Update app.py

app.py

@@ -8,8 +8,8 @@ import lightning as L
 import torch
 
 # support running without installing as a package
-wd = Path(__file__).parent.parent.resolve()
-sys.path.append(str(wd))
+# wd = Path(__file__).parent.parent.resolve()
+# sys.path.append(str(wd))
 
 from generate import generate
 from lit_llama import Tokenizer
@@ -32,7 +32,7 @@ torch.set_float32_matmul_precision("high")
 def model_load(
     adapter_path: Path = Path("out/adapter/alpaca/lit-llama-adapter-finetuned_15k.pth"),
     pretrained_path: Path = Path("checkpoints/lit-llama/7B/lit-llama.pth"),
-    quantize: Optional[str] = None,
+    quantize: Optional[str] = "llm.int8",
 ):
 
     fabric = L.Fabric(devices=1)
@@ -110,11 +110,12 @@ def instruct_generate(
 # 配置具体参数
 # pretrained_path = hf_hub_download(
 #     repo_id="Gary3410/pretrain_lit_llama", filename="lit-llama.pth")
-tokenizer_path = hf_hub_download(
-    repo_id="Gary3410/pretrain_lit_llama", filename="tokenizer.model")
-adapter_path = hf_hub_download(
-    repo_id="Gary3410/pretrain_lit_llama", filename="lit-llama-adapter-finetuned_15k.pth")
-# adapter_path = "lit-llama-adapter-finetuned_15k.pth"
+# tokenizer_path = hf_hub_download(
+#     repo_id="Gary3410/pretrain_lit_llama", filename="tokenizer.model")
+# adapter_path = hf_hub_download(
+#     repo_id="Gary3410/pretrain_lit_llama", filename="lit-llama-adapter-finetuned_15k.pth")
+adapter_path = "lit-llama-adapter-finetuned_15k.pth"
+tokenizer_path = "tokenizer.model"
 pretrained_path = "lit-llama.pth"
 example_path = "example.json"
 # 1024如果不够, 调整为512