Upload DogeForCausalLM

config.json

@@ -1,43 +1,43 @@
-{
-  "_name_or_path": "SmallDoge/Doge-60M",
-  "architectures": [
-    "DogeForCausalLM"
-  ],
-  "attention_dropout": 0.0,
-  "auto_map": {
-    "AutoConfig": "configuration_doge.DogeConfig",
-    "AutoModelForCausalLM": "modeling_doge.DogeForCausalLM"
-  },
-  "bos_token_id": 0,
-  "dynamic_mask_ratio": 0.0,
-  "eos_token_id": 1,
-  "expert_retrieval_size": 256,
-  "hidden_act": "silu",
-  "hidden_bias": false,
-  "hidden_dropout": 0.0,
-  "hidden_size": 512,
-  "initializer_range": 0.02,
-  "intermediate_size": 1024,
-  "is_moe": false,
-  "max_position_embeddings": 2048,
-  "model_type": "doge",
-  "num_attention_heads": 4,
-  "num_cdmoe_experts": 16348,
-  "num_cdmoe_experts_per_head": 8,
-  "num_cdmoe_heads": 4,
-  "num_hidden_layers": 16,
-  "num_key_value_heads": 2,
-  "pad_token_id": 2,
-  "rms_norm_eps": 1e-06,
-  "rope_scaling": {
-    "factor": 4.0,
-    "original_max_position_embeddings": 2048,
-    "rope_type": "dynamic"
-  },
-  "rope_theta": 10000.0,
-  "tie_word_embeddings": true,
-  "torch_dtype": "float32",
-  "transformers_version": "4.48.3",
-  "use_cache": true,
-  "vocab_size": 32768
-}
+{
+  "_name_or_path": "SmallDoge/Doge-60M",
+  "architectures": [
+    "DogeForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_doge.DogeConfig",
+    "AutoModelForCausalLM": "modeling_doge.DogeForCausalLM"
+  },
+  "bos_token_id": 0,
+  "dynamic_mask_ratio": 0.0,
+  "eos_token_id": 1,
+  "expert_retrieval_size": 256,
+  "hidden_act": "silu",
+  "hidden_bias": false,
+  "hidden_dropout": 0.0,
+  "hidden_size": 512,
+  "initializer_range": 0.02,
+  "intermediate_size": 1024,
+  "is_moe": false,
+  "max_position_embeddings": 2048,
+  "model_type": "doge",
+  "num_attention_heads": 4,
+  "num_cdmoe_experts": 16348,
+  "num_cdmoe_experts_per_head": 8,
+  "num_cdmoe_heads": 4,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 2,
+  "pad_token_id": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": {
+    "factor": 4.0,
+    "original_max_position_embeddings": 2048,
+    "rope_type": "dynamic"
+  },
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.3",
+  "use_cache": true,
+  "vocab_size": 32768
+}

modeling_doge.py

@@ -864,27 +864,10 @@ class DogeModel(DogePreTrainedModel):
         past_key_values: Cache,
         output_attentions: bool,
     ):
-        if self.config._attn_implementation == "flash_attention_2":
-            if attention_mask is not None and (attention_mask == 0.0).any():
-                return attention_mask
-            return None
-
-        # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in
-        # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail
-        # to infer the attention mask.
+        # We have to provide attention_mask for dynamic mask computation
         past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
         using_static_cache = isinstance(past_key_values, StaticCache)
 
-        # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward
-        if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions:
-            if AttentionMaskConverter._ignore_causal_mask_sdpa(
-                attention_mask,
-                inputs_embeds=input_tensor,
-                past_key_values_length=past_seen_tokens,
-                is_training=self.training,
-            ):
-                return None
-
         dtype, device = input_tensor.dtype, input_tensor.device
         sequence_length = input_tensor.shape[1]
         if using_static_cache: