bep40 commited on
Commit
53ddae0
·
verified ·
1 Parent(s): 5e3cda1

Upload patch_models.py

Browse files
Files changed (1) hide show
  1. patch_models.py +90 -15
patch_models.py CHANGED
@@ -1,43 +1,120 @@
1
- """Patch backend: replace model IDs + add OpenRouter routing for openai/-prefixed models."""
2
 
3
  import ast
4
  import os
 
5
 
 
6
  LLM_PARAMS_FILE = "/app/agent/core/llm_params.py"
7
 
8
- # === Step 1: Patch model_ids.py — add new model constants ===
9
  IDS_FILE = "/app/agent/core/model_ids.py"
10
  with open(IDS_FILE, "r") as f:
11
  content = f.read()
12
 
13
  old_models_block = 'CLAUDE_OPUS_48_MODEL_ID = "anthropic/claude-opus-4.8:fal-ai"\nGPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"\nKIMI_K27_CODE_MODEL_ID = "moonshotai/Kimi-K2.7-Code:novita"\nMINIMAX_M3_MODEL_ID = "MiniMaxAI/MiniMax-M3:novita"\nGLM_52_MODEL_ID = "zai-org/GLM-5.2:novita"\nDEEPSEEK_V4_PRO_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:novita"'
14
 
15
- new_models_block = 'GPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"\nKIMI_K27_CODE_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro"\nMINIMAX_M3_MODEL_ID = "deepseek-ai/DeepSeek-V4-Flash"\nGLM_52_MODEL_ID = "openai/deepseek/deepseek-v4-flash"\nDEEPSEEK_V4_PRO_MODEL_ID = "nvidia/nemotron-3-super-120b-a12b:free"\n\n# === PATCH: New model IDs ===\nGEMMA_4_31B_FREE_MODEL_ID = "openai/google/gemma-4-31b-it:free"\nTENCENT_HY3_FREE_MODEL_ID = "openai/tencent/hy3:free"\nLLAMA_3_3_70B_FREE_MODEL_ID = "openai/meta-llama/llama-3.3-70b-instruct:free"\nLLAMA_3_1_8B_MODEL_ID = "openai/meta-llama/llama-3.1-8b-instruct"'
16
 
17
  content = content.replace(old_models_block, new_models_block)
18
 
19
  old_hosted = "HOSTED_MODEL_IDS = {\n CLAUDE_OPUS_48_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GLM_52_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n}"
20
- new_hosted = "HOSTED_MODEL_IDS = {\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GLM_52_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n GEMMA_4_31B_FREE_MODEL_ID,\n TENCENT_HY3_FREE_MODEL_ID,\n LLAMA_3_3_70B_FREE_MODEL_ID,\n LLAMA_3_1_8B_MODEL_ID,\n}"
21
  content = content.replace(old_hosted, new_hosted)
22
 
23
  with open(IDS_FILE, "w") as f:
24
  f.write(content)
25
  print("OK: model_ids.py patched")
26
 
27
- # === Step 2: Patch agent.py — expand AVAILABLE_MODELS ===
28
- AGENT_FILE = "/app/backend/routes/agent.py"
29
- with open(AGENT_FILE, "r") as f:
30
  content = f.read()
31
 
 
 
 
 
 
 
 
 
 
 
32
  # Update imports
33
  old_import = "from agent.core.model_ids import (\n CLAUDE_OPUS_48_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n GLM_52_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n strip_huggingface_model_prefix,\n)"
34
- new_import = "from agent.core.model_ids import (\n DEEPSEEK_V4_PRO_MODEL_ID,\n GLM_52_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GEMMA_4_31B_FREE_MODEL_ID,\n TENCENT_HY3_FREE_MODEL_ID,\n LLAMA_3_3_70B_FREE_MODEL_ID,\n LLAMA_3_1_8B_MODEL_ID,\n strip_huggingface_model_prefix,\n)"
35
  content = content.replace(old_import, new_import)
36
 
