Upload patch_models.py

patch_models.py

@@ -1,4 +1,4 @@
-"""Patch backend: replace AVAILABLE_MODELS with OpenRouter models + routing."""
+"""Patch backend: OpenRouter routing + DeepSeek V4 Flash 0731."""
 
 import ast
 import os
@@ -6,61 +6,36 @@ import re
 
 AGENT_FILE = "/app/backend/routes/agent.py"
 LLM_PARAMS_FILE = "/app/agent/core/llm_params.py"
-
-# === Step 1: Patch model_ids.py ===
 IDS_FILE = "/app/agent/core/model_ids.py"
+
 with open(IDS_FILE) as f:
     content = f.read()
 
-# --- Remove any previous NO_TOOLS_MODELS addition first (clean slate) ---
+# Remove NO_TOOLS_MODELS if present from previous patches
 if "NO_TOOLS_MODELS" in content:
     content = re.sub(r'NO_TOOLS_MODELS\s*=\s*\{[^}]*\}\n?', '', content)
     print("OK: Removed old NO_TOOLS_MODELS")
 
-# Swap out old model definitions for new ones
+# Fix model IDs for OpenRouter
 content = content.replace(
     'KIMI_K27_CODE_MODEL_ID = "moonshotai/Kimi-K2.7-Code:novita"',
-    'KIMI_K27_CODE_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro"',
+    'KIMI_K27_CODE_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro"'
 )
 content = content.replace(
     'MINIMAX_M3_MODEL_ID = "MiniMaxAI/MiniMax-M3:novita"',
-    'MINIMAX_M3_MODEL_ID = "deepseek-ai/DeepSeek-V4-Flash"',
+    'MINIMAX_M3_MODEL_ID = "deepseek-ai/DeepSeek-V4-Flash"'
 )
 content = content.replace(
     'GLM_52_MODEL_ID = "zai-org/GLM-5.2:novita"',
-    'GLM_52_MODEL_ID = "openai/deepseek/deepseek-v4-flash"',
+    'GLM_52_MODEL_ID = "openai/deepseek/deepseek-v4-flash"'
 )
 content = content.replace(
     'DEEPSEEK_V4_PRO_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:novita"',
-    'DEEPSEEK_V4_PRO_MODEL_ID = "nvidia/nemotron-3-super-120b-a12b:free"',
-)
-
-# Replace all existing deepseek model IDs with 0731 version
-content = content.replace(
-    '~deepseek/deepseek-v4-flash-latest',
-    '~deepseek/deepseek-v4-flash-0731'
-)
-content = content.replace(
-    'deepseek/deepseek-v4-flash-latest',
-    '~deepseek/deepseek-v4-flash-0731'
-)
-content = content.replace(
-    'deepseek/deepseek-v4-flash',
-    'deepseek/deepseek-v4-flash'  # keep as-is
+    'DEEPSEEK_V4_PRO_MODEL_ID = "nvidia/nemotron-3-super-120b-a12b:free"'
 )
 
-# Add missing model ID constants if not present
-if "GEMMA_4_31B_FREE_MODEL_ID" not in content:
-    # Find where model constants end (after DEEPSEEK_V4_PRO_MODEL_ID)
-    pos = content.find('DEEPSEEK_V4_PRO_MODEL_ID')
-    if pos >= 0:
-        # Find the next line break after this constant definition
-        next_newline = content.find('\n', pos)
-        if next_newline >= 0:
-            before = content[:next_newline + 1]
-            after = content[next_newline + 1:]
-            new_constants = '''
-TENCENT_HY3_FREE_MODEL_ID = "openai/tencent/hy3:free"
+# Add missing constants before DEEPSEEK_V4_PRO_MODEL_ID line
+insert_before = '''TENCENT_HY3_FREE_MODEL_ID = "openai/tencent/hy3:free"
 GEMMA_4_31B_FREE_MODEL_ID = "openai/google/gemma-4-31b-it:free"
 LLAMA_3_3_70B_FREE_MODEL_ID = "openai/meta-llama/llama-3.3-70b-instruct:free"
 LLAMA_3_1_8B_MODEL_ID = "openai/meta-llama/llama-3.1-8b-instruct"
@@ -69,21 +44,20 @@ LAGUNA_S21_FREE_MODEL_ID = "openai/poolside/laguna-s-2.1:free"
 NEX_N2_MINI_MODEL_ID = "openai/nex-agi/nex-n2-mini"
 LING_3_0_FLASH_FREE_MODEL_ID = "openai/inclusionai/ling-3.0-flash:free"
 '''
-            content = before + new_constants + after
+if "TENCENT_HY3_FREE_MODEL_ID" not in content:
+    content = content.replace('DEEPSEEK_V4_PRO_MODEL_ID', insert_before + 'DEEPSEEK_V4_PRO_MODEL_ID')
+
+# Replace all references to latest deepseek models with 0731 version
+content = content.replace('"~deepseek/deepseek-v4-flash-latest"', '"deepseek/deepseek-v4-flash-0731"')
+content = content.replace('"~deepseek/deepseek-v4-flash-0731"', '"deepseek/deepseek-v4-flash-0731"')
 
