File size: 14,691 Bytes
f8d0193
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
import asyncio
import json
import os
from typing import Dict, List, Optional, Union

import anthropic
import httpcore
import httpx
from anthropic import NOT_GIVEN
from requests.exceptions import ProxyError

from .base_api import AsyncBaseAPILLM, BaseAPILLM


class ClaudeAPI(BaseAPILLM):

    is_api: bool = True

    def __init__(
        self,
        model_type: str = 'claude-3-5-sonnet-20241022',
        retry: int = 5,
        key: Union[str, List[str]] = 'ENV',
        proxies: Optional[Dict] = None,
        meta_template: Optional[Dict] = [
            dict(role='system', api_role='system'),
            dict(role='user', api_role='user'),
            dict(role='assistant', api_role='assistant'),
            dict(role='environment', api_role='user'),
        ],
        temperature: float = NOT_GIVEN,
        max_new_tokens: int = 512,
        top_p: float = NOT_GIVEN,
        top_k: int = NOT_GIVEN,
        repetition_penalty: float = 0.0,
        stop_words: Union[List[str], str] = None,
    ):

        super().__init__(
            meta_template=meta_template,
            model_type=model_type,
            retry=retry,
            temperature=temperature,
            max_new_tokens=max_new_tokens,
            top_p=top_p,
            top_k=top_k,
            repetition_penalty=repetition_penalty,
            stop_words=stop_words,
        )

        key = os.getenv('Claude_API_KEY') if key == 'ENV' else key

        if isinstance(key, str):
            self.keys = [key]
        else:
            self.keys = list(set(key))
        self.clients = {key: anthropic.AsyncAnthropic(proxies=proxies, api_key=key) for key in self.keys}

        # record invalid keys and skip them when requesting API
        # - keys have insufficient_quota
        self.invalid_keys = set()

        self.key_ctr = 0
        self.model_type = model_type
        self.proxies = proxies

    def chat(
        self,
        inputs: Union[List[dict], List[List[dict]]],
        session_ids: Union[int, List[int]] = None,
        **gen_params,
    ) -> Union[str, List[str]]:
        """Generate responses given the contexts.

        Args:
            inputs (Union[List[dict], List[List[dict]]]): a list of messages
                or list of lists of messages
            gen_params: additional generation configuration

        Returns:
            Union[str, List[str]]: generated string(s)
        """
        assert isinstance(inputs, list)
        gen_params = {**self.gen_params, **gen_params}
        import nest_asyncio

        nest_asyncio.apply()

        async def run_async_tasks():
            tasks = [
                self._chat(self.template_parser(messages), **gen_params)
                for messages in ([inputs] if isinstance(inputs[0], dict) else inputs)
            ]
            return await asyncio.gather(*tasks)

        try:
            loop = asyncio.get_running_loop()
            # If the event loop is already running, schedule the task
            future = asyncio.ensure_future(run_async_tasks())
            ret = loop.run_until_complete(future)
        except RuntimeError:
            # If no running event loop, start a new one
            ret = asyncio.run(run_async_tasks())
        return ret[0] if isinstance(inputs[0], dict) else ret

    def generate_request_data(self, model_type, messages, gen_params):
        """
        Generates the request data for different model types.

        Args:
            model_type (str): The type of the model (e.g., 'gpt', 'internlm', 'qwen').
            messages (list): The list of messages to be sent to the model.
            gen_params (dict): The generation parameters.
            json_mode (bool): Flag to determine if the response format should be JSON.

        Returns:
            tuple: A tuple containing the header and the request data.
        """
        # Copy generation parameters to avoid modifying the original dictionary
        gen_params = gen_params.copy()

