hz/ebiz-dify-ai
1
0
Fork 0
mirror of http://112.124.100.131/huang.ze/ebiz-dify-ai.git synced 2025-12-24 02:03:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix/price calc (#862)

Browse Source
This commit is contained in:
Krasus.Chen
2023-08-19 16:41:35 +08:00
committed by GitHub
parent 1c552ff23a
commit fd0fc8f4fe
22 changed files with 288 additions and 230 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
api/core/model_providers/models/llm/anthropic_model.py
View File

@@ -54,32 +54,6 @@ class AnthropicModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return max(self._client.get_num_tokens_from_messages(prompts) - len(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
model_unit_prices = {
'claude-instant-1': {
'prompt': decimal.Decimal('1.63'),
'completion': decimal.Decimal('5.51'),
},
'claude-2': {
'prompt': decimal.Decimal('11.02'),
'completion': decimal.Decimal('32.68'),
},
}
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = model_unit_prices[self.name]['prompt']
else:
unit_price = model_unit_prices[self.name]['completion']
tokens_per_1m = (decimal.Decimal(tokens) / 1000000).quantize(decimal.Decimal('0.000001'),
rounding=decimal.ROUND_HALF_UP)
total_price = tokens_per_1m * unit_price
return total_price.quantize(decimal.Decimal('0.00000001'), rounding=decimal.ROUND_HALF_UP)
def get_currency(self):
return 'USD'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
for k, v in provider_model_kwargs.items():

49
api/core/model_providers/models/llm/azure_openai_model.py
View File

@@ -29,7 +29,6 @@ class AzureOpenAIModel(BaseLLM):
self.model_mode = ModelMode.COMPLETION
else:
self.model_mode = ModelMode.CHAT
super().__init__(model_provider, name, model_kwargs, streaming, callbacks)
def _init_client(self) -> Any:
@@ -83,6 +82,15 @@ class AzureOpenAIModel(BaseLLM):
"""
prompts = self._get_prompt_from_messages(messages)
return self._client.generate([prompts], stop, callbacks)
@property
def base_model_name(self) -> str:
"""
get base model name (not deployment)
:return: str
"""
return self.credentials.get("base_model_name")
def get_num_tokens(self, messages: List[PromptMessage]) -> int:
"""
@@ -97,45 +105,6 @@ class AzureOpenAIModel(BaseLLM):
else:
return max(self._client.get_num_tokens_from_messages(prompts) - len(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
model_unit_prices = {
'gpt-4': {
'prompt': decimal.Decimal('0.03'),
'completion': decimal.Decimal('0.06'),
},
'gpt-4-32k': {
'prompt': decimal.Decimal('0.06'),
'completion': decimal.Decimal('0.12')
},
'gpt-35-turbo': {
'prompt': decimal.Decimal('0.0015'),
'completion': decimal.Decimal('0.002')
},
'gpt-35-turbo-16k': {
'prompt': decimal.Decimal('0.003'),
'completion': decimal.Decimal('0.004')
},
'text-davinci-003': {
'prompt': decimal.Decimal('0.02'),
'completion': decimal.Decimal('0.02')
},
}
base_model_name = self.credentials.get("base_model_name")
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = model_unit_prices[base_model_name]['prompt']
else:
unit_price = model_unit_prices[base_model_name]['completion']
tokens_per_1k = (decimal.Decimal(tokens) / 1000).quantize(decimal.Decimal('0.001'),
rounding=decimal.ROUND_HALF_UP)
total_price = tokens_per_1k * unit_price
return total_price.quantize(decimal.Decimal('0.0000001'), rounding=decimal.ROUND_HALF_UP)
def get_currency(self):
return 'USD'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
if self.name == 'text-davinci-003':

