Update raven_modeling_minimal.py

Update to support 4.48 cache changes.

raven_modeling_minimal.py

@@ -1,4 +1,4 @@
-"""Minimal modeling.py file for HF compatibility and funny zero-shot experiments. Use only for inference."""
+"""Minimal modeling.py file for HF compatibility and funny zero-shot experiments. Best used for inference, finetuning should work, but is untested with this implementation."""
 
 import torch
 import math
@@ -627,7 +627,7 @@ class RavenForCausalLM(RavenPreTrainedModel, GenerationMixin):
         model_inputs["cache_position"] = cache_position
         current_input_length = input_ids.shape[1]
         if past_key_values is not None:
-            if type(past_key_values) == DynamicCache:
+            if type(past_key_values) != HuginnDynamicCache:
                 # Need to use custom cache, detect and replace HF dynamic cache if generate injects it
                 assert past_key_values.get_seq_length() == 0
                 past_key_values = HuginnDynamicCache()