37
- # Replace _available_models() — use openai/ prefix for OpenRouter routing
38
- old_available = 'def _available_models() -> list[dict[str, Any]]:\n models = [\n {\n "id": CLAUDE_OPUS_48_MODEL_ID,\n "label": "Claude Opus 4.8",\n },\n {\n "id": DEFAULT_GPT_MODEL_ID,\n "label": "GPT-5.5",\n },\n {\n "id": KIMI_K27_CODE_MODEL_ID,\n "label": "Kimi K2.7 Code",\n },\n {\n "id": MINIMAX_M3_MODEL_ID,\n "label": "MiniMax M3",\n },\n {\n "id": DEFAULT_MODEL_ID,\n "label": "GLM 5.2",\n "recommended": True,\n },\n {\n "id": DEEPSEEK_V4_PRO_MODEL_ID,\n "label": "DeepSeek V4 Pro",\n },\n ]\n return models'
39
- new_available = 'def _available_models() -> list[dict[str, Any]]:\n models = [\n {\n "id": GEMMA_4_31B_FREE_MODEL_ID,\n "label": "Gemma 4 31B:free",\n "recommended": True,\n },\n {\n "id": LLAMA_3_3_70B_FREE_MODEL_ID,\n "label": "Llama 3.3 70B:free",\n "recommended": True,\n },\n {\n "id": LLAMA_3_1_8B_MODEL_ID,\n "label": "Llama 3.1 8B",\n },\n {\n "id": TENCENT_HY3_FREE_MODEL_ID,\n "label": "Tencent HY3:free",\n },\n {\n "id": DEFAULT_GPT_MODEL_ID,\n "label": "GPT-5.5",\n },\n {\n "id": KIMI_K27_CODE_MODEL_ID,\n "label": "DeepSeek V4 Pro",\n },\n {\n "id": MINIMAX_M3_MODEL_ID,\n "label": "DeepSeek V4 Flash",\n },\n {\n "id": DEFAULT_MODEL_ID,\n "label": "DeepSeek V4 Flash",\n },\n {\n "id": DEEPSEEK_V4_PRO_MODEL_ID,\n "label": "Nemotron 3 Super 120B",\n },\n ]\n return models'
40
- content = content.replace(old_available, new_available)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
 
42
  # Update title generation
43
  old_title = '"openai/gpt-oss-120b:cerebras",'
@@ -56,11 +133,9 @@ with open(AGENT_FILE, "w") as f:
56
  print("OK: agent.py patched")
57
 
58
  # === Step 3: Patch _resolve_llm_params for OpenRouter routing ===
59
- # Route ALL openai/-prefixed models to OpenRouter
60
  with open(LLM_PARAMS_FILE) as f:
61
  llm_content = f.read()
62
 
63
- # Find where normalized_model is used and insert routing
64
  if "normalized_model.startswith" not in llm_content:
65
  api_key_find = "api_key = _resolve_hf_router_token(session_hf_token)"
66
  if api_key_find in llm_content:
@@ -78,7 +153,7 @@ if "normalized_model.startswith" not in llm_content:
78
  llm_content = llm_content[:line_end] + routing_insert + llm_content[line_end:]
79
  print("OK: Patched _resolve_llm_params (catch-all openai/ -> OR)")
80
  else:
81
- print("WARN: Could not find api_key line in llm_params.py")
82
  else:
83
  print("OK: llm_params.py already patched")
84
 
 
1
+ """Patch backend: replace AVAILABLE_MODELS with OpenRouter models + routing."""
2
 
3
  import ast
4
  import os
5
+ import re
6
 
7
+ AGENT_FILE = "/app/backend/routes/agent.py"
8
  LLM_PARAMS_FILE = "/app/agent/core/llm_params.py"
9
 
10
+ # === Step 1: Patch model_ids.py ===
11
  IDS_FILE = "/app/agent/core/model_ids.py"
12
  with open(IDS_FILE, "r") as f:
13
  content = f.read()
14
 
15
  old_models_block = 'CLAUDE_OPUS_48_MODEL_ID = "anthropic/claude-opus-4.8:fal-ai"\nGPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"\nKIMI_K27_CODE_MODEL_ID = "moonshotai/Kimi-K2.7-Code:novita"\nMINIMAX_M3_MODEL_ID = "MiniMaxAI/MiniMax-M3:novita"\nGLM_52_MODEL_ID = "zai-org/GLM-5.2:novita"\nDEEPSEEK_V4_PRO_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:novita"'
16
 
