vodorod/Dorod-Sky
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

O3 Mini support (#1709)

Browse Source
This commit is contained in:
Shuchang Zheng
2025-02-04 05:07:10 +08:00
committed by GitHub
parent e0e868445d
commit 59756cb1d2
3 changed files with 39 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
skyvern/forge/sdk/api/llm/api_handler_factory.py
View File

@@ -155,7 +155,12 @@ class LLMAPIHandlerFactory:
LOG.exception("Failed to calculate LLM cost", error=str(e)) LOG.exception("Failed to calculate LLM cost", error=str(e))
llm_cost = 0 llm_cost = 0
prompt_tokens = response.get("usage", {}).get("prompt_tokens", 0) prompt_tokens = response.get("usage", {}).get("prompt_tokens", 0)
completion_tokens = response.get("usage", {}).get("completion_tokens", 0)
# TODO (suchintan): Properly support reasoning tokens
reasoning_tokens = response.get("usage", {}).get("reasoning_tokens", 0)
LOG.info("Reasoning tokens", reasoning_tokens=reasoning_tokens)
completion_tokens = response.get("usage", {}).get("completion_tokens", 0) + reasoning_tokens
if step: if step:
await app.DATABASE.update_step( await app.DATABASE.update_step(
@@ -388,10 +393,15 @@ class LLMAPIHandlerFactory:
@staticmethod @staticmethod
def get_api_parameters(llm_config: LLMConfig | LLMRouterConfig) -> dict[str, Any]: def get_api_parameters(llm_config: LLMConfig | LLMRouterConfig) -> dict[str, Any]:
return { params: dict[str, Any] = {"max_completion_tokens": llm_config.max_completion_tokens}
"max_tokens": llm_config.max_output_tokens,
"temperature": settings.LLM_CONFIG_TEMPERATURE, if llm_config.temperature is not None:
} params["temperature"] = llm_config.temperature
if llm_config.reasoning_effort is not None:
params["reasoning_effort"] = llm_config.reasoning_effort
return params
@classmethod @classmethod
def register_custom_handler(cls, llm_key: str, handler: LLMAPIHandler) -> None: def register_custom_handler(cls, llm_key: str, handler: LLMAPIHandler) -> None:

24
skyvern/forge/sdk/api/llm/config_registry.py
View File

@@ -80,7 +80,19 @@ if settings.ENABLE_OPENAI:
LLMConfigRegistry.register_config( LLMConfigRegistry.register_config(
"OPENAI_GPT4O", "OPENAI_GPT4O",
LLMConfig( LLMConfig(
"gpt-4o", ["OPENAI_API_KEY"], supports_vision=True, add_assistant_prefix=False, max_output_tokens=16384 "gpt-4o", ["OPENAI_API_KEY"], supports_vision=True, add_assistant_prefix=False, max_completion_tokens=16384
),
)
LLMConfigRegistry.register_config(
"OPENAI_O3_MINI",
LLMConfig(
"o3-mini",
["OPENAI_API_KEY"],
supports_vision=False,
add_assistant_prefix=False,
max_completion_tokens=16384,
temperature=None, # Temperature isn't supported in the O-model series
reasoning_effort="high",
), ),
) )
LLMConfigRegistry.register_config( LLMConfigRegistry.register_config(
@@ -90,7 +102,7 @@ if settings.ENABLE_OPENAI:
["OPENAI_API_KEY"], ["OPENAI_API_KEY"],
supports_vision=True, supports_vision=True,
add_assistant_prefix=False, add_assistant_prefix=False,
max_output_tokens=16384, max_completion_tokens=16384,
), ),
) )
LLMConfigRegistry.register_config( LLMConfigRegistry.register_config(
@@ -100,7 +112,7 @@ if settings.ENABLE_OPENAI:
["OPENAI_API_KEY"], ["OPENAI_API_KEY"],
supports_vision=True, supports_vision=True,
add_assistant_prefix=False, add_assistant_prefix=False,
max_output_tokens=16384, max_completion_tokens=16384,
), ),
) )
@@ -149,7 +161,7 @@ if settings.ENABLE_ANTHROPIC:
["ANTHROPIC_API_KEY"], ["ANTHROPIC_API_KEY"],
supports_vision=True, supports_vision=True,
add_assistant_prefix=True, add_assistant_prefix=True,
max_output_tokens=8192, max_completion_tokens=8192,
), ),
) )
@@ -275,7 +287,7 @@ if settings.ENABLE_GEMINI:
["GEMINI_API_KEY"], ["GEMINI_API_KEY"],
supports_vision=True, supports_vision=True,
add_assistant_prefix=False, add_assistant_prefix=False,
max_output_tokens=8192, max_completion_tokens=8192,
), ),
) )
LLMConfigRegistry.register_config( LLMConfigRegistry.register_config(
@@ -285,7 +297,7 @@ if settings.ENABLE_GEMINI:
["GEMINI_API_KEY"], ["GEMINI_API_KEY"],
supports_vision=True, supports_vision=True,
add_assistant_prefix=False, add_assistant_prefix=False,
max_output_tokens=8192, max_completion_tokens=8192,
), ),
) )

8
skyvern/forge/sdk/api/llm/models.py
View File

@@ -36,7 +36,9 @@ class LLMConfigBase:
@dataclass(frozen=True) @dataclass(frozen=True)
class LLMConfig(LLMConfigBase): class LLMConfig(LLMConfigBase):
litellm_params: Optional[LiteLLMParams] = field(default=None) litellm_params: Optional[LiteLLMParams] = field(default=None)
max_output_tokens: int = SettingsManager.get_settings().LLM_CONFIG_MAX_TOKENS max_completion_tokens: int = SettingsManager.get_settings().LLM_CONFIG_MAX_TOKENS
temperature: float | None = SettingsManager.get_settings().LLM_CONFIG_TEMPERATURE
reasoning_effort: str | None = None
@dataclass(frozen=True) @dataclass(frozen=True)
@@ -72,7 +74,9 @@ class LLMRouterConfig(LLMConfigBase):
allowed_fails: int | None = None allowed_fails: int | None = None
allowed_fails_policy: AllowedFailsPolicy | None = None allowed_fails_policy: AllowedFailsPolicy | None = None
cooldown_time: float | None = None cooldown_time: float | None = None
max_output_tokens: int = SettingsManager.get_settings().LLM_CONFIG_MAX_TOKENS max_completion_tokens: int = SettingsManager.get_settings().LLM_CONFIG_MAX_TOKENS
reasoning_effort: str | None = None
temperature: float | None = SettingsManager.get_settings().LLM_CONFIG_TEMPERATURE
class LLMAPIHandler(Protocol): class LLMAPIHandler(Protocol):