 with open(IDS_FILE, "w") as f:
     f.write(content)
 print("OK: model_ids.py patched")
 
-# Validate syntax
-try:
-    ast.parse(content)
-    print("OK: model_ids.py syntax OK")
-except SyntaxError as e:
-    print(f"FAIL: model_ids.py syntax error: {e}")
-    raise
+ast.parse(content)  # validate syntax
+print("OK: model_ids.py syntax OK")
 
-# === Step 2: Patch agent.py ===
 with open(AGENT_FILE) as f:
     content = f.read()
 
@@ -96,7 +70,7 @@ content = content.replace(
     "DEFAULT_GPT_MODEL_ID = LAGUNA_S21_FREE_MODEL_ID"
 )
 
-# Replace import block
+# Update imports
 old_import = (
     "from agent.core.model_ids import (\n"
     "    CLAUDE_OPUS_48_MODEL_ID,\n"
@@ -128,25 +102,24 @@ new_import = (
 )
 content = content.replace(old_import, new_import)
 
-# Replace _available_models function
-new_func = (
-    "def _available_models() -> list[dict[str, Any]]:\n"
-    "    models = [\n"
-    '        {"id": TENCENT_HY3_FREE_MODEL_ID, "label": "Tencent HY3:free", "recommended": True},\n'
-    '        {"id": GEMMA_4_31B_FREE_MODEL_ID, "label": "Gemma 4 31B:free", "recommended": True},\n'
-    '        {"id": LLAMA_3_3_70B_FREE_MODEL_ID, "label": "Llama 3.3 70B:free", "recommended": True},\n'
-    '        {"id": LAGUNA_M1_FREE_MODEL_ID, "label": "Laguna M.1:free", "recommended": True},\n'
-    '        {"id": DEFAULT_GPT_MODEL_ID, "label": "Laguna S 2.1:free", "recommended": True},\n'
-    '        {"id": NEX_N2_MINI_MODEL_ID, "label": "nex-agi/nex-n2-mini", "recommended": True},\n'
-    '        {"id": LING_3_0_FLASH_FREE_MODEL_ID, "label": "inclusionai/ling-3.0-flash:free", "recommended": True},\n'
-    '        {"id": LLAMA_3_1_8B_MODEL_ID, "label": "Llama 3.1 8B"},\n'
-    '        {"id": KIMI_K27_CODE_MODEL_ID, "label": "DeepSeek V4 Pro"},\n'
-    '        {"id": MINIMAX_M3_MODEL_ID, "label": "DeepSeek V4 Flash"},\n'
-    '        {"id": DEFAULT_MODEL_ID, "label": "DeepSeek V4 Flash"},\n'
-    '        {"id": DEEPSEEK_V4_PRO_MODEL_ID, "label": "Nemotron 3 Super 120B"},\n'
-    "    ]\n"
-    "    return models"
-)
+# Update _available_models to include DeepSeek V4 Flash 0731
+new_func = '''def _available_models() -> list[dict[str, Any]]:
+    models = [
+        {"id": TENCENT_HY3_FREE_MODEL_ID, "label": "Tencent HY3:free", "recommended": True},
+        {"id": GEMMA_4_31B_FREE_MODEL_ID, "label": "Gemma 4 31B:free", "recommended": True},
+        {"id": LLAMA_3_3_70B_FREE_MODEL_ID, "label": "Llama 3.3 70B:free", "recommended": True},
+        {"id": LAGUNA_M1_FREE_MODEL_ID, "label": "Laguna M.1:free", "recommended": True},
+        {"id": DEFAULT_GPT_MODEL_ID, "label": "Laguna S 2.1:free", "recommended": True},
+        {"id": NEX_N2_MINI_MODEL_ID, "label": "nex-agi/nex-n2-mini", "recommended": True},
+        {"id": LING_3_0_FLASH_FREE_MODEL_ID, "label": "inclusionai/ling-3.0-flash:free", "recommended": True},
+        {"id": LLAMA_3_1_8B_MODEL_ID, "label": "Llama 3.1 8B"},
+        {"id": KIMI_K27_CODE_MODEL_ID, "label": "DeepSeek V4 Pro"},
+        {"id": MINIMAX_M3_MODEL_ID, "label": "DeepSeek V4 Flash"},
+        {"id": DEFAULT_MODEL_ID, "label": "DeepSeek V4 Flash"},
+        {"id": DEEPSEEK_V4_PRO_MODEL_ID, "label": "Nemotron 3 Super 120B"},
+        {"id": "deepseek/deepseek-v4-flash-0731", "label": "DeepSeek V4 Flash 0731", "recommended": True},
+    ]
+    return models'''
 
