大修LLMReq
This commit is contained in:
UnCLAS-Prommer
@@ -1,5 +1,5 @@
|
||||
[inner]
|
||||
version = "0.2.1"
|
||||
version = "1.0.0"
|
||||
|
||||
# 配置文件版本号迭代规则同bot_config.toml
|
||||
#
|
||||
@@ -42,53 +42,31 @@ version = "0.2.1"
|
||||
# - 未配置新字段时会自动回退到基于模型名称的推断
|
||||
|
||||
[request_conf] # 请求配置（此配置项数值均为默认值，如想修改，请取消对应条目的注释）
|
||||
#max_retry = 2 # 最大重试次数（单个模型API调用失败，最多重试的次数）
|
||||
#timeout = 10 # API调用的超时时长（超过这个时长，本次请求将被视为“请求超时”，单位：秒）
|
||||
#retry_interval = 10 # 重试间隔（如果API调用失败，重试的间隔时间，单位：秒）
|
||||
#default_temperature = 0.7 # 默认的温度（如果bot_config.toml中没有设置temperature参数，默认使用这个值）
|
||||
#default_max_tokens = 1024 # 默认的最大输出token数（如果bot_config.toml中没有设置max_tokens参数，默认使用这个值）
|
||||
max_retry = 2 # 最大重试次数（单个模型API调用失败，最多重试的次数）
|
||||
timeout = 10 # API调用的超时时长（超过这个时长，本次请求将被视为“请求超时”，单位：秒）
|
||||
retry_interval = 10 # 重试间隔（如果API调用失败，重试的间隔时间，单位：秒）
|
||||
default_temperature = 0.7 # 默认的温度（如果bot_config.toml中没有设置temperature参数，默认使用这个值）
|
||||
default_max_tokens = 1024 # 默认的最大输出token数（如果bot_config.toml中没有设置max_tokens参数，默认使用这个值）
|
||||
|
||||
|
||||
[[api_providers]] # API服务提供商（可以配置多个）
|
||||
name = "DeepSeek" # API服务商名称（可随意命名，在models的api-provider中需使用这个命名）
|
||||
base_url = "https://api.deepseek.cn/v1" # API服务商的BaseURL
|
||||
# 支持多个API Key，实现自动切换和负载均衡
|
||||
api_keys = [ # API Key列表（多个key支持错误自动切换和负载均衡）
|
||||
"sk-your-first-key-here",
|
||||
"sk-your-second-key-here",
|
||||
"sk-your-third-key-here"
|
||||
]
|
||||
# 向后兼容：如果只有一个key，也可以使用单个key字段
|
||||
#key = "******" # API Key （可选，默认为None）
|
||||
api_key = "sk-your-first-key-here"
|
||||
client_type = "openai" # 请求客户端（可选，默认值为"openai"，使用gimini等Google系模型时请配置为"gemini"）
|
||||
|
||||
[[api_providers]] # 特殊：Google的Gimini使用特殊API，与OpenAI格式不兼容，需要配置client为"gemini"
|
||||
name = "Google"
|
||||
base_url = "https://api.google.com/v1"
|
||||
# Google API同样支持多key配置
|
||||
api_keys = [
|
||||
"your-google-api-key-1",
|
||||
"your-google-api-key-2"
|
||||
]
|
||||
api_key = "your-google-api-key-1"
|
||||
client_type = "gemini"
|
||||
|
||||
[[api_providers]]
|
||||
name = "SiliconFlow"
|
||||
base_url = "https://api.siliconflow.cn/v1"
|
||||
# 单个key的示例（向后兼容）
|
||||
key = "******"
|
||||
#
|
||||
#[[api_providers]]
|
||||
#name = "LocalHost"
|
||||
#base_url = "https://localhost:8888"
|
||||
#key = "lm-studio"
|
||||
|
||||
|
||||
[[models]] # 模型（可以配置多个）
|
||||
# 模型标识符（API服务商提供的模型标识符）
|
||||
model_identifier = "deepseek-chat"
|
||||
# 模型名称（可随意命名，在bot_config.toml中需使用这个命名）
|
||||
#（可选，若无该字段，则将自动使用model_identifier填充）
|
||||
name = "deepseek-v3"
|
||||
# API服务商名称（对应在api_providers中配置的服务商名称）
|
||||
api_provider = "DeepSeek"
|
||||
@@ -111,20 +89,15 @@ price_out = 8.0
|
||||
model_identifier = "deepseek-reasoner"
|
||||
name = "deepseek-r1"
|
||||
api_provider = "DeepSeek"
|
||||
# 推理模型的配置示例
|
||||
task_type = "llm_reasoning"
|
||||
capabilities = ["text", "tool_calling", "reasoning"]
