[API] Update API (#624)

* update api

* update generation_kwargs impl

* update api

* refactor

---------
Co-authored-by: Leymore <zfz-960727@163.com>

configs/eval_api_360.py → configs/api_examples/eval_api_360.py

@@ -5,8 +5,8 @@ from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,
@@ -33,4 +33,4 @@ infer = dict(
         task=dict(type=OpenICLInferTask)),
 )
 
-work_dir ="./output/360GPT_S2_V9"
+work_dir ="./output/api_360GPT_S2_V9"
\ No newline at end of file

configs/eval_api_baichuan.py → configs/api_examples/eval_api_baichuan.py

 from mmengine.config import read_base
 from opencompass.models import BaiChuan
-
 from opencompass.partitioners import NaivePartitioner
 from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,

configs/api_examples/eval_api_baidu.py

+from mmengine.config import read_base
+from opencompass.models import ERNIEBot
+from opencompass.partitioners import NaivePartitioner
+from opencompass.runners.local_api import LocalAPIRunner
+from opencompass.tasks import OpenICLInferTask
+
+with read_base():
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
+
+datasets = [
+    *ceval_datasets,
+]
+
+models = [
+    dict(
+        abbr='erniebot',
+        type=ERNIEBot,
+        path='erniebot',
+        key='xxxxxx',  # please give you key
+        secretkey='xxxxxxxxx',  # please give your group_id
+        url='xxxxxxxxx',
+        query_per_second=1,
+        max_out_len=2048,
+        max_seq_len=2048,
+        batch_size=8),
+]
+
+infer = dict(
+    partitioner=dict(type=NaivePartitioner),
+    runner=dict(
+        type=LocalAPIRunner,
+        max_num_workers=2,
+        concurrent_users=2,
+        task=dict(type=OpenICLInferTask)),
+)
+
+work_dir = "outputs/api_erniebot/"
\ No newline at end of file

configs/api_examples/eval_api_bytedance.py

+from mmengine.config import read_base
+from opencompass.models import ByteDance
+from opencompass.partitioners import NaivePartitioner
+from opencompass.runners.local_api import LocalAPIRunner
+from opencompass.tasks import OpenICLInferTask
+
+with read_base():
+    # from .datasets.collections.chat_medium import datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
+
+datasets = [
+    *ceval_datasets,
+]
+
+models = [
+    dict(
+        abbr='skylark-pro-public',
+        type=ByteDance,
+        path='skylark-pro-public',
+        accesskey="xxxxxxx",
+        secretkey="xxxxxxx",
+        url='xxxxxx',
+        query_per_second=1,
+        max_out_len=2048,
+        max_seq_len=2048,
+        batch_size=8),
+]
+
+infer = dict(
+    partitioner=dict(type=NaivePartitioner),
+    runner=dict(
+        type=LocalAPIRunner,
+        max_num_workers=2,
+        concurrent_users=2,
+        task=dict(type=OpenICLInferTask)),
+)
+
+work_dir = "outputs/api_bytedance/"
\ No newline at end of file

configs/eval_minimax.py → configs/api_examples/eval_api_minimax.py

 from mmengine.config import read_base
 from opencompass.models import MiniMax
 from opencompass.partitioners import NaivePartitioner
-from opencompass.runners import LocalRunner
 from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
-    # from .datasets.collections.chat_medium import datasets
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,
@@ -34,4 +32,6 @@ infer = dict(
         max_num_workers=4,
         concurrent_users=4,
         task=dict(type=OpenICLInferTask)),
-)
+)
\ No newline at end of file
+
+work_dir = "outputs/api_minimax/"
\ No newline at end of file

