|
| 1 | +# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates |
| 2 | +# Licensed under the 【火山方舟】原型应用软件自用许可协议 |
| 3 | +# you may not use this file except in compliance with the License. |
| 4 | +# You may obtain a copy of the License at |
| 5 | +# https://www.volcengine.com/docs/82379/1433703 |
| 6 | +# Unless required by applicable law or agreed to in writing, software |
| 7 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 8 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 9 | +# See the License for the specific language governing permissions and |
| 10 | +# limitations under the License. |
| 11 | + |
| 12 | +""" |
| 13 | +DeepDoubao |
| 14 | +""" |
| 15 | + |
| 16 | +import logging |
| 17 | +import os |
| 18 | +from typing import AsyncIterable, Union |
| 19 | + |
| 20 | +from arkitect.core.component.llm import BaseChatLanguageModel |
| 21 | +from arkitect.core.component.llm.model import ( |
| 22 | + ArkChatCompletionChunk, |
| 23 | + ArkChatParameters, |
| 24 | + ArkChatRequest, |
| 25 | + ArkChatResponse, |
| 26 | + Response, |
| 27 | + ArkMessage, |
| 28 | + BotUsage, |
| 29 | +) |
| 30 | +from arkitect.launcher.local.serve import launch_serve |
| 31 | +from arkitect.telemetry.trace import task |
| 32 | +from volcenginesdkarkruntime.types.completion_usage import ( |
| 33 | + CompletionUsage, |
| 34 | + PromptTokensDetails, |
| 35 | + CompletionTokensDetails, |
| 36 | +) |
| 37 | + |
| 38 | +logger = logging.getLogger(__name__) |
| 39 | + |
| 40 | +DEEPSEEK_R1_ENDPOINT = "<ENDPOINT_ID_FOR_DEEPSEEK_R1>" |
| 41 | +DOUBAO_ENDPOINT = "<ENDPOINT_ID_FOR_DOUBAO>" |
| 42 | + |
| 43 | + |
| 44 | +def merge_usage(usage1: CompletionUsage, usage2: CompletionUsage) -> CompletionUsage: |
| 45 | + usage = CompletionUsage( |
| 46 | + prompt_tokens=usage1.prompt_tokens + usage2.prompt_tokens, |
| 47 | + completion_tokens=usage1.completion_tokens + usage2.completion_tokens, |
| 48 | + total_tokens=usage1.total_tokens + usage2.total_tokens, |
| 49 | + ) |
| 50 | + if usage1.prompt_tokens_details or usage2.prompt_tokens_details: |
| 51 | + usage.prompt_tokens_details = PromptTokensDetails( |
| 52 | + cached_tokens=usage1.prompt_tokens_details.cached_tokens |
| 53 | + + usage2.prompt_tokens_details.cached_tokens |
| 54 | + ) |
| 55 | + if usage1.completion_tokens_details or usage2.completion_tokens_details: |
| 56 | + r1 = usage1.completion_tokens_details.reasoning_tokens |
| 57 | + r2 = usage2.completion_tokens_details.reasoning_tokens |
| 58 | + usage.completion_tokens_details = CompletionTokensDetails( |
| 59 | + reasoning_tokens=r1 if r2 is None else r2 if r1 is None else r1 + r2 |
| 60 | + ) |
| 61 | + return usage |
| 62 | + |
| 63 | + |
| 64 | +@task() |
| 65 | +async def default_model_calling( |
| 66 | + request: ArkChatRequest, |
| 67 | +) -> AsyncIterable[Union[ArkChatCompletionChunk, ArkChatResponse]]: |
| 68 | + parameters_r1 = ArkChatParameters(**request.__dict__) |
| 69 | + parameters_r1.max_tokens = ( |
| 70 | + 1 # Set max_tokens to 1, so R1 model will only output reasoning content. |
| 71 | + ) |
| 72 | + deepseek = BaseChatLanguageModel( |
| 73 | + endpoint_id=DEEPSEEK_R1_ENDPOINT, |
| 74 | + messages=request.messages, |
| 75 | + parameters=parameters_r1, |
| 76 | + ) |
| 77 | + reasoning_content = "" |
| 78 | + reasoning_usage = CompletionUsage( |
| 79 | + completion_tokens=0, |
| 80 | + total_tokens=0, |
| 81 | + prompt_tokens=0, |
| 82 | + ) |
| 83 | + if request.stream: |
| 84 | + async for chunk in deepseek.astream(): |
| 85 | + if chunk.usage: |
| 86 | + reasoning_usage = chunk.usage |
| 87 | + if len(chunk.choices) > 0 and chunk.choices[0].delta.reasoning_content: |
| 88 | + yield chunk |
| 89 | + reasoning_content += chunk.choices[0].delta.reasoning_content |
| 90 | + else: |
| 91 | + response = await deepseek.arun() |
| 92 | + reasoning_content = response.choices[0].message.reasoning_content |
| 93 | + if response.usage: |
| 94 | + reasoning_usage = response.usage |
| 95 | + |
| 96 | + parameters_doubao = ArkChatParameters(**request.__dict__) |
| 97 | + doubao = BaseChatLanguageModel( |
| 98 | + endpoint_id=DOUBAO_ENDPOINT, |
| 99 | + messages=request.messages |
| 100 | + + [ |
| 101 | + ArkMessage( |
| 102 | + role="assistant", |
| 103 | + content="思考过程如下：\n" |
| 104 | + + reasoning_content |
| 105 | + + "\n请根据以上思考过程，给出完整的回答：\n", |
| 106 | + ) |
| 107 | + ], |
| 108 | + parameters=parameters_doubao, |
| 109 | + ) |
| 110 | + if request.stream: |
| 111 | + async for chunk in doubao.astream(): |
| 112 | + if chunk.usage: |
| 113 | + chunk.bot_usage = BotUsage(model_usage=[reasoning_usage, chunk.usage]) |
| 114 | + chunk.usage = merge_usage(chunk.usage, reasoning_usage) |
| 115 | + yield chunk |
| 116 | + else: |
| 117 | + response = await doubao.arun() |
| 118 | + response.choices[0].message.reasoning_content = reasoning_content |
| 119 | + if response.usage: |
| 120 | + response.bot_usage = BotUsage(model_usage=[reasoning_usage, response.usage]) |
| 121 | + response.usage = merge_usage(response.usage, reasoning_usage) |
| 122 | + yield response |
| 123 | + |
| 124 | + |
| 125 | +@task() |
| 126 | +async def main(request: ArkChatRequest) -> AsyncIterable[Response]: |
| 127 | + async for resp in default_model_calling(request): |
| 128 | + yield resp |
| 129 | + |
| 130 | + |
| 131 | +if __name__ == "__main__": |
| 132 | + port = os.getenv("_FAAS_RUNTIME_PORT") |
| 133 | + launch_serve( |
| 134 | + package_path="main", |
| 135 | + port=int(port) if port else 8888, |
| 136 | + health_check_path="/v1/ping", |
| 137 | + endpoint_path="/api/v3/bots/chat/completions", |
| 138 | + ) |
0 commit comments