Fix: float causal mask, weight tying, attention_mask, config aliases

modeling_konkan.py

@@ -117,4 +117,31 @@ class KonkanGPT(PreTrainedModel, GenerationMixin):
         return CausalLMOutput(loss=loss, logits=logits)
 
     def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs):
-        return {"input_ids": input_ids, "attention_mask": attention_mask}
+        return {"input_ids": input_ids, "attention_mask": attention_mask}
+    
+    def chat(self, tokenizer, query, max_new_tokens=200, temperature=0.7):
+        device = next(self.parameters()).device
+        
+        # 1. Apply Template Inbuilt
+        messages = [{"role": "user", "content": query}]
+        prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+        
+        # 2. Tokenize Inbuilt
+        inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).to(device)
+        
+        # 3. Generate
+        outputs = self.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            do_sample=True,
+            temperature=temperature,
+            top_p=0.9,
+            repetition_penalty=1.2,
+            pad_token_id=tokenizer.pad_token_id,
+            eos_token_id=tokenizer.eos_token_id
+        )
+        
+        # 4. Clean and Return
+        decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        response = decoded.split("assistant")[-1].strip()
+        return response.replace("|>", "").strip()