|
||||
# 保留向后兼容的model_flags字段（已废弃，建议使用capabilities）
|
||||
model_flags = [ "text", "tool_calling", "reasoning",]
|
||||
price_in = 4.0
|
||||
price_out = 16.0
|
||||
has_thinking = true # 有无思考参数
|
||||
enable_thinking = true # 是否启用思考
|
||||
|
||||
[[models]]
|
||||
model_identifier = "Pro/deepseek-ai/DeepSeek-V3"
|
||||
name = "siliconflow-deepseek-v3"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_normal"
|
||||
capabilities = ["text", "tool_calling"]
|
||||
price_in = 2.0
|
||||
price_out = 8.0
|
||||
|
||||
@@ -132,8 +105,6 @@ price_out = 8.0
|
||||
model_identifier = "Pro/deepseek-ai/DeepSeek-R1"
|
||||
name = "siliconflow-deepseek-r1"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_reasoning"
|
||||
capabilities = ["text", "tool_calling", "reasoning"]
|
||||
price_in = 4.0
|
||||
price_out = 16.0
|
||||
|
||||
@@ -141,8 +112,6 @@ price_out = 16.0
|
||||
model_identifier = "Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"
|
||||
name = "deepseek-r1-distill-qwen-32b"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_reasoning"
|
||||
capabilities = ["text", "tool_calling", "reasoning"]
|
||||
price_in = 4.0
|
||||
price_out = 16.0
|
||||
|
||||
@@ -150,8 +119,6 @@ price_out = 16.0
|
||||
model_identifier = "Qwen/Qwen3-8B"
|
||||
name = "qwen3-8b"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_normal"
|
||||
capabilities = ["text"]
|
||||
price_in = 0
|
||||
price_out = 0
|
||||
|
||||
@@ -159,8 +126,6 @@ price_out = 0
|
||||
model_identifier = "Qwen/Qwen3-14B"
|
||||
name = "qwen3-14b"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_normal"
|
||||
capabilities = ["text", "tool_calling"]
|
||||
price_in = 0.5
|
||||
price_out = 2.0
|
||||
|
||||
@@ -168,8 +133,6 @@ price_out = 2.0
|
||||
model_identifier = "Qwen/Qwen3-30B-A3B"
|
||||
name = "qwen3-30b"
|
||||
api_provider = "SiliconFlow"
|
||||
task_type = "llm_normal"
|
||||
capabilities = ["text", "tool_calling"]
|
||||
price_in = 0.7
|
||||
price_out = 2.8
|
||||
|
||||
@@ -177,11 +140,6 @@ price_out = 2.8
|
||||
model_identifier = "Qwen/Qwen2.5-VL-72B-Instruct"
|
||||
name = "qwen2.5-vl-72b"
|
||||
api_provider = "SiliconFlow"
|
||||
# 视觉模型的配置示例
|
||||
task_type = "vision"
|
||||
capabilities = ["vision", "text"]
|
||||
# 保留向后兼容的model_flags字段（已废弃，建议使用capabilities）
|
||||
model_flags = [ "vision", "text",]
|
||||
price_in = 4.13
|
||||
price_out = 4.13
|
||||
|
||||
@@ -189,11 +147,6 @@ price_out = 4.13
|
||||
model_identifier = "FunAudioLLM/SenseVoiceSmall"
|
||||
name = "sensevoice-small"
|
||||
api_provider = "SiliconFlow"
|
||||
# 语音模型的配置示例
|
||||
task_type = "speech"
|
||||
capabilities = ["speech"]
|
||||
# 保留向后兼容的model_flags字段（已废弃，建议使用capabilities）
|
||||
model_flags = [ "audio",]
|
||||
price_in = 0
|
||||
price_out = 0
|
||||
|
||||
@@ -210,11 +163,73 @@ price_in = 0