17
+ new_models_block = 'GPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"\nKIMI_K27_CODE_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro"\nMINIMAX_M3_MODEL_ID = "deepseek-ai/DeepSeek-V4-Flash"\nGLM_52_MODEL_ID = "openai/deepseek/deepseek-v4-flash"\nDEEPSEEK_V4_PRO_MODEL_ID = "nvidia/nemotron-3-super-120b-a12b:free"\n\n# === PATCH: OpenRouter models ===\nGEMMA_4_31B_FREE_MODEL_ID = "openai/google/gemma-4-31b-it:free"\nTENCENT_HY3_FREE_MODEL_ID = "openai/tencent/hy3:free"\nLLAMA_3_3_70B_FREE_MODEL_ID = "openai/meta-llama/llama-3.3-70b-instruct:free"\nLLAMA_3_1_8B_MODEL_ID = "openai/meta-llama/llama-3.1-8b-instruct"\nLAGUNA_M1_FREE_MODEL_ID = "openai/poolside/laguna-m.1:free"'
18
 
19
  content = content.replace(old_models_block, new_models_block)
20
 
21
  old_hosted = "HOSTED_MODEL_IDS = {\n CLAUDE_OPUS_48_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GLM_52_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n}"
22
+ new_hosted = "HOSTED_MODEL_IDS = {\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GLM_52_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n GEMMA_4_31B_FREE_MODEL_ID,\n TENCENT_HY3_FREE_MODEL_ID,\n LLAMA_3_3_70B_FREE_MODEL_ID,\n LLAMA_3_1_8B_MODEL_ID,\n LAGUNA_M1_FREE_MODEL_ID,\n}"
23
  content = content.replace(old_hosted, new_hosted)
24
 
25
  with open(IDS_FILE, "w") as f:
26
  f.write(content)
27
  print("OK: model_ids.py patched")
28
 
29
+ # === Step 2: Patch agent.py — replace _available_models() via regex ===
30
+ with open(AGENT_FILE) as f:
 
31
  content = f.read()
32
 
33
+ # Make DEFAULT_MODEL_ID and DEFAULT_GPT_MODEL_ID point to our models
34
+ content = content.replace(
35
+ "DEFAULT_MODEL_ID = GLM_52_MODEL_ID",
36
+ 'DEFAULT_MODEL_ID = "openai/tencent/hy3:free"'
37
+ )
38
+ content = content.replace(
39
+ "DEFAULT_GPT_MODEL_ID = GPT_55_MODEL_ID",
40
+ 'DEFAULT_GPT_MODEL_ID = "openai/gpt-oss-120b"'
41
+ )
42
+
43
  # Update imports
44
  old_import = "from agent.core.model_ids import (\n CLAUDE_OPUS_48_MODEL_ID,\n DEEPSEEK_V4_PRO_MODEL_ID,\n GLM_52_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n strip_huggingface_model_prefix,\n)"
45
+ new_import = "from agent.core.model_ids import (\n DEEPSEEK_V4_PRO_MODEL_ID,\n GLM_52_MODEL_ID,\n GPT_55_MODEL_ID,\n KIMI_K27_CODE_MODEL_ID,\n MINIMAX_M3_MODEL_ID,\n GEMMA_4_31B_FREE_MODEL_ID,\n TENCENT_HY3_FREE_MODEL_ID,\n LLAMA_3_3_70B_FREE_MODEL_ID,\n LLAMA_3_1_8B_MODEL_ID,\n LAGUNA_M1_FREE_MODEL_ID,\n strip_huggingface_model_prefix,\n)"
46
  content = content.replace(old_import, new_import)
47
 