 func_match = re.search(
     r"def _available_models\(\)\s*->\s*list\[dict\[str,\s*Any\]\]:.*?return models",
@@ -154,7 +127,7 @@ func_match = re.search(
 )
 if func_match:
     content = content[:func_match.start()] + new_func + content[func_match.end():]
-    print("OK: Replaced _available_models()")
+    print("OK: Replaced _available_models() with DeepSeek V4 Flash 0731")
 else:
     old_func = (
         "def _available_models() -> list[dict[str, Any]]:\n"
@@ -172,15 +145,16 @@ else:
         content = content.replace(old_func, new_func)
         print("OK: String replace of _available_models()")
     else:
-        print("WARN: Cannot find _available_models(), skipping")
+        print("WARN: Could not find _available_models() pattern, searching...")
+        idx = content.find("def _available_models")
+        if idx >= 0:
+            print(f"Found at index {idx}, context: {content[idx:idx+400]}")
 
-# Update title gen model
 content = content.replace(
     '"openai/gpt-oss-120b:cerebras",',
     '"huggingface/deepseek-ai/DeepSeek-V4-Pro",'
 )
 
-# Validate syntax
 try:
     ast.parse(content)
     print("OK: agent.py syntax OK")
@@ -190,9 +164,8 @@ except SyntaxError as e:
 
 with open(AGENT_FILE, "w") as f:
     f.write(content)
-print("OK: agent.py patched")
 
-# === Step 3: Patch _resolve_llm_params ===
+# Patch llm_params for OpenRouter routing
 with open(LLM_PARAMS_FILE) as f:
     llm_content = f.read()
 
@@ -201,6 +174,13 @@ if "normalized_model.startswith" not in llm_content:
     if api_key_find in llm_content:
         line_end = llm_content.find("\n", llm_content.find(api_key_find) + len(api_key_find)) + 1
         routing_insert = (
+            '    # Route deepseek/ models to OpenRouter\n'
+            '    if normalized_model.startswith("deepseek/"):\n'
+            '        return {\n'
+            '            "model": normalized_model,\n'
+            '            "api_base": "https://openrouter.ai/api/v1",\n'
+            '            "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "",\n'
+            '        }\n\n'
             '    # Route openai/-prefixed models to OpenRouter\n'
             '    if normalized_model.startswith("openai/"):\n'
             '        return {\n'
@@ -210,7 +190,7 @@ if "normalized_model.startswith" not in llm_content:
             '        }\n\n'
         )
         llm_content = llm_content[:line_end] + routing_insert + llm_content[line_end:]
-        print("OK: Patched _resolve_llm_params")
+        print("OK: Patched _resolve_llm_params with deepseek/ routing")
     else:
         print("WARN: Could not find api_key line")
 else: