youngkingdom/dify
1
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-05-03 17:08:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge tag '0.15.7' into e-260

Browse Source 
0.15.7
This commit is contained in:
Garfield Dai
2025-04-28 17:17:26 +08:00
parent 2fce4a338c 5f7f851b17
commit 849994d35e
26 changed files with 319 additions and 215 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
api/core/agent/cot_agent_runner.py
View File

@ -104,7 +104,6 @@ class CotAgentRunner(BaseAgentRunner, ABC):
# recalc llm max tokens
prompt_messages = self._organize_prompt_messages()
self.recalc_llm_max_tokens(self.model_config, prompt_messages)
# invoke model
chunks = model_instance.invoke_llm(
prompt_messages=prompt_messages,

1
api/core/agent/fc_agent_runner.py
View File

@ -84,7 +84,6 @@ class FunctionCallAgentRunner(BaseAgentRunner):
# recalc llm max tokens
prompt_messages = self._organize_prompt_messages()
self.recalc_llm_max_tokens(self.model_config, prompt_messages)
# invoke model
chunks: Union[Generator[LLMResultChunk, None, None], LLMResult] = model_instance.invoke_llm(
prompt_messages=prompt_messages,

14
api/core/app/apps/agent_chat/app_runner.py
View File

@ -55,20 +55,6 @@ class AgentChatAppRunner(AppRunner):
query = application_generate_entity.query
files = application_generate_entity.files
# Pre-calculate the number of tokens of the prompt messages,
# and return the rest number of tokens by model context token size limit and max token size limit.
# If the rest number of tokens is not enough, raise exception.
# Include: prompt template, inputs, query(optional), files(optional)
# Not Include: memory, external data, dataset context
self.get_pre_calculate_rest_tokens(
app_record=app_record,
model_config=application_generate_entity.model_conf,
prompt_template_entity=app_config.prompt_template,
inputs=inputs,
files=files,
query=query,
)
memory = None
if application_generate_entity.conversation_id:
# get memory of conversation (read-only)