        # Hold out 100 tokens due to potential errors in token calculation
        max_tokens = min(gen_params.pop('max_new_tokens'), 4096)
        if max_tokens <= 0:
            return '', ''
        gen_params.pop('repetition_penalty')
        if 'stop_words' in gen_params:
            gen_params['stop_sequences'] = gen_params.pop('stop_words')
        # Common parameters processing
        gen_params['max_tokens'] = max_tokens
        gen_params.pop('skip_special_tokens', None)
        gen_params.pop('session_id', None)

        system = None
        if messages[0]['role'] == 'system':
            system = messages.pop(0)
            system = system['content']
        for message in messages:
            message.pop('name', None)
        data = {'model': model_type, 'messages': messages, **gen_params}
        if system:
            data['system'] = system
        return data

    async def _chat(self, messages: List[dict], **gen_params) -> str:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """
        assert isinstance(messages, list)

        data = self.generate_request_data(model_type=self.model_type, messages=messages, gen_params=gen_params)
        max_num_retries = 0

        while max_num_retries < self.retry:
            if len(self.invalid_keys) == len(self.keys):
                raise RuntimeError('All keys have insufficient quota.')
            # find the next valid key
            while True:
                self.key_ctr += 1
                if self.key_ctr == len(self.keys):
                    self.key_ctr = 0

                if self.keys[self.key_ctr] not in self.invalid_keys:
                    break

            key = self.keys[self.key_ctr]
            client = self.clients[key]

            try:
                response = await client.messages.create(**data)
                response = json.loads(response.json())
                return response['content'][0]['text'].strip()
            except (anthropic.RateLimitError, anthropic.APIConnectionError) as e:
                print(f'API请求错误: {e}')
                await asyncio.sleep(5)

            except (httpcore.ProxyError, ProxyError) as e:

                print(f'代理服务器错误: {e}')
                await asyncio.sleep(5)
            except httpx.TimeoutException as e:
                print(f'请求超时: {e}')
                await asyncio.sleep(5)

            except KeyboardInterrupt:
                raise

            except Exception as error:
                if error.body['error']['message'] == 'invalid x-api-key':
                    self.invalid_keys.add(key)
                    self.logger.warn(f'invalid key: {key}')
                elif error.body['error']['type'] == 'overloaded_error':
                    await asyncio.sleep(5)
                elif error.body['error']['message'] == 'Internal server error':
                    await asyncio.sleep(5)
                elif error.body['error']['message'] == (
                    'Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to '
                    'upgrade or purchase credits.'
                ):
                    self.invalid_keys.add(key)
                    print(f'API has no quota: {key}, Valid keys: {len(self.keys) - len(self.invalid_keys)}')
            max_num_retries += 1

        raise RuntimeError(
            'Calling Claude failed after retrying for ' f'{max_num_retries} times. Check the logs for ' 'details.'
        )


class AsyncClaudeAPI(AsyncBaseAPILLM):

    is_api: bool = True

    def __init__(
        self,
        model_type: str = 'claude-3-5-sonnet-20241022',
        retry: int = 5,
        key: Union[str, List[str]] = 'ENV',
        proxies: Optional[Dict] = None,
        meta_template: Optional[Dict] = [
            dict(role='system', api_role='system'),
            dict(role='user', api_role='user'),
            dict(role='assistant', api_role='assistant'),
            dict(role='environment', api_role='user'),
        ],
        temperature: float = NOT_GIVEN,
        max_new_tokens: int = 512,
        top_p: float = NOT_GIVEN,
        top_k: int = NOT_GIVEN,
        repetition_penalty: float = 0.0,
        stop_words: Union[List[str], str] = None,
    ):

        super().__init__(
            model_type=model_type,
            retry=retry,
            meta_template=meta_template,
            temperature=temperature,
            max_new_tokens=max_new_tokens,
            top_p=top_p,
            top_k=top_k,
            repetition_penalty=repetition_penalty,
            stop_words=stop_words,
        )

        key = os.getenv('Claude_API_KEY') if key == 'ENV' else key

        if isinstance(key, str):
            self.keys = [key]
        else:
            self.keys = list(set(key))
        self.clients = {key: anthropic.AsyncAnthropic(proxies=proxies, api_key=key) for key in self.keys}

        # record invalid keys and skip them when requesting API
        # - keys have insufficient_quota
        self.invalid_keys = set()

        self.key_ctr = 0
        self.model_type = model_type
        self.proxies = proxies

    async def chat(
        self,
        inputs: Union[List[dict], List[List[dict]]],
        session_ids: Union[int, List[int]] = None,
        **gen_params,
    ) -> Union[str, List[str]]:
        """Generate responses given the contexts.