configs/api_examples/eval_api_moonshot.py

+from mmengine.config import read_base
+from opencompass.models import MoonShot
+from opencompass.partitioners import NaivePartitioner
+from opencompass.runners.local_api import LocalAPIRunner
+from opencompass.tasks import OpenICLInferTask
+
+with read_base():
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
+
+datasets = [
+    *ceval_datasets,
+]
+
+models = [
+    dict(
+        abbr='moonshot-v1-32k',
+        type=MoonShot,
+        path='moonshot-v1-32k',
+        key='xxxxxxx',
+        url= 'xxxxxxxx',
+        query_per_second=1,
+        max_out_len=2048,
+        max_seq_len=2048,
+        batch_size=8),
+]
+
+infer = dict(
+    partitioner=dict(type=NaivePartitioner),
+    runner=dict(
+        type=LocalAPIRunner,
+        max_num_workers=4,
+        concurrent_users=4,
+        task=dict(type=OpenICLInferTask)),
+)
+
+work_dir = "outputs/api_moonshot/"
\ No newline at end of file

configs/eval_api_pangu.py → configs/api_examples/eval_api_pangu.py

 from mmengine.config import read_base
 from opencompass.models import PanGu
-
 from opencompass.partitioners import NaivePartitioner
 from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,

configs/eval_api_sensetime.py → configs/api_examples/eval_api_sensetime.py

@@ -5,8 +5,8 @@ from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,
@@ -33,3 +33,5 @@ infer = dict(
         concurrent_users=2,
         task=dict(type=OpenICLInferTask)),
 )
+
+work_dir = "outputs/api_sensetime/"
\ No newline at end of file

configs/eval_xunfei.py → configs/api_examples/eval_api_xunfei.py

 from mmengine.config import read_base
 from opencompass.models.xunfei_api import XunFei
 from opencompass.partitioners import NaivePartitioner
-from opencompass.runners import LocalRunner
 from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
     # from .datasets.collections.chat_medium import datasets
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,
@@ -47,4 +46,6 @@ infer = dict(
         max_num_workers=2,
         concurrent_users=2,
         task=dict(type=OpenICLInferTask)),
-)
+)
\ No newline at end of file
+
+work_dir = "outputs/api_xunfei/"
\ No newline at end of file

configs/eval_zhipu.py → configs/api_examples/eval_api_zhipu.py

 from mmengine.config import read_base
 from opencompass.models import ZhiPuAI
 from opencompass.partitioners import NaivePartitioner
-from opencompass.runners import LocalRunner
 from opencompass.runners.local_api import LocalAPIRunner
 from opencompass.tasks import OpenICLInferTask
 
 with read_base():
     # from .datasets.collections.chat_medium import datasets
-    from .summarizers.medium import summarizer
-    from .datasets.ceval.ceval_gen import ceval_datasets
+    from ..summarizers.medium import summarizer
+    from ..datasets.ceval.ceval_gen import ceval_datasets
 
 datasets = [
     *ceval_datasets,
@@ -33,4 +32,6 @@ infer = dict(
         max_num_workers=2,
         concurrent_users=2,
         task=dict(type=OpenICLInferTask)),
-)
+)
\ No newline at end of file
+
+work_dir = "outputs/api_zhipu/"
\ No newline at end of file

opencompass/models/__init__.py

@@ -15,6 +15,7 @@ from .lightllm_api import LightllmAPI  # noqa: F401
 from .llama2 import Llama2, Llama2Chat  # noqa: F401, F403
 from .minimax_api import MiniMax  # noqa: F401
 from .modelscope import ModelScope, ModelScopeCausalLM  # noqa: F401, F403
+from .moonshot_api import MoonShot  # noqa: F401
 from .openai_api import OpenAI  # noqa: F401
 from .pangu_api import PanGu  # noqa: F401
 from .sensetime_api import SenseTime  # noqa: F401

opencompass/models/ai360_api.py

@@ -160,7 +160,6 @@ class AI360GPT(BaseAPIModel):
                     or raw_response.status_code == 429
                     or raw_response.status_code == 500):
                 print(raw_response.text)
-                # return ''
                 continue
             print(raw_response)
             max_num_retries += 1

opencompass/models/baichuan_api.py

@@ -150,14 +150,14 @@ class BaiChuan(BaseAPIModel):
                 self.wait()
                 continue
             if raw_response.status_code == 200 and response['code'] == 0:
-                # msg = json.load(response.text)
-                # response
+
                 msg = response['data']['messages'][0]['content']
                 return msg
 
             if response['code'] != 0:
                 print(response)
-                return ''
+                time.sleep(1)
+                continue
             print(response)
             max_num_retries += 1
 

opencompass/models/bytedance_api.py

+import time
 from concurrent.futures import ThreadPoolExecutor
 from typing import Dict, List, Optional, Union
 