102
api/core/app/apps/base_app_runner.py
View File

@ -15,10 +15,8 @@ from core.app.features.annotation_reply.annotation_reply import AnnotationReplyF
from core.app.features.hosting_moderation.hosting_moderation import HostingModerationFeature
from core.external_data_tool.external_data_fetch import ExternalDataFetch
from core.memory.token_buffer_memory import TokenBufferMemory
from core.model_manager import ModelInstance
from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
from core.model_runtime.entities.message_entities import AssistantPromptMessage, PromptMessage
from core.model_runtime.entities.model_entities import ModelPropertyKey
from core.model_runtime.errors.invoke import InvokeBadRequestError
from core.moderation.input_moderation import InputModeration
from core.prompt.advanced_prompt_transform import AdvancedPromptTransform
@ -31,106 +29,6 @@ if TYPE_CHECKING:
class AppRunner:
def get_pre_calculate_rest_tokens(
self,
app_record: App,
model_config: ModelConfigWithCredentialsEntity,
prompt_template_entity: PromptTemplateEntity,
inputs: Mapping[str, str],
files: Sequence["File"],
query: Optional[str] = None,
) -> int:
"""
Get pre calculate rest tokens
:param app_record: app record
:param model_config: model config entity
:param prompt_template_entity: prompt template entity
:param inputs: inputs
:param files: files
:param query: query
:return:
"""
# Invoke model
model_instance = ModelInstance(
provider_model_bundle=model_config.provider_model_bundle, model=model_config.model
)
model_context_tokens = model_config.model_schema.model_properties.get(ModelPropertyKey.CONTEXT_SIZE)
max_tokens = 0
for parameter_rule in model_config.model_schema.parameter_rules:
if parameter_rule.name == "max_tokens" or (
parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
):
max_tokens = (
model_config.parameters.get(parameter_rule.name)
or model_config.parameters.get(parameter_rule.use_template or "")
) or 0
if model_context_tokens is None:
return -1
if max_tokens is None:
max_tokens = 0
# get prompt messages without memory and context
prompt_messages, stop = self.organize_prompt_messages(
app_record=app_record,
model_config=model_config,
prompt_template_entity=prompt_template_entity,
inputs=inputs,
files=files,
query=query,
)
prompt_tokens = model_instance.get_llm_num_tokens(prompt_messages)
rest_tokens: int = model_context_tokens - max_tokens - prompt_tokens
if rest_tokens < 0:
raise InvokeBadRequestError(
"Query or prefix prompt is too long, you can reduce the prefix prompt, "
"or shrink the max token, or switch to a llm with a larger token limit size."
)
return rest_tokens
def recalc_llm_max_tokens(
self, model_config: ModelConfigWithCredentialsEntity, prompt_messages: list[PromptMessage]
):
# recalc max_tokens if sum(prompt_token + max_tokens) over model token limit
model_instance = ModelInstance(
provider_model_bundle=model_config.provider_model_bundle, model=model_config.model
)
model_context_tokens = model_config.model_schema.model_properties.get(ModelPropertyKey.CONTEXT_SIZE)
max_tokens = 0
for parameter_rule in model_config.model_schema.parameter_rules:
if parameter_rule.name == "max_tokens" or (
parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
):
max_tokens = (
model_config.parameters.get(parameter_rule.name)
or model_config.parameters.get(parameter_rule.use_template or "")
) or 0
if model_context_tokens is None:
return -1
if max_tokens is None:
max_tokens = 0
prompt_tokens = model_instance.get_llm_num_tokens(prompt_messages)
if prompt_tokens + max_tokens > model_context_tokens:
max_tokens = max(model_context_tokens - prompt_tokens, 16)
for parameter_rule in model_config.model_schema.parameter_rules:
if parameter_rule.name == "max_tokens" or (
parameter_rule.use_template and parameter_rule.use_template == "max_tokens"
):
model_config.parameters[parameter_rule.name] = max_tokens
def organize_prompt_messages(
self,
app_record: App,

17
api/core/app/apps/chat/app_runner.py
View File

@ -50,20 +50,6 @@ class ChatAppRunner(AppRunner):
query = application_generate_entity.query
files = application_generate_entity.files
# Pre-calculate the number of tokens of the prompt messages,
# and return the rest number of tokens by model context token size limit and max token size limit.
# If the rest number of tokens is not enough, raise exception.
# Include: prompt template, inputs, query(optional), files(optional)
# Not Include: memory, external data, dataset context
self.get_pre_calculate_rest_tokens(
app_record=app_record,
model_config=application_generate_entity.model_conf,
prompt_template_entity=app_config.prompt_template,
inputs=inputs,
files=files,
query=query,
)
memory = None
if application_generate_entity.conversation_id:
# get memory of conversation (read-only)
@ -194,9 +180,6 @@ class ChatAppRunner(AppRunner):
if hosting_moderation_result:
return
# Re-calculate the max tokens if sum(prompt_token + max_tokens) over model token limit
self.recalc_llm_max_tokens(model_config=application_generate_entity.model_conf, prompt_messages=prompt_messages)
# Invoke model
model_instance = ModelInstance(
provider_model_bundle=application_generate_entity.model_conf.provider_model_bundle,

