cyberes
/
local-llm-server
Archived
1
1
Fork 0
Code Issues Pull Requests Packages Projects Releases 1 Wiki Activity

test

This commit is contained in:
Cyberes 2023-10-02 22:45:07 -06:00
parent aed5db4968
commit cd325216e2
2 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
llm_server/llm/openai/oai_to_vllm.py
View File

@ -26,6 +26,7 @@ def oai_to_vllm(request_json_body, hashes: bool, mode):
def format_oai_err(err_msg): def format_oai_err(err_msg):
print('OAI ERROR MESSAGE:', err_msg)
return jsonify({ return jsonify({
"error": { "error": {
"message": err_msg, "message": err_msg,

3
llm_server/routes/openai_request_handler.py
View File

@ -25,6 +25,7 @@ class OpenAIRequestHandler(RequestHandler):
self.prompt = None self.prompt = None
def handle_request(self) -> Tuple[flask.Response, int]: def handle_request(self) -> Tuple[flask.Response, int]:
print('recieved request')
assert not self.used assert not self.used
if opts.openai_silent_trim: if opts.openai_silent_trim:
@ -66,11 +67,13 @@ class OpenAIRequestHandler(RequestHandler):
model = self.request_json_body.get('model') model = self.request_json_body.get('model')
if success: if success:
print('sent success response')
return self.build_openai_response(self.prompt, backend_response.json['results'][0]['text'], model=model), backend_response_status_code return self.build_openai_response(self.prompt, backend_response.json['results'][0]['text'], model=model), backend_response_status_code
else: else:
return backend_response, backend_response_status_code return backend_response, backend_response_status_code
def handle_ratelimited(self, do_log: bool = True): def handle_ratelimited(self, do_log: bool = True):
print('OAI ratelimited:', self.client_ip, self.request.headers)
_, default_backend_info = get_model_choices() _, default_backend_info = get_model_choices()
w = int(default_backend_info['estimated_wait']) if default_backend_info['estimated_wait'] > 0 else 2 w = int(default_backend_info['estimated_wait']) if default_backend_info['estimated_wait'] > 0 else 2
response = jsonify({ response = jsonify({