app.py

@@ -2,17 +2,18 @@
 # pylint: disable=invalid-name, line-too-long,broad-exception-caught, protected-access
 import os
 import time
+from pathlib import Path
+
+import torch
+from loguru import logger
+from transformers import AutoModel, AutoTokenizer
 
 # ruff: noqa: E402
 # os.system("pip install --upgrade torch transformers sentencepiece scipy cpm_kernels accelerate bitsandbytes loguru")
 
 # os.system("pip install torch transformers sentencepiece loguru")
 
-from pathlib import Path
 
-import torch
-from loguru import logger
-from transformers import AutoModel, AutoTokenizer
 
 # fix timezone in Linux
 os.environ["TZ"] = "Asia/Shanghai"
@@ -50,13 +51,15 @@ logger.debug("done load")
 # tokenizer = AutoTokenizer.from_pretrained("openchat/openchat_v2_w")
 # model = AutoModelForCausalLM.from_pretrained("openchat/openchat_v2_w", load_in_8bit_fp32_cpu_offload=True, load_in_8bit=True)
 
-# model_path = model.config._dict["model_name_or_path"]
-# logger.debug(f"{model.config=} {type(model.config)=} {model_path=}")
-logger.debug(f"{model.config=}, {type(model.config)=} ")
+# locate model file cache
+cache_loc = Path("~/.cache/huggingface/hub").expanduser()
+model_cache_path = [elm for elm in Path(cache_loc).rglob("*") if Path(model_name).name in elm.as_posix() and "pytorch_model.bin" in elm.as_posix()]
 
-# model_size_gb = Path(model_path).stat().st_size / 2**30
+logger.debug(f"{model_cache_path=}")
 
-# logger.info(f"{model_name=} {model_size_gb=:.2f} GB")
+if model_cache_path:
+    model_size_gb = Path(model_cache_path).stat().st_size / 2**30
+    logger.info(f"{model_name=} {model_size_gb=:.2f} GB")
 
 # with gr.Blocks() as demo:
 #     chatbot = gr.Chatbot()