17
api/core/app/apps/completion/app_runner.py
View File

@ -43,20 +43,6 @@ class CompletionAppRunner(AppRunner):
query = application_generate_entity.query
files = application_generate_entity.files
# Pre-calculate the number of tokens of the prompt messages,
# and return the rest number of tokens by model context token size limit and max token size limit.
# If the rest number of tokens is not enough, raise exception.
# Include: prompt template, inputs, query(optional), files(optional)
# Not Include: memory, external data, dataset context
self.get_pre_calculate_rest_tokens(
app_record=app_record,
model_config=application_generate_entity.model_conf,
prompt_template_entity=app_config.prompt_template,
inputs=inputs,
files=files,
query=query,
)
# organize all inputs and template to prompt messages
# Include: prompt template, inputs, query(optional), files(optional)
prompt_messages, stop = self.organize_prompt_messages(
@ -152,9 +138,6 @@ class CompletionAppRunner(AppRunner):
if hosting_moderation_result:
return
# Re-calculate the max tokens if sum(prompt_token + max_tokens) over model token limit
self.recalc_llm_max_tokens(model_config=application_generate_entity.model_conf, prompt_messages=prompt_messages)
# Invoke model
model_instance = ModelInstance(
provider_model_bundle=application_generate_entity.model_conf.provider_model_bundle,

2
api/core/memory/token_buffer_memory.py
View File

@ -26,7 +26,7 @@ class TokenBufferMemory:
self.model_instance = model_instance
def get_history_prompt_messages(
self, max_token_limit: int = 2000, message_limit: Optional[int] = None
self, max_token_limit: int = 100000, message_limit: Optional[int] = None
) -> Sequence[PromptMessage]:
"""
Get history prompt messages.

115
api/core/model_runtime/model_providers/bedrock/llm/eu.anthropic.claude-3.7-sonnet-v1.yaml Normal file
View File

@ -0,0 +1,115 @@
model: us.anthropic.claude-3-7-sonnet-20250219-v1:0
label:
en_US: Claude 3.7 Sonnet(US.Cross Region Inference)
icon: icon_s_en.svg
model_type: llm
features:
- agent-thought
- vision
- tool-call
- stream-tool-call
model_properties:
mode: chat
context_size: 200000
# docs: https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
parameter_rules:
- name: enable_cache
label:
zh_Hans: 启用提示缓存
en_US: Enable Prompt Cache
type: boolean
required: false
default: true
help:
zh_Hans: 启用提示缓存可以提高性能并降低成本。Claude 3.7 Sonnet支持在system、messages和tools字段中使用缓存检查点。
en_US: Enable prompt caching to improve performance and reduce costs. Claude 3.7 Sonnet supports cache checkpoints in system, messages, and tools fields.
- name: reasoning_type
label:
zh_Hans: 推理配置
en_US: Reasoning Type
type: boolean
required: false
default: false
placeholder:
zh_Hans: 设置推理配置
en_US: Set reasoning configuration
help:
zh_Hans: 控制模型的推理能力。启用时，temperature将固定为1且top_p将被禁用。
en_US: Controls the model's reasoning capability. When enabled, temperature will be fixed to 1 and top_p will be disabled.
- name: reasoning_budget
show_on:
- variable: reasoning_type
value: true
label:
zh_Hans: 推理预算
en_US: Reasoning Budget
type: int
default: 1024
min: 0
max: 128000
help:
zh_Hans: 推理的预算限制（最小1024），必须小于max_tokens。仅在推理类型为enabled时可用。
en_US: Budget limit for reasoning (minimum 1024), must be less than max_tokens. Only available when reasoning type is enabled.
- name: max_tokens
use_template: max_tokens
required: true
label:
zh_Hans: 最大token数
en_US: Max Tokens
type: int
default: 8192
min: 1
max: 128000
help:
zh_Hans: 停止前生成的最大令牌数。请注意，Anthropic Claude 模型可能会在达到 max_tokens 的值之前停止生成令牌。不同的 Anthropic Claude 模型对此参数具有不同的最大值。
en_US: The maximum number of tokens to generate before stopping. Note that Anthropic Claude models might stop generating tokens before reaching the value of max_tokens. Different Anthropic Claude models have different maximum values for this parameter.
- name: temperature
use_template: temperature
required: false
label:
zh_Hans: 模型温度
en_US: Model Temperature
type: float
default: 1
min: 0.0
max: 1.0
help:
zh_Hans: 生成内容的随机性。当推理功能启用时，该值将被固定为1。
en_US: The amount of randomness injected into the response. When reasoning is enabled, this value will be fixed to 1.
- name: top_p
show_on:
- variable: reasoning_type
value: disabled
use_template: top_p
label:
zh_Hans: Top P
en_US: Top P
required: false
type: float
default: 0.999
min: 0.000
max: 1.000
help:
zh_Hans: 在核采样中的概率阈值。当推理功能启用时，该参数将被禁用。
en_US: The probability threshold in nucleus sampling. When reasoning is enabled, this parameter will be disabled.
- name: top_k
label:
zh_Hans: 取样数量
en_US: Top k
required: false
type: int
default: 0
min: 0
# tip docs from aws has error, max value is 500
max: 500
help:
zh_Hans: 对于每个后续标记，仅从前 K 个选项中进行采样。使用 top_k 删除长尾低概率响应。
en_US: Only sample from the top K options for each subsequent token. Use top_k to remove long tail low probability responses.
- name: response_format
use_template: response_format
pricing:
input: '0.003'
output: '0.015'
unit: '0.001'
currency: USD