73
api/core/model_providers/models/llm/base.py
View File

@@ -1,5 +1,6 @@
from abc import abstractmethod
from typing import List, Optional, Any, Union
import decimal
from langchain.callbacks.manager import Callbacks
from langchain.schema import LLMResult, SystemMessage, AIMessage, HumanMessage, BaseMessage, ChatGeneration
@@ -10,6 +11,8 @@ from core.model_providers.models.entity.message import PromptMessage, MessageTyp
from core.model_providers.models.entity.model_params import ModelType, ModelKwargs, ModelMode, ModelKwargsRules
from core.model_providers.providers.base import BaseModelProvider
from core.third_party.langchain.llms.fake import FakeLLM
import logging
logger = logging.getLogger(__name__)
class BaseLLM(BaseProviderModel):
@@ -60,6 +63,39 @@ class BaseLLM(BaseProviderModel):
def _init_client(self) -> Any:
raise NotImplementedError
@property
def base_model_name(self) -> str:
"""
get llm base model name
:return: str
"""
return self.name
@property
def price_config(self) -> dict:
def get_or_default():
default_price_config = {
'prompt': decimal.Decimal('0'),
'completion': decimal.Decimal('0'),
'unit': decimal.Decimal('0'),
'currency': 'USD'
}
rules = self.model_provider.get_rules()
price_config = rules['price_config'][self.base_model_name] if 'price_config' in rules else default_price_config
price_config = {
'prompt': decimal.Decimal(price_config['prompt']),
'completion': decimal.Decimal(price_config['completion']),
'unit': decimal.Decimal(price_config['unit']),
'currency': price_config['currency']
}
return price_config
self._price_config = self._price_config if hasattr(self, '_price_config') else get_or_default()
logger.debug(f"model: {self.name} price_config: {self._price_config}")
return self._price_config
def run(self, messages: List[PromptMessage],
stop: Optional[List[str]] = None,
callbacks: Callbacks = None,
@@ -161,25 +197,48 @@ class BaseLLM(BaseProviderModel):
"""
raise NotImplementedError
@abstractmethod
def get_token_price(self, tokens: int, message_type: MessageType):
def calc_tokens_price(self, tokens:int, message_type: MessageType):
"""
get token price.
calc tokens total price.
:param tokens:
:param message_type:
:return:
"""
raise NotImplementedError
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = self.price_config['prompt']
else:
unit_price = self.price_config['completion']
unit = self.price_config['unit']
total_price = tokens * unit_price * unit
total_price = total_price.quantize(decimal.Decimal('0.0000001'), rounding=decimal.ROUND_HALF_UP)
logging.debug(f"tokens={tokens}, unit_price={unit_price}, unit={unit}, total_price:{total_price}")
return total_price
def get_tokens_unit_price(self, message_type: MessageType):
"""
get token price.
:param message_type:
:return: decimal.Decimal('0.0001')
"""
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = self.price_config['prompt']
else:
unit_price = self.price_config['completion']
unit_price = unit_price.quantize(decimal.Decimal('0.0001'), rounding=decimal.ROUND_HALF_UP)
logging.debug(f"unit_price={unit_price}")
return unit_price
@abstractmethod
def get_currency(self):
"""
get token currency.
:return:
:return: get from price config, default 'USD'
"""
raise NotImplementedError
currency = self.price_config['currency']
return currency
def get_model_kwargs(self):
return self.model_kwargs

3
api/core/model_providers/models/llm/chatglm_model.py
View File

@@ -47,9 +47,6 @@ class ChatGLMModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return max(self._client.get_num_tokens(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
return decimal.Decimal('0')
def get_currency(self):
return 'RMB'

7
api/core/model_providers/models/llm/huggingface_hub_model.py
View File

@@ -62,13 +62,6 @@ class HuggingfaceHubModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return self._client.get_num_tokens(prompts)
def get_token_price(self, tokens: int, message_type: MessageType):
# not support calc price
return decimal.Decimal('0')
def get_currency(self):
return 'USD'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
self.client.model_kwargs = provider_model_kwargs

3
api/core/model_providers/models/llm/minimax_model.py
View File

@@ -51,9 +51,6 @@ class MinimaxModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return max(self._client.get_num_tokens(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
return decimal.Decimal('0')
def get_currency(self):
return 'RMB'