|
||||
price_out = 0
|
||||
|
||||
|
||||
[task_model_usage]
|
||||
llm_reasoning = {model="deepseek-r1", temperature=0.8, max_tokens=1024, max_retry=0}
|
||||
llm_normal = {model="deepseek-r1", max_tokens=1024, max_retry=0}
|
||||
embedding = "siliconflow-bge-m3"
|
||||
#schedule = [
|
||||
# "deepseek-v3",
|
||||
# "deepseek-r1",
|
||||
#]
|
||||
[model.utils] # 在麦麦的一些组件中使用的模型，例如表情包模块，取名模块，关系模块，是麦麦必须的模型
|
||||
model_list = ["siliconflow-deepseek-v3","qwen3-8b"]
|
||||
temperature = 0.2 # 模型温度，新V3建议0.1-0.3
|
||||
max_tokens = 800 # 最大输出token数
|
||||
|
||||
[model.utils_small] # 在麦麦的一些组件中使用的小模型，消耗量较大，建议使用速度较快的小模型
|
||||
model_name = "qwen3-8b" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.7
|
||||
max_tokens = 800
|
||||
|
||||
[model.replyer_1] # 首要回复模型，还用于表达器和表达方式学习
|
||||
model_name = "siliconflow-deepseek-v3" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.2 # 模型温度，新V3建议0.1-0.3
|
||||
max_tokens = 800
|
||||
|
||||
[model.replyer_2] # 次要回复模型
|
||||
model_name = "siliconflow-deepseek-r1" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.7 # 模型温度
|
||||
max_tokens = 800
|
||||
|
||||
[model.planner] #决策：负责决定麦麦该做什么的模型
|
||||
model_name = "siliconflow-deepseek-v3" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.3
|
||||
max_tokens = 800
|
||||
|
||||
[model.emotion] #负责麦麦的情绪变化
|
||||
model_name = "siliconflow-deepseek-v3" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.3
|
||||
max_tokens = 800
|
||||
|
||||
[model.memory] # 记忆模型
|
||||
model_name = "qwen3-30b" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.7
|
||||
max_tokens = 800
|
||||
enable_thinking = false # 是否启用思考
|
||||
|
||||
[model.vlm] # 图像识别模型
|
||||
model_name = "qwen2.5-vl-72b" # 对应 model_config.toml 中的模型名称
|
||||
max_tokens = 800
|
||||
|
||||
[model.voice] # 语音识别模型
|
||||
model_name = "sensevoice-small" # 对应 model_config.toml 中的模型名称
|
||||
|
||||
[model.tool_use] #工具调用模型，需要使用支持工具调用的模型
|
||||
model_name = "qwen3-14b" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.7
|
||||
max_tokens = 800
|
||||
enable_thinking = false # 是否启用思考（qwen3 only）
|
||||
|
||||
#嵌入模型
|
||||
[model.embedding]
|
||||
model_name = "bge-m3" # 对应 model_config.toml 中的模型名称
|
||||
|
||||
#------------LPMM知识库模型------------
|
||||
|
||||
[model.lpmm_entity_extract] # 实体提取模型
|
||||
model_name = "siliconflow-deepseek-v3" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.2
|
||||
max_tokens = 800
|
||||
|
||||
[model.lpmm_rdf_build] # RDF构建模型
|
||||
model_name = "siliconflow-deepseek-v3" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.2
|
||||
max_tokens = 800
|
||||
|
||||
[model.lpmm_qa] # 问答模型
|
||||
model_name = "deepseek-r1-distill-qwen-32b" # 对应 model_config.toml 中的模型名称
|
||||
temperature = 0.7
|
||||
max_tokens = 800
|
||||
enable_thinking = false # 是否启用思考
|
||||
Reference in New Issue
Block a user