48
+ # Replace _available_models() using regex (giống control space)
49
+ func_pattern = re.compile(
50
+ r'def _available_models\(\) -> list\[dict\[str, Any\]\]:\s*\n'
51
+ r'\s+models\s*=\s*\[.*?\]\s*\n\s+return models',
52
+ re.DOTALL
53
+ )
54
+
55
+ new_func = '''def _available_models() -> list[dict[str, Any]]:
56
+ models = [
57
+ {
58
+ "id": TENCENT_HY3_FREE_MODEL_ID,
59
+ "label": "Tencent HY3:free",
60
+ "recommended": True,
61
+ },
62
+ {
63
+ "id": GEMMA_4_31B_FREE_MODEL_ID,
64
+ "label": "Gemma 4 31B:free",
65
+ "recommended": True,
66
+ },
67
+ {
68
+ "id": LLAMA_3_3_70B_FREE_MODEL_ID,
69
+ "label": "Llama 3.3 70B:free",
70
+ "recommended": True,
71
+ },
72
+ {
73
+ "id": LAGUNA_M1_FREE_MODEL_ID,
74
+ "label": "Laguna M.1:free",
75
+ "recommended": True,
76
+ },
77
+ {
78
+ "id": LLAMA_3_1_8B_MODEL_ID,
79
+ "label": "Llama 3.1 8B",
80
+ },
81
+ {
82
+ "id": DEFAULT_GPT_MODEL_ID,
83
+ "label": "GPT-5.5",
84
+ },
85
+ {
86
+ "id": KIMI_K27_CODE_MODEL_ID,
87
+ "label": "DeepSeek V4 Pro",
88
+ },
89
+ {
90
+ "id": MINIMAX_M3_MODEL_ID,
91
+ "label": "DeepSeek V4 Flash",
92
+ },
93
+ {
94
+ "id": DEFAULT_MODEL_ID,
95
+ "label": "DeepSeek V4 Flash",
96
+ },
97
+ {
98
+ "id": DEEPSEEK_V4_PRO_MODEL_ID,
99
+ "label": "Nemotron 3 Super 120B",
100
+ },
101
+ ]
102
+ return models'''
103
+
104
+ func_match = func_pattern.search(content)
105
+ if func_match:
106
+ content = content[:func_match.start()] + new_func + content[func_match.end():]
107
+ print("OK: Replaced _available_models() via regex")
108
+ else:
109
+ print("WARN: Regex failed, trying string replace...")
110
+ old_available = 'def _available_models() -> list[dict[str, Any]]:\n models = [\n {\n "id": CLAUDE_OPUS_48_MODEL_ID,\n "label": "Claude Opus 4.8",\n },\n {\n "id": DEFAULT_GPT_MODEL_ID,\n "label": "GPT-5.5",\n },\n {\n "id": KIMI_K27_CODE_MODEL_ID,\n "label": "Kimi K2.7 Code",\n },\n {\n "id": MINIMAX_M3_MODEL_ID,\n "label": "MiniMax M3",\n },\n {\n "id": DEFAULT_MODEL_ID,\n "label": "GLM 5.2",\n "recommended": True,\n },\n {\n "id": DEEPSEEK_V4_PRO_MODEL_ID,\n "label": "DeepSeek V4 Pro",\n },\n ]\n return models'
111
+ if old_available in content:
112
+ content = content.replace(old_available, new_func)
113
+ print("OK: String replace worked")
114
+ else:
115
+ print("FAIL: Cannot find _available_models()!")
116
+ import sys
117
+ sys.exit(1)
118
 
119
  # Update title generation
120
  old_title = '"openai/gpt-oss-120b:cerebras",'
 
133
  print("OK: agent.py patched")
134
 
135
  # === Step 3: Patch _resolve_llm_params for OpenRouter routing ===
 
136
  with open(LLM_PARAMS_FILE) as f:
137
  llm_content = f.read()
138
 
 
139
  if "normalized_model.startswith" not in llm_content:
140
  api_key_find = "api_key = _resolve_hf_router_token(session_hf_token)"
141
  if api_key_find in llm_content:
 
153
  llm_content = llm_content[:line_end] + routing_insert + llm_content[line_end:]
154
  print("OK: Patched _resolve_llm_params (catch-all openai/ -> OR)")
155
  else:
156
+ print("WARN: Could not find api_key line")
157
  else:
158
  print("OK: llm_params.py already patched")
159