1
api/core/model_runtime/model_providers/bedrock/llm/llm.py
View File

@ -58,6 +58,7 @@ class BedrockLargeLanguageModel(LargeLanguageModel):
# TODO There is invoke issue: context limit on Cohere Model, will add them after fixed.
CONVERSE_API_ENABLED_MODEL_INFO = [
{"prefix": "anthropic.claude-v2", "support_system_prompts": True, "support_tool_use": False},
{"prefix": "us.deepseek", "support_system_prompts": True, "support_tool_use": False},
{"prefix": "anthropic.claude-v1", "support_system_prompts": True, "support_tool_use": False},
{"prefix": "us.anthropic.claude-3", "support_system_prompts": True, "support_tool_use": True},
{"prefix": "eu.anthropic.claude-3", "support_system_prompts": True, "support_tool_use": True},

63
api/core/model_runtime/model_providers/bedrock/llm/us.deepseek-r1.yaml Normal file
View File

@ -0,0 +1,63 @@
model: us.deepseek.r1-v1:0
label:
en_US: DeepSeek-R1(US.Cross Region Inference)
icon: icon_s_en.svg
model_type: llm
features:
- agent-thought
- vision
- tool-call
- stream-tool-call
model_properties:
mode: chat
context_size: 32768
parameter_rules:
- name: max_tokens
use_template: max_tokens
required: true
label:
zh_Hans: 最大token数
en_US: Max Tokens
type: int
default: 8192
min: 1
max: 128000
help:
zh_Hans: 停止前生成的最大令牌数。
en_US: The maximum number of tokens to generate before stopping.
- name: temperature
use_template: temperature
required: false
label:
zh_Hans: 模型温度
en_US: Model Temperature
type: float
default: 1
min: 0.0
max: 1.0
help:
zh_Hans: 生成内容的随机性。当推理功能启用时，该值将被固定为1。
en_US: The amount of randomness injected into the response. When reasoning is enabled, this value will be fixed to 1.
- name: top_p
show_on:
- variable: reasoning_type
value: disabled
use_template: top_p
label:
zh_Hans: Top P
en_US: Top P
required: false
type: float
default: 0.999
min: 0.000
max: 1.000
help:
zh_Hans: 在核采样中的概率阈值。当推理功能启用时，该参数将被禁用。
en_US: The probability threshold in nucleus sampling. When reasoning is enabled, this parameter will be disabled.
- name: response_format
use_template: response_format
pricing:
input: '0.001'
output: '0.005'
unit: '0.001'
currency: USD

1
api/core/model_runtime/model_providers/openai/llm/_position.yaml
View File

@ -1,3 +1,4 @@
- gpt-4.1
- o1
- o1-2024-12-17
- o1-mini

60
api/core/model_runtime/model_providers/openai/llm/gpt-4-1.yaml Normal file
View File

