update

mlc-chat-config.json

@@ -22,9 +22,9 @@
     "max_batch_size": 80
   },
   "vocab_size": 32064,
-  "context_window_size": 4096,
+  "context_window_size": 2048,
   "sliding_window_size": -1,
-  "prefill_chunk_size": 2048,
+  "prefill_chunk_size": 1024,
   "attention_sink_size": -1,
   "tensor_parallel_shards": 1,
   "pipeline_parallel_stages": 1,
@@ -86,4 +86,4 @@
     32001,
     32007
   ]
-}
+}