Create generate.py

generate.py

@@ -0,0 +1,15 @@
+import torch
+
+def generate(model, model_inputs, generation_config, **kwargs)
+    cur_length = model_inputs["input_ids"].shape[1]
+    max_length = generation_config.max_length or cur_length + generation_config.max_new_tokens
+    
+    while cur_length < max_length:
+        logits = model(model_inputs["input_ids"]).logits
+        next_token_logits = logits[:, -1, :]
+        next_tokens = torch.argmax(next_token_logits)
+        model_inputs["input_ids"] = torch.cat((model_inputs["input_ids"], next_tokens), dim=-1)
+        cur_length += 1
+
+    return model_inputs["input_ids"]
+