sys prompt fix

Author: DataBoySu

README.md

@@ -1,4 +1,3 @@
-<!--b-->
 <div align="center">
   <a href="README.md">🇺🇸 English</a> |
   <a href="locales/README.de.md">🇩🇪 Deutsch</a> |
@@ -14,16 +13,15 @@
 <div style="text-align:center; margin:18px 0;">
   <img src="monitor/api/static/logo.png" alt="MyGPU logo"/>
 </div>
-<!--e-->
+
 > *MyGPU: Lightweight GPU Management Utility: a compact `nvidia-smi` wrapper with an elegant web dashboard.*
-<!-- b-->
 
 ![License](https://img.shields.io/badge/license-MIT-blue.svg)
 ![Python](https://img.shields.io/badge/python-3.10%2B-blue)
 ![Version](https://img.shields.io/badge/version-1.2.3-blue)
 ![Platform](https://img.shields.io/badge/platform-Windows-lightgrey)
 ![cuda 12.x](https://img.shields.io/badge/CUDA-12.x-0f9d58?logo=nvidia)
-<!--e-->
+
 ## Gallery
 
 <details>

scripts/translate_eastern.py

@@ -27,28 +27,33 @@
 with open(README_PATH, "r", encoding="utf-8") as f:
     original_text = f.read()
 
-# --- PRE-PROCESSING ---
+# --- UNIVERSAL PRE-PROCESSING ---
 protected_blocks = []
 
 def protect_match(match):
-    placeholder = f"__PB_{len(protected_blocks)}__"
+    placeholder = f"[[PB_{len(protected_blocks)}]]"
     protected_blocks.append(match.group(0))
     return placeholder
 
 text_to_translate = original_text
-text_to_translate = re.sub(r'(<!--\s*b\s*-->.*?<!--\s*e\s*-->)', protect_match, text_to_translate, flags=re.DOTALL)
 
+# 1. Protect ALL Code Blocks (Triple Backticks)
+text_to_translate = re.sub(r'(```.*?```)', protect_match, text_to_translate, flags=re.DOTALL)
+
+# 2. Protect ALL HTML tags (e.g., <div...>, <img...>, <a...>)
+text_to_translate = re.sub(r'(<[^>]+>)', protect_match, text_to_translate)
+
+# 3. Protect ALL Markdown Images and Badges
+text_to_translate = re.sub(r'(!\[[^\]]*\]\([^\)]+\))', protect_match, text_to_translate)
 # Specialized Prompt for CJK/Eastern Languages
 prompt = f"""<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>
 You are a professional technical {target_lang_name} translator.
 Your task is to translate the README into {target_lang_name}.
-Keep the markdown format and HTML tags the same.
+Keep the markdown and html tag element in original format.
 CRITICAL INSTRUCTIONS:
-1. **Placeholders**: Keep `__PB_0__`, `__PB_1__` etc. exactly as is. These are protected blocks.
-2. **Preserve HTML**: Do NOT translate or modify any HTML tags (lines starting with `<`). Output them exactly as they are in the source.
-3. **Preserve Images**: Do NOT translate or modify any Markdown images or badges (lines starting with `![`). Output them exactly as they are in the source. Do NOT add spaces inside URLs.
-4. **Translation**: Translate only the text content (paragraphs, headers, lists) into {target_lang_name}.
-5. **Technical Terms**: Keep terms like GPU, CLI, VRAM, SSH, Docker, API, CUDA in English.
+1. **Technical Preservation**: Keep all product names and technical jargon in English.
+2. **System Tags**: Return any text like [[PB_X]] exactly as is. Do not translate or change brackets.
+3. **Accuracy**: Maintain 100% technical context.
 6. **Context**: 
    - 'Enforcement' = Policy restriction (e.g., JA: 制限/強制).
    - 'Headless' = Server without GUI/display.
@@ -72,11 +77,6 @@ def protect_match(match):
         lines = lines[:-1]
     translated_content = "\n".join(lines).strip()
 
