diff --git a/benchmarks/analyze_sharegpt.py b/benchmarks/analyze_sharegpt.py index 4394dbae..19f90e05 100644 --- a/benchmarks/analyze_sharegpt.py +++ b/benchmarks/analyze_sharegpt.py @@ -54,7 +54,8 @@ def tokens_in_input_str(s): return return_val convo_numbers = [] - loaded_share_gpt = json.load(open('/home/hanq/data/ShareGPT_V3_unfiltered_cleaned_split.json', 'r')) + # Please update with your own data file path + loaded_share_gpt = json.load(open('~/data/ShareGPT_V3_unfiltered_cleaned_split.json', 'r')) for example in loaded_share_gpt: if len(example['conversations']) < 2: continue diff --git a/run_server.py b/run_server.py index d6ad06d7..05e7ed09 100644 --- a/run_server.py +++ b/run_server.py @@ -68,7 +68,7 @@ def main(argv: Sequence[str]): os.environ['XLA_FLAGS'] = '--xla_dump_to=/tmp/xla_logs --xla_dump_hlo_as_text' # No devices for local cpu test. A None for prefill and a None for generate. devices = server_lib.get_devices() - print('HERE 1') + print(f"devices: {devices}") engine = je.create_pytorch_engine( devices=devices, tokenizer_path=_TOKENIZER_PATH.value, @@ -85,7 +85,7 @@ def main(argv: Sequence[str]): interleaved_slices=(_PLATFORM.value, ), interleaved_engine_create_fns=(lambda a: engine, ), ) - print('HERE 2') + print(f"server_config: {server_config}") # We separate credential from run so that we can unit test it with local credentials. # TODO: Add grpc credentials for OSS. @@ -95,7 +95,7 @@ def main(argv: Sequence[str]): config=server_config, devices=devices, ) - print('HANQ....') + print("Started jetstream_server....") jetstream_server.wait_for_termination()