Update app_modules/utils.py

app_modules/utils.py

@@ -102,8 +102,6 @@ def convert_mdtext(md_text):
             else:
                 result.append(mdtex2html.convert(non_code, extensions=["tables"]))
         if code.strip():
-            # _, code = detect_language(code)  # 暂时去除代码高亮功能，因为在大段代码的情况下会出现问题
-            # code = code.replace("\n\n", "\n") # 暂时去除代码中的空行，因为在大段代码的情况下会出现问题
             code = f"\n```{code}\n\n```"
             code = markdown_to_html_with_syntax_highlight(code)
             result.append(code)
@@ -152,10 +150,10 @@ def convert_to_markdown(text):
     for line in lines:
         if in_code_block is False and line.startswith("```"):
             in_code_block = True
-            markdown_text += "```\n"
+            markdown_text += f"{line}\n"
         elif in_code_block is True and line.startswith("```"):
             in_code_block = False
-            markdown_text += "```\n"
+            markdown_text += f"{line}\n"
         elif in_code_block:
             markdown_text += f"{line}\n"
         else:
@@ -321,7 +319,7 @@ def is_stop_word_or_prefix(s: str, stop_words: list) -> bool:
 
 
 
-def load_tokenizer_and_model(base_model,adapter_model,load_8bit=False):
+def load_tokenizer_and_model(base_model,adapter_model=None,load_8bit=False):
     if torch.cuda.is_available():
         device = "cuda"
     else:
@@ -340,32 +338,35 @@ def load_tokenizer_and_model(base_model,adapter_model,load_8bit=False):
             torch_dtype=torch.float16,
             device_map="auto",
         )
-        model = PeftModel.from_pretrained(
-            model,
-            adapter_model,
-            torch_dtype=torch.float16,
-        )
+        if adapter_model is not None:
+            model = PeftModel.from_pretrained(
+                model,
+                adapter_model,
+                torch_dtype=torch.float16,
+            )
     elif device == "mps":
         model = LlamaForCausalLM.from_pretrained(
             base_model,
             device_map={"": device},
             torch_dtype=torch.float16,
         )
-        model = PeftModel.from_pretrained(
-            model,
-            adapter_model,
-            device_map={"": device},
-            torch_dtype=torch.float16,
-        )
+        if adapter_model is not None:
+            model = PeftModel.from_pretrained(
+                model,
+                adapter_model,
+                device_map={"": device},
+                torch_dtype=torch.float16,
+            )
     else:
         model = LlamaForCausalLM.from_pretrained(
             base_model, device_map={"": device}, low_cpu_mem_usage=True
         )
-        model = PeftModel.from_pretrained(
-            model,
-            adapter_model,
-            device_map={"": device},
-        )
+        if adapter_model is not None:
+            model = PeftModel.from_pretrained(
+                model,
+                adapter_model,
+                device_map={"": device},
+            )
 
     if not load_8bit:
         model.half()  # seems to fix bugs for some users.