Support configurable OpenAI reasoning effort
This commit is contained in:
1 parent
00038cd1ae
commit
70ed09ad2b
2 files changed
+8
No files matched your search
@@ -132,6 +132,10 @@ class LLMSession:
|
|||||||
self.max_retries = max(0, int(cfg.get('max_retries', 2)))
|
self.max_retries = max(0, int(cfg.get('max_retries', 2)))
|
||||||
self.connect_timeout = max(1, int(cfg.get('connect_timeout', 10)))
|
self.connect_timeout = max(1, int(cfg.get('connect_timeout', 10)))
|
||||||
self.read_timeout = max(5, int(cfg.get('read_timeout', 120)))
|
self.read_timeout = max(5, int(cfg.get('read_timeout', 120)))
|
||||||
|
effort = cfg.get('reasoning_effort')
|
||||||
|
effort = None if effort is None else str(effort).strip().lower()
|
||||||
|
self.reasoning_effort = effort if effort in ['none', 'minimal','low', 'medium', 'high', 'xhigh'] else None
|
||||||
|
if effort and self.reasoning_effort is None: print(f"[WARN] Invalid reasoning_effort {effort!r}, ignored.")
|
||||||
mode = str(cfg.get('api_mode', 'chat_completions')).strip().lower().replace('-', '_')
|
mode = str(cfg.get('api_mode', 'chat_completions')).strip().lower().replace('-', '_')
|
||||||
if mode in ["responses", "response"]: self.api_mode = "responses"
|
if mode in ["responses", "response"]: self.api_mode = "responses"
|
||||||
else: self.api_mode = "chat_completions"
|
else: self.api_mode = "chat_completions"
|
||||||
@@ -176,9 +180,11 @@ class LLMSession:
|
|||||||
if self.api_mode == "responses":
|
if self.api_mode == "responses":
|
||||||
url = auto_make_url(self.api_base, "responses")
|
url = auto_make_url(self.api_base, "responses")
|
||||||
payload = {"model": model, "input": self._to_responses_input(messages), "temperature": temperature, "stream": True}
|
payload = {"model": model, "input": self._to_responses_input(messages), "temperature": temperature, "stream": True}
|
||||||
|
if self.reasoning_effort: payload["reasoning"] = {"effort": self.reasoning_effort}
|
||||||
else:
|
else:
|
||||||
url = auto_make_url(self.api_base, "chat/completions")
|
url = auto_make_url(self.api_base, "chat/completions")
|
||||||
payload = {"model": model, "messages": messages, "temperature": temperature, "stream": True}
|
payload = {"model": model, "messages": messages, "temperature": temperature, "stream": True}
|
||||||
|
if self.reasoning_effort: payload["reasoning_effort"] = self.reasoning_effort
|
||||||
for attempt in range(self.max_retries + 1):
|
for attempt in range(self.max_retries + 1):
|
||||||
streamed_any = False
|
streamed_any = False
|
||||||
try:
|
try:
|
||||||
|
|||||||
@@ -4,6 +4,8 @@ oai_config = {
|
|||||||
'apibase':"http://243.55.19.137:2001",
|
'apibase':"http://243.55.19.137:2001",
|
||||||
'model':"openai/gpt-5.1",
|
'model':"openai/gpt-5.1",
|
||||||
'api_mode':"chat_completions", # optional: "chat_completions" | "responses"
|
'api_mode':"chat_completions", # optional: "chat_completions" | "responses"
|
||||||
|
# 'reasoning_effort': "low", # optional: none | low | medium | high | xhigh;
|
||||||
|
# Only OPENAI models support this parameter, and it may not be supported by all models. It is used to control the amount of reasoning effort the model should use when generating a response. Higher values may result in more accurate and detailed responses, but may also take longer to generate.
|
||||||
'max_retries': 2, # optional: retries for 429/timeout/5xx
|
'max_retries': 2, # optional: retries for 429/timeout/5xx
|
||||||
'connect_timeout': 10, # optional: seconds
|
'connect_timeout': 10, # optional: seconds
|
||||||
'read_timeout': 120 # optional: seconds (stream read)
|
'read_timeout': 120 # optional: seconds (stream read)
|
||||||
|
|||||||
Reference in new issue
Block a user