add anthropic_api

tests/metagpt/provider/mock_llm_config.py

@@ -56,3 +56,7 @@ mock_llm_config_spark = LLMConfig(
 mock_llm_config_qianfan = LLMConfig(api_type="qianfan", access_key="xxx", secret_key="xxx", model="ERNIE-Bot-turbo")
 
 mock_llm_config_dashscope = LLMConfig(api_type="dashscope", api_key="xxx", model="qwen-max")
+
+mock_llm_config_anthropic = LLMConfig(
+    api_type="anthropic", api_key="xxx", base_url="https://api.anthropic.com", model="claude-3-opus-20240229"
+)

tests/metagpt/provider/req_resp_const.py

@@ -3,6 +3,14 @@
 # @Desc   : default request & response data for provider unittest
 
 
+from anthropic.types import (
+    ContentBlock,
+    ContentBlockDeltaEvent,
+    Message,
+    MessageStartEvent,
+    TextDelta,
+)
+from anthropic.types import Usage as AnthropicUsage
 from dashscope.api_entities.dashscope_response import (
     DashScopeAPIResponse,
     GenerationOutput,
@@ -130,6 +138,38 @@ def get_dashscope_response(name: str) -> GenerationResponse:
     )
 
 
+# For Anthropic
+def get_anthropic_response(name: str, stream: bool = False) -> Message:
+    if stream:
+        return [
+            MessageStartEvent(
+                message=Message(
+                    id="xxx",
+                    model=name,
+                    role="assistant",
+                    type="message",
+                    content=[ContentBlock(text="", type="text")],
+                    usage=AnthropicUsage(input_tokens=10, output_tokens=10),
+                ),
+                type="message_start",
+            ),
+            ContentBlockDeltaEvent(
+                index=0,
+                delta=TextDelta(text=resp_cont_tmpl.format(name=name), type="text_delta"),
+                type="content_block_delta",
+            ),
+        ]
+    else:
+        return Message(
+            id="xxx",
+            model=name,
+            role="assistant",
+            type="message",
+            content=[ContentBlock(text=resp_cont_tmpl.format(name=name), type="text")],
+            usage=AnthropicUsage(input_tokens=10, output_tokens=10),
+        )
+
+
 # For llm general chat functions call
 async def llm_general_chat_funcs_test(llm: BaseLLM, prompt: str, messages: list[dict], resp_cont: str):
     resp = await llm.aask(prompt, stream=False)

tests/metagpt/provider/test_anthropic_api.py

@@ -2,31 +2,45 @@
 # -*- coding: utf-8 -*-
 # @Desc   : the unittest of Claude2
 
-
 import pytest
 from anthropic.resources.completions import Completion
 
-from metagpt.provider.anthropic_api import Claude2
-from tests.metagpt.provider.mock_llm_config import mock_llm_config
-from tests.metagpt.provider.req_resp_const import prompt, resp_cont_tmpl
+from metagpt.provider.anthropic_api import AnthropicLLM
+from tests.metagpt.provider.mock_llm_config import mock_llm_config_anthropic
+from tests.metagpt.provider.req_resp_const import (
+    get_anthropic_response,
+    llm_general_chat_funcs_test,
+    messages,
+    prompt,
+    resp_cont_tmpl,
+)
 
-resp_cont = resp_cont_tmpl.format(name="Claude")
+name = "claude-3-opus-20240229"
+resp_cont = resp_cont_tmpl.format(name=name)
 
 
-def mock_anthropic_completions_create(self, model: str, prompt: str, max_tokens_to_sample: int) -> Completion:
-    return Completion(id="xx", completion=resp_cont, model="claude-2", stop_reason="stop_sequence", type="completion")
+async def mock_anthropic_messages_create(
+    self, messages: list[dict], model: str, stream: bool = True, max_tokens: int = None, system: str = None
+) -> Completion:
+    if stream:
 
+        async def aresp_iterator():
+            resps = get_anthropic_response(name, stream=True)
+            for resp in resps:
+                yield resp
 
-async def mock_anthropic_acompletions_create(self, model: str, prompt: str, max_tokens_to_sample: int) -> Completion:
-    return Completion(id="xx", completion=resp_cont, model="claude-2", stop_reason="stop_sequence", type="completion")
-
-
-def test_claude2_ask(mocker):
-    mocker.patch("anthropic.resources.completions.Completions.create", mock_anthropic_completions_create)
-    assert resp_cont == Claude2(mock_llm_config).ask(prompt)
+        return aresp_iterator()
+    else:
+        return get_anthropic_response(name)
 
 
 @pytest.mark.asyncio
-async def test_claude2_aask(mocker):
-    mocker.patch("anthropic.resources.completions.AsyncCompletions.create", mock_anthropic_acompletions_create)
-    assert resp_cont == await Claude2(mock_llm_config).aask(prompt)
+async def test_anthropic_acompletion(mocker):
+    mocker.patch("anthropic.resources.messages.AsyncMessages.create", mock_anthropic_messages_create)
+
+    anthropic_llm = AnthropicLLM(mock_llm_config_anthropic)
+
+    resp = await anthropic_llm.acompletion(messages)
+    assert resp.content[0].text == resp_cont
+
+    await llm_general_chat_funcs_test(anthropic_llm, prompt, messages, resp_cont)

tests/metagpt/provider/test_base_llm.py

@@ -27,9 +27,15 @@ class MockBaseLLM(BaseLLM):
     def completion(self, messages: list[dict], timeout=3):
         return get_part_chat_completion(name)
 
+    async def _achat_completion(self, messages: list[dict], timeout=3):
+        pass
+
     async def acompletion(self, messages: list[dict], timeout=3):
         return get_part_chat_completion(name)
 
+    async def _achat_completion_stream(self, messages: list[dict], timeout: int = 3) -> str:
+        pass
+
     async def acompletion_text(self, messages: list[dict], stream=False, timeout=3) -> str:
         return default_resp_cont