summaryrefslogtreecommitdiffstats
path: root/tool/vercel.py
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--tool/vercel.py103
1 files changed, 103 insertions, 0 deletions
diff --git a/tool/vercel.py b/tool/vercel.py
new file mode 100644
index 00000000..7b87e298
--- /dev/null
+++ b/tool/vercel.py
@@ -0,0 +1,103 @@
+import json
+import re
+from typing import Any
+
+import quickjs
+from curl_cffi import requests
+
+session = requests.Session(impersonate="chrome107")
+
+
+def get_model_info() -> dict[str, Any]:
+ url = "https://sdk.vercel.ai"
+ response = session.get(url)
+ html = response.text
+ paths_regex = r"static\/chunks.+?\.js"
+ separator_regex = r'"\]\)<\/script><script>self\.__next_f\.push\(\[.,"'
+
+ paths = re.findall(paths_regex, html)
+ paths = [re.sub(separator_regex, "", path) for path in paths]
+ paths = list(set(paths))
+
+ urls = [f"{url}/_next/{path}" for path in paths]
+ scripts = [session.get(url).text for url in urls]
+
+ for script in scripts:
+ models_regex = r'let .="\\n\\nHuman:\",r=(.+?),.='
+ matches = re.findall(models_regex, script)
+
+ if matches:
+ models_str = matches[0]
+ stop_sequences_regex = r"(?<=stopSequences:{value:\[)\D(?<!\])"
+ models_str = re.sub(
+ stop_sequences_regex, re.escape('"\\n\\nHuman:"'), models_str
+ )
+
+ context = quickjs.Context() # type: ignore
+ json_str: str = context.eval(f"({models_str})").json() # type: ignore
+ return json.loads(json_str) # type: ignore
+
+ return {}
+
+
+def convert_model_info(models: dict[str, Any]) -> dict[str, Any]:
+ model_info: dict[str, Any] = {}
+ for model_name, params in models.items():
+ default_params = params_to_default_params(params["parameters"])
+ model_info[model_name] = {"id": params["id"], "default_params": default_params}
+ return model_info
+
+
+def params_to_default_params(parameters: dict[str, Any]):
+ defaults: dict[str, Any] = {}
+ for key, parameter in parameters.items():
+ if key == "maximumLength":
+ key = "maxTokens"
+ defaults[key] = parameter["value"]
+ return defaults
+
+
+def get_model_names(model_info: dict[str, Any]):
+ model_names = model_info.keys()
+ model_names = [
+ name
+ for name in model_names
+ if name not in ["openai:gpt-4", "openai:gpt-3.5-turbo"]
+ ]
+ model_names.sort()
+ return model_names
+
+
+def print_providers(model_names: list[str]):
+ for name in model_names:
+ split_name = re.split(r":|/", name)
+ base_provider = split_name[0]
+ variable_name = split_name[-1].replace("-", "_").replace(".", "")
+ line = f'{variable_name} = Model(name="{name}", base_provider="{base_provider}", best_provider=Vercel,)\n'
+ print(line)
+
+
+def print_convert(model_names: list[str]):
+ for name in model_names:
+ split_name = re.split(r":|/", name)
+ key = split_name[-1]
+ variable_name = split_name[-1].replace("-", "_").replace(".", "")
+ # "claude-instant-v1": claude_instant_v1,
+ line = f' "{key}": {variable_name},'
+ print(line)
+
+
+def main():
+ model_info = get_model_info()
+ model_info = convert_model_info(model_info)
+ print(json.dumps(model_info, indent=2))
+
+ model_names = get_model_names(model_info)
+ print("-------" * 40)
+ print_providers(model_names)
+ print("-------" * 40)
+ print_convert(model_names)
+
+
+if __name__ == "__main__":
+ main()