-# Restore Protected Blocks
-for i, block in enumerate(protected_blocks):
-    placeholder = f"__PB_{i}__"
-    translated_content = translated_content.replace(placeholder, block)
-
 # 2. Path Correction
 # Prepend ../ to relative paths
 translated_content = re.sub(r'(\[.*?\]\()(?!(?:http|/|#|\.\./))', r'\1../', translated_content)

scripts/translate_western.py

@@ -28,28 +28,33 @@
     original_text = f.read()
 
 # --- PRE-PROCESSING ---
+# --- UNIVERSAL PRE-PROCESSING ---
 protected_blocks = []
 
 def protect_match(match):
-    placeholder = f"__PB_{len(protected_blocks)}__"
+    placeholder = f"[[PB_{len(protected_blocks)}]]"
     protected_blocks.append(match.group(0))
     return placeholder
 
 text_to_translate = original_text
-text_to_translate = re.sub(r'(<!--\s*b\s*-->.*?<!--\s*e\s*-->)', protect_match, text_to_translate, flags=re.DOTALL)
 
+# 1. Protect ALL Code Blocks (Triple Backticks)
+text_to_translate = re.sub(r'(```.*?```)', protect_match, text_to_translate, flags=re.DOTALL)
+
+# 2. Protect ALL HTML tags (e.g., <div...>, <img...>, <a...>)
+text_to_translate = re.sub(r'(<[^>]+>)', protect_match, text_to_translate)
+
+# 3. Protect ALL Markdown Images and Badges
+text_to_translate = re.sub(r'(!\[[^\]]*\]\([^\)]+\))', protect_match, text_to_translate)
+
+# --- REFINED AGNOSTIC PROMPT ---
 prompt = f"""<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>
-You are a professional technical {target_lang_name} translator.
-Your task is to translate the README into {target_lang_name}.
-Keep the markdown format and HTML tags the same.
-CRITICAL INSTRUCTIONS:
-1. **Placeholders**: Keep `__PB_0__`, `__PB_1__` etc. exactly as is. These are protected blocks.
-2. **Preserve HTML**: Do NOT translate or modify any HTML tags (lines starting with `<`). Output them exactly as they are in the source.
-3. **Preserve Images**: Do NOT translate or modify any Markdown images or badges (lines starting with `![`). Output them exactly as they are in the source.
-4. **Translation**: Translate only the text content (paragraphs, headers, lists) into {target_lang_name}.
-5. **Technical Terms**: Keep terms like GPU, CLI, VRAM, SSH, Docker, API, CUDA in English.
-6. **No Conversational Text**: Output only the final Markdown file content. No code fences.
-<|END_OF_TURN_TOKEN|>
+You are a professional technical translator. Translate the provided text into professional {target_lang_name}.
+RULES:
+1. **Technical Preservation**: Keep all product names and technical jargon in English.
+2. **System Tags**: Return any text like [[PB_X]] exactly as is. Do not translate or change brackets.
+3. **Accuracy**: Maintain 100% technical context.
+4. **No Talk**: Output ONLY the translated Markdown.<|END_OF_TURN_TOKEN|>
 <|START_OF_TURN_TOKEN|><|USER_TOKEN|>
 {text_to_translate}<|END_OF_TURN_TOKEN|>
 <|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"""
@@ -67,11 +72,6 @@ def protect_match(match):
         lines = lines[:-1]
     translated_content = "\n".join(lines).strip()
 
-# Restore Protected Blocks
-for i, block in enumerate(protected_blocks):
-    placeholder = f"__PB_{i}__"
-    translated_content = translated_content.replace(placeholder, block)
-
 # 2. Path Correction
 # Prepend ../ to relative paths
 translated_content = re.sub(r'(\[.*?\]\()(?!(?:http|/|#|\.\./))', r'\1../', translated_content)