@@ -165,8 +166,9 @@ class ByteDance(BaseAPIModel):
 
             if isinstance(response, MaasException):
                 print(response)
-                return ''
-            print(response)
+                time.sleep(1)
+                continue
+
             max_num_retries += 1
 
         raise RuntimeError(response)

opencompass/models/minimax_api.py

+import time
 from concurrent.futures import ThreadPoolExecutor
 from typing import Dict, List, Optional, Union
 
@@ -168,7 +169,8 @@ class MiniMax(BaseAPIModel):
                     or response.status_code == 1039
                     or response.status_code == 2013):
                 print(response.text)
-                return ''
+                time.sleep(1)
+                continue
             print(response)
             max_num_retries += 1
 

opencompass/models/moonshot_api.py

+import time
+from concurrent.futures import ThreadPoolExecutor
+from typing import Dict, List, Optional, Union
+
+import requests
+
+from opencompass.utils.prompt import PromptList
+
+from .base_api import BaseAPIModel
+
+PromptType = Union[PromptList, str]
+
+
+class MoonShot(BaseAPIModel):
+    """Model wrapper around MoonShot.
+
+    Documentation:
+
+    Args:
+        path (str): The name of MoonShot model.
+            e.g. `moonshot-v1-32k`
+        key (str): Authorization key.
+        query_per_second (int): The maximum queries allowed per second
+            between two consecutive calls of the API. Defaults to 1.
+        max_seq_len (int): Unused here.
+        meta_template (Dict, optional): The model's meta prompt
+            template if needed, in case the requirement of injecting or
+            wrapping of any meta instructions.
+        retry (int): Number of retires if the API call fails. Defaults to 2.
+    """
+
+    def __init__(
+        self,
+        path: str,
+        key: str,
+        url: str,
+        query_per_second: int = 2,
+        max_seq_len: int = 2048,
+        meta_template: Optional[Dict] = None,
+        retry: int = 2,
+    ):
+        super().__init__(path=path,
+                         max_seq_len=max_seq_len,
+                         query_per_second=query_per_second,
+                         meta_template=meta_template,
+                         retry=retry)
+        self.headers = {
+            'Content-Type': 'application/json',
+            'Authorization': 'Bearer ' + key,
+        }
+        self.url = url
+        self.model = path
+
+    def generate(
+        self,
+        inputs: List[str or PromptList],
+        max_out_len: int = 512,
+    ) -> List[str]:
+        """Generate results given a list of inputs.
+
+        Args:
+            inputs (List[str or PromptList]): A list of strings or PromptDicts.
+                The PromptDict should be organized in OpenCompass'
+                API format.
+            max_out_len (int): The maximum length of the output.
+
+        Returns:
+            List[str]: A list of generated strings.
+        """
+        with ThreadPoolExecutor() as executor:
+            results = list(
+                executor.map(self._generate, inputs,
+                             [max_out_len] * len(inputs)))
+        self.flush()
+        return results
+
+    def _generate(
+        self,
+        input: str or PromptList,
+        max_out_len: int = 512,
+    ) -> str:
+        """Generate results given an input.
+
+        Args:
+            inputs (str or PromptList): A string or PromptDict.
+                The PromptDict should be organized in OpenCompass'
+                API format.
+            max_out_len (int): The maximum length of the output.
+
+        Returns:
+            str: The generated string.
+        """
+        assert isinstance(input, (str, PromptList))
+
+        if isinstance(input, str):
+            messages = [{'role': 'user', 'content': input}]
+        else:
+            messages = []
+            for item in input:
+                msg = {'content': item['prompt']}
+                if item['role'] == 'HUMAN':
+                    msg['role'] = 'user'
+                elif item['role'] == 'BOT':
+                    msg['role'] = 'assistant'
+
+                messages.append(msg)
+
+        system = {
+            'role':
+            'system',
+            'content':
+            '你是 Kimi，由 Moonshot AI 提供的人工智能助手，你更擅长中文和英文的对话。'
+            '你会为用户提供安全，有帮助，准确的回答。同时，你会拒绝一些涉及恐怖主义，种族歧视，'
+            '黄色暴力等问题的回答。Moonshot AI 为专有名词，不可翻译成其他语言。'
+        }
+
+        messages.insert(0, system)
+
+        data = {
+            'model': self.model,
+            'messages': messages,
+        }
+
+        max_num_retries = 0
+        while max_num_retries < self.retry:
+            self.acquire()
+            raw_response = requests.request('POST',
+                                            url=self.url,
+                                            headers=self.headers,
+                                            json=data)
+
+            response = raw_response.json()
+            self.release()
+
+            if response is None:
+                print('Connection error, reconnect.')
+                # if connect error, frequent requests will casuse
+                # continuous unstable network, therefore wait here
+                # to slow down the request
+                self.wait()
+                continue
+
+            if raw_response.status_code == 200:
+                # msg = json.load(response.text)
+                # response
+                msg = response['choices'][0]['message']['content']
+                return msg
+
+            if raw_response.status_code == 403:
+                print('请求被拒绝 api_key错误')
+                continue
+            elif raw_response.status_code == 400:
+                print('请求失败，状态码:', raw_response)
+                time.sleep(1)
+                continue
+            elif raw_response.status_code == 429:
+                print('请求失败，状态码:', raw_response)
+                time.sleep(3)
+                continue
+
+            max_num_retries += 1
+
+        raise RuntimeError(raw_response)