        Args:
            inputs (Union[List[dict], List[List[dict]]]): a list of messages
                or list of lists of messages
            gen_params: additional generation configuration

        Returns:
            Union[str, List[str]]: generated string(s)
        """
        assert isinstance(inputs, list)
        gen_params = {**self.gen_params, **gen_params}
        tasks = [
            self._chat(messages, **gen_params) for messages in ([inputs] if isinstance(inputs[0], dict) else inputs)
        ]
        ret = await asyncio.gather(*tasks)
        return ret[0] if isinstance(inputs[0], dict) else ret

    def generate_request_data(self, model_type, messages, gen_params):
        """
        Generates the request data for different model types.

        Args:
            model_type (str): The type of the model (e.g., 'gpt', 'internlm', 'qwen').
            messages (list): The list of messages to be sent to the model.
            gen_params (dict): The generation parameters.
            json_mode (bool): Flag to determine if the response format should be JSON.

        Returns:
            tuple: A tuple containing the header and the request data.
        """
        # Copy generation parameters to avoid modifying the original dictionary
        gen_params = gen_params.copy()

        # Hold out 100 tokens due to potential errors in token calculation
        max_tokens = min(gen_params.pop('max_new_tokens'), 4096)
        if max_tokens <= 0:
            return '', ''
        gen_params.pop('repetition_penalty')
        if 'stop_words' in gen_params:
            gen_params['stop_sequences'] = gen_params.pop('stop_words')
        # Common parameters processing
        gen_params['max_tokens'] = max_tokens
        gen_params.pop('skip_special_tokens', None)
        gen_params.pop('session_id', None)

        system = None
        if messages[0]['role'] == 'system':
            system = messages.pop(0)
            system = system['content']
        for message in messages:
            message.pop('name', None)
        data = {'model': model_type, 'messages': messages, **gen_params}
        if system:
            data['system'] = system
        return data

    async def _chat(self, messages: List[dict], **gen_params) -> str:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """
        assert isinstance(messages, list)
        messages = self.template_parser(messages)
        data = self.generate_request_data(model_type=self.model_type, messages=messages, gen_params=gen_params)
        max_num_retries = 0

        while max_num_retries < self.retry:
            if len(self.invalid_keys) == len(self.keys):
                raise RuntimeError('All keys have insufficient quota.')
            # find the next valid key
            while True:
                self.key_ctr += 1
                if self.key_ctr == len(self.keys):
                    self.key_ctr = 0

                if self.keys[self.key_ctr] not in self.invalid_keys:
                    break

            key = self.keys[self.key_ctr]
            client = self.clients[key]

            try:
                response = await client.messages.create(**data)
                response = json.loads(response.json())
                return response['content'][0]['text'].strip()
            except (anthropic.RateLimitError, anthropic.APIConnectionError) as e:
                print(f'API请求错误: {e}')
                await asyncio.sleep(5)

            except (httpcore.ProxyError, ProxyError) as e:

                print(f'代理服务器错误: {e}')
                await asyncio.sleep(5)
            except httpx.TimeoutException as e:
                print(f'请求超时: {e}')
                await asyncio.sleep(5)

            except KeyboardInterrupt:
                raise

            except Exception as error:
                if error.body['error']['message'] == 'invalid x-api-key':
                    self.invalid_keys.add(key)
                    self.logger.warn(f'invalid key: {key}')
                elif error.body['error']['type'] == 'overloaded_error':
                    await asyncio.sleep(5)
                elif error.body['error']['message'] == 'Internal server error':
                    await asyncio.sleep(5)
                elif error.body['error']['message'] == (
                    'Your credit balance is too low to access the Anthropic API. Please go to Plans & Billing to'
                    ' upgrade or purchase credits.'
                ):
                    self.invalid_keys.add(key)
                    print(f'API has no quota: {key}, Valid keys: {len(self.keys) - len(self.invalid_keys)}')
                else:
                    raise error
            max_num_retries += 1

        raise RuntimeError(
            'Calling Claude failed after retrying for ' f'{max_num_retries} times. Check the logs for ' 'details.'
        )