@ -0,0 +1,60 @@
model: gpt-4.1
label:
zh_Hans: gpt-4.1
en_US: gpt-4.1
model_type: llm
features:
- multi-tool-call
- agent-thought
- stream-tool-call
- vision
model_properties:
mode: chat
context_size: 1047576
parameter_rules:
- name: temperature
use_template: temperature
- name: top_p
use_template: top_p
- name: presence_penalty
use_template: presence_penalty
- name: frequency_penalty
use_template: frequency_penalty
- name: max_tokens
use_template: max_tokens
default: 512
min: 1
max: 32768
- name: reasoning_effort
label:
zh_Hans: 推理工作
en_US: Reasoning Effort
type: string
help:
zh_Hans: 限制推理模型的推理工作
en_US: Constrains effort on reasoning for reasoning models
required: false
options:
- low
- medium
- high
- name: response_format
label:
zh_Hans: 回复格式
en_US: Response Format
type: string
help:
zh_Hans: 指定模型必须输出的格式
en_US: specifying the format that the model must output
required: false
options:
- text
- json_object
- json_schema
- name: json_schema
use_template: json_schema
pricing:
input: '2.00'
output: '8.00'
unit: '0.000001'
currency: USD

2
api/core/model_runtime/model_providers/openai/llm/llm.py
View File

@ -1057,7 +1057,7 @@ class OpenAILargeLanguageModel(_CommonOpenAI, LargeLanguageModel):
model = "gpt-4o"
try:
encoding = tiktoken.encoding_for_model(model)
encoding = tiktoken.get_encoding(model)
except KeyError:
logger.warning("Warning: model not found. Using cl100k_base encoding.")
model = "cl100k_base"

8
api/core/workflow/nodes/code/code_node.py
View File

@ -195,7 +195,7 @@ class CodeNode(BaseNode[CodeNodeData]):
if output_config.type == "object":
# check if output is object
if not isinstance(result.get(output_name), dict):
if isinstance(result.get(output_name), type(None)):
if result.get(output_name) is None:
transformed_result[output_name] = None
else:
raise OutputValidationError(
@ -223,7 +223,7 @@ class CodeNode(BaseNode[CodeNodeData]):
elif output_config.type == "array[number]":
# check if array of number available
if not isinstance(result[output_name], list):
if isinstance(result[output_name], type(None)):
if result[output_name] is None:
transformed_result[output_name] = None
else:
raise OutputValidationError(
@ -244,7 +244,7 @@ class CodeNode(BaseNode[CodeNodeData]):
elif output_config.type == "array[string]":
# check if array of string available
if not isinstance(result[output_name], list):
if isinstance(result[output_name], type(None)):
if result[output_name] is None:
transformed_result[output_name] = None
else:
raise OutputValidationError(
@ -265,7 +265,7 @@ class CodeNode(BaseNode[CodeNodeData]):
elif output_config.type == "array[object]":
# check if array of object available
if not isinstance(result[output_name], list):
if isinstance(result[output_name], type(None)):
if result[output_name] is None:
transformed_result[output_name] = None
else:
raise OutputValidationError(

4
api/core/workflow/nodes/llm/node.py
View File

@ -968,14 +968,12 @@ def _handle_memory_chat_mode(
*,
memory: TokenBufferMemory | None,
memory_config: MemoryConfig | None,
model_config: ModelConfigWithCredentialsEntity,
model_config: ModelConfigWithCredentialsEntity, # TODO(-LAN-): Needs to remove
) -> Sequence[PromptMessage]:
memory_messages: Sequence[PromptMessage] = []
# Get messages from memory for chat model
if memory and memory_config:
rest_tokens = _calculate_rest_token(prompt_messages=[], model_config=model_config)
memory_messages = memory.get_history_prompt_messages(
max_token_limit=rest_tokens,
message_limit=memory_config.window.size if memory_config.window.enabled else None,
)
return memory_messages