opencompass/models/pangu_api.py

+import time
 from concurrent.futures import ThreadPoolExecutor
 from typing import Dict, List, Optional, Union
 
@@ -56,6 +57,14 @@ class PanGu(BaseAPIModel):
         self.project_name = project_name
         self.model = path
 
+        token_response = self._get_token()
+        if token_response.status_code == 201:
+            self.token = token_response.headers['X-Subject-Token']
+            print('请求成功！')
+        else:
+            self.token = None
+            print('token生成失败')
+
     def generate(
         self,
         inputs: List[str or PromptList],
@@ -139,16 +148,18 @@ class PanGu(BaseAPIModel):
 
         data = {'messages': messages, 'stream': False}
 
-        token_response = self._get_token()
-        if token_response.status_code == 201:
-            token = token_response.headers['X-Subject-Token']
-            print('请求成功！')
-        else:
-            msg = 'token生成失败'
-            print(msg)
-            return ''
-
-        headers = {'Content-Type': 'application/json', 'X-Auth-Token': token}
+        # token_response = self._get_token()
+        # if token_response.status_code == 201:
+        #     self.token = token_response.headers['X-Subject-Token']
+        #     print('请求成功！')
+        # else:
+        #     self.token = None
+        #     print('token生成失败')
+
+        headers = {
+            'Content-Type': 'application/json',
+            'X-Auth-Token': self.token
+        }
 
         max_num_retries = 0
         while max_num_retries < self.retry:
@@ -175,7 +186,9 @@ class PanGu(BaseAPIModel):
 
             if (raw_response.status_code != 200):
                 print(response['error_msg'])
-                return ''
+                # return ''
+                time.sleep(1)
+                continue
             print(response)
             max_num_retries += 1
 

opencompass/models/sensetime_api.py

@@ -127,9 +127,14 @@ class SenseTime(BaseAPIModel):
                 return msg
 
             if (raw_response.status_code != 200):
-                print(raw_response.text)
-                time.sleep(1)
-                continue
+                if response['error']['code'] == 18:
+                    # security issue
+                    return 'error:unsafe'
+                else:
+                    print(raw_response.text)
+                    time.sleep(1)
+                    continue
+
             print(response)
             max_num_retries += 1