Upload folder using huggingface_hub

config.json

@@ -6,12 +6,12 @@
     "hidden_size": 768,
     "int8": false,
     "lora_target_modules": null,
-    "max_batch_size": 16,
+    "max_batch_size": 32,
     "max_beam_width": 1,
     "max_draft_len": 0,
-    "max_input_len": 512,
+    "max_input_len": 1024,
     "max_num_tokens": null,
-    "max_output_len": 200,
+    "max_output_len": 2048,
     "max_position_embeddings": 1024,
     "max_prompt_embedding_table_size": 0,
     "name": "gpt",

gpt_float16_tp1_rank0.engine

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f92cab6ffe048accc1475d3187147c664daa820d31c09fa42e8c0fc54f1a5dc3
-size 328056100
+oid sha256:50ffa47bfbe312d8d243e97995a83e0494c312d03d0ba18dd2577b7b9a684e24
+size 328061308