Update modeling_i3.py

modeling_i3.py

@@ -1,6 +1,6 @@
 # modeling_i3.py
-import torch
 from transformers import PreTrainedModel
+from transformers.modeling_outputs import CausalLMOutputWithPast
 from configuration_i3 import I3Config
 from i3_architecture import i3Model  # your actual i3 implementation
 
@@ -18,18 +18,26 @@ class I3ForCausalLM(PreTrainedModel):
             rank=config.rank,
             d_state=config.d_state,
         )
+        self.lm_head = torch.nn.Linear(config.d_model, config.vocab_size, bias=False)
         self.post_init()
 
-    def forward(self, input_ids, labels=None):
-        logits, loss = self.model(input_ids, labels)
-        return {"loss": loss, "logits": logits}
+    def forward(self, input_ids, labels=None, attention_mask=None, **kwargs):
+        outputs = self.model(input_ids)
+        logits = self.lm_head(outputs)
+        
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            loss_fct = torch.nn.CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), 
+                           shift_labels.view(-1))
 
-    @torch.no_grad()
-    def generate(self, input_ids, max_new_tokens=50, temperature=1.0, top_k=None):
-        return self.model.generate(input_ids, max_new_tokens, temperature, top_k)
-
-# AutoClass registration (optional but recommended)
-from transformers import AutoConfig, AutoModelForCausalLM
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+        )
 
-AutoConfig.register("i3", I3Config)
-AutoModelForCausalLM.register(I3Config, I3ForCausalLM)
+    def prepare_inputs_for_generation(self, input_ids, **kwargs):
+        return {"input_ids": input_ids}