41
api/core/model_providers/models/llm/openai_model.py
View File

@@ -46,7 +46,8 @@ class OpenAIModel(BaseLLM):
self.model_mode = ModelMode.COMPLETION
else:
self.model_mode = ModelMode.CHAT
# TODO load price config from configs(db)
super().__init__(model_provider, name, model_kwargs, streaming, callbacks)
def _init_client(self) -> Any:
@@ -117,44 +118,6 @@ class OpenAIModel(BaseLLM):
else:
return max(self._client.get_num_tokens_from_messages(prompts) - len(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
model_unit_prices = {
'gpt-4': {
'prompt': decimal.Decimal('0.03'),
'completion': decimal.Decimal('0.06'),
},
'gpt-4-32k': {
'prompt': decimal.Decimal('0.06'),
'completion': decimal.Decimal('0.12')
},
'gpt-3.5-turbo': {
'prompt': decimal.Decimal('0.0015'),
'completion': decimal.Decimal('0.002')
},
'gpt-3.5-turbo-16k': {
'prompt': decimal.Decimal('0.003'),
'completion': decimal.Decimal('0.004')
},
'text-davinci-003': {
'prompt': decimal.Decimal('0.02'),
'completion': decimal.Decimal('0.02')
},
}
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = model_unit_prices[self.name]['prompt']
else:
unit_price = model_unit_prices[self.name]['completion']
tokens_per_1k = (decimal.Decimal(tokens) / 1000).quantize(decimal.Decimal('0.001'),
rounding=decimal.ROUND_HALF_UP)
total_price = tokens_per_1k * unit_price
return total_price.quantize(decimal.Decimal('0.0000001'), rounding=decimal.ROUND_HALF_UP)
def get_currency(self):
return 'USD'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
if self.name in COMPLETION_MODELS:

7
api/core/model_providers/models/llm/replicate_model.py
View File

@@ -81,13 +81,6 @@ class ReplicateModel(BaseLLM):
return self._client.get_num_tokens(prompts)
def get_token_price(self, tokens: int, message_type: MessageType):
# replicate only pay for prediction seconds
return decimal.Decimal('0')
def get_currency(self):
return 'USD'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
self.client.input = provider_model_kwargs

3
api/core/model_providers/models/llm/spark_model.py
View File

@@ -50,9 +50,6 @@ class SparkModel(BaseLLM):
contents = [message.content for message in messages]
return max(self._client.get_num_tokens("".join(contents)), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
return decimal.Decimal('0')
def get_currency(self):
return 'RMB'

3
api/core/model_providers/models/llm/tongyi_model.py
View File

@@ -53,9 +53,6 @@ class TongyiModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return max(self._client.get_num_tokens(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
return decimal.Decimal('0')
def get_currency(self):
return 'RMB'

31
api/core/model_providers/models/llm/wenxin_model.py
View File

@@ -16,6 +16,7 @@ class WenxinModel(BaseLLM):
def _init_client(self) -> Any:
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, self.model_kwargs)
# TODO load price_config from configs(db)
return Wenxin(
streaming=self.streaming,
callbacks=self.callbacks,
@@ -48,36 +49,6 @@ class WenxinModel(BaseLLM):
prompts = self._get_prompt_from_messages(messages)
return max(self._client.get_num_tokens(prompts), 0)
def get_token_price(self, tokens: int, message_type: MessageType):
model_unit_prices = {
'ernie-bot': {
'prompt': decimal.Decimal('0.012'),
'completion': decimal.Decimal('0.012'),
},
'ernie-bot-turbo': {
'prompt': decimal.Decimal('0.008'),
'completion': decimal.Decimal('0.008')
},
'bloomz-7b': {
'prompt': decimal.Decimal('0.006'),
'completion': decimal.Decimal('0.006')
}
}
if message_type == MessageType.HUMAN or message_type == MessageType.SYSTEM:
unit_price = model_unit_prices[self.name]['prompt']
else:
unit_price = model_unit_prices[self.name]['completion']
tokens_per_1k = (decimal.Decimal(tokens) / 1000).quantize(decimal.Decimal('0.001'),
rounding=decimal.ROUND_HALF_UP)
total_price = tokens_per_1k * unit_price
return total_price.quantize(decimal.Decimal('0.0000001'), rounding=decimal.ROUND_HALF_UP)
def get_currency(self):
return 'RMB'
def _set_model_kwargs(self, model_kwargs: ModelKwargs):
provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
for k, v in provider_model_kwargs.items():