_attention": create_sliding_window_causal_mask(**mask_kwargs),
+            }
+
+        hidden_states = inputs_embeds
+
+        # Apply muP input scaling if enabled
+        if self.config.mup_enabled:
+            hidden_states = hidden_states * (self.config.hidden_size**0.5)
+
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        for decoder_layer in self.layers:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask_mapping[decoder_layer.attention_type],
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+class AfmoeForCausalLM(AfmoePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = AfmoeModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        token_type_ids: Optional[torch.Tensor] = None,  # will be ignored
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> Union[Tuple, MoeCausalLMOutputWithPast]:
+        outputs: MoeModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits
+        slice_indices = (
+            slice(-logits_to_keep, None)
+            if isinstance(logits_to_keep, int)
+            else logits_to_keep
+        )
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
+
+
+        return MoeCausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )
+
+
+__all__ = [
+    "AfmoeForCausalLM",
+    "AfmoeModel",
+    "AfmoePreTrainedModel",
+]