fix_bug_for_config_model

tests/metagpt/provider/mock_llm_config.py

@@ -15,6 +15,7 @@ mock_llm_config = LLMConfig(
     app_id="mock_app_id",
     api_secret="mock_api_secret",
     domain="mock_domain",
+    model="mock_model",
 )
 
 

tests/metagpt/provider/test_base_llm.py

@@ -8,7 +8,6 @@
 
 import pytest
 
-from metagpt.configs.compress_msg_config import CompressType
 from metagpt.configs.llm_config import LLMConfig
 from metagpt.const import IMAGES
 from metagpt.provider.base_llm import BaseLLM
@@ -26,6 +25,7 @@ name = "GPT"
 class MockBaseLLM(BaseLLM):
     def __init__(self, config: LLMConfig = None):
         self.config = config or mock_llm_config
+        self.model = mock_llm_config.model
 
     def completion(self, messages: list[dict], timeout=3):
         return get_part_chat_completion(name)
@@ -108,64 +108,6 @@ async def test_async_base_llm():
     # assert resp == default_resp_cont
 
 
-@pytest.mark.parametrize("compress_type", list(CompressType))
-def test_compress_messages_no_effect(compress_type):
-    base_llm = MockBaseLLM()
-    messages = [
-        {"role": "system", "content": "first system msg"},
-        {"role": "system", "content": "second system msg"},
-    ]
-    for i in range(5):
-        messages.append({"role": "user", "content": f"u{i}"})
-        messages.append({"role": "assistant", "content": f"a{i}"})
-    compressed = base_llm.compress_messages(messages, compress_type=compress_type)
-    # should take no effect for short context
-    assert compressed == messages
-
-
-@pytest.mark.parametrize("compress_type", CompressType.cut_types())
-def test_compress_messages_long(compress_type):
-    base_llm = MockBaseLLM()
-    base_llm.config.model = "test_llm"
-    max_token_limit = 100
-
-    messages = [
-        {"role": "system", "content": "first system msg"},
-        {"role": "system", "content": "second system msg"},
-    ]
-    for i in range(100):
-        messages.append({"role": "user", "content": f"u{i}" * 10})  # ~2x10x0.5 = 10 tokens
-        messages.append({"role": "assistant", "content": f"a{i}" * 10})
-    compressed = base_llm.compress_messages(messages, compress_type=compress_type, max_token=max_token_limit)
-
-    print(compressed)
-    print(len(compressed))
-    assert 3 <= len(compressed) < len(messages)
-    assert compressed[0]["role"] == "system" and compressed[1]["role"] == "system"
-    assert compressed[2]["role"] != "system"
-
-
-def test_long_messages_no_compress():
-    base_llm = MockBaseLLM()
-    messages = [{"role": "user", "content": "1" * 10000}] * 10000
-    compressed = base_llm.compress_messages(messages)
-    assert len(compressed) == len(messages)
-
-
-@pytest.mark.parametrize("compress_type", CompressType.cut_types())
-def test_compress_messages_long_no_sys_msg(compress_type):
-    base_llm = MockBaseLLM()
-    base_llm.config.model = "test_llm"
-    max_token_limit = 100
-
-    messages = [{"role": "user", "content": "1" * 10000}]
-    compressed = base_llm.compress_messages(messages, compress_type=compress_type, max_token=max_token_limit)
-
-    print(compressed)
-    assert compressed
-    assert len(compressed[0]["content"]) < len(messages[0]["content"])
-
-
 def test_format_msg(mocker):
     base_llm = MockBaseLLM()
     messages = [UserMessage(content="req"), AIMessage(content="rsp")]
@@ -175,7 +117,7 @@ def test_format_msg(mocker):
 
 def test_format_msg_w_images(mocker):
     base_llm = MockBaseLLM()
-    base_llm.config.model = "gpt-4o"
+    base_llm.model = "gpt-4o"
     msg_w_images = UserMessage(content="req1")
     msg_w_images.add_metadata(IMAGES, ["base64 string 1", "base64 string 2"])
     msg_w_empty_images = UserMessage(content="req2")

tests/metagpt/provider/test_bedrock_api.py

@@ -22,13 +22,13 @@ usage = {
 }
 
 
-def mock_invoke_model(self: BedrockLLM, *args, **kwargs) -> dict:
+async def mock_invoke_model(self: BedrockLLM, *args, **kwargs) -> dict:
     provider = self.config.model.split(".")[0]
     self._update_costs(usage, self.config.model)
     return BEDROCK_PROVIDER_RESPONSE_BODY[provider]
 
 
-def mock_invoke_model_stream(self: BedrockLLM, *args, **kwargs) -> dict:
+async def mock_invoke_model_stream(self: BedrockLLM, *args, **kwargs) -> dict:
     # use json object to mock EventStream
     def dict2bytes(x):
         return json.dumps(x).encode("utf-8")

tests/metagpt/provider/test_openai.py

@@ -169,7 +169,7 @@ async def test_openai_acompletion(mocker):
 
 def test_count_tokens():
     llm = LLM()
-    llm.config.model = "gpt-4o"
+    llm.model = "gpt-4o"
     messages = [
         llm._system_msg("some system msg"),
         llm._system_msg("some system message 2"),
@@ -184,7 +184,7 @@ def test_count_tokens():
 
 def test_count_tokens_long():
     llm = LLM()
-    llm.config.model = "gpt-4-0613"
+    llm.model = "gpt-4-0613"
     test_msg_content = " ".join([str(i) for i in range(100000)])
     messages = [
         llm._system_msg("You are a helpful assistant"),
@@ -193,7 +193,7 @@ def test_count_tokens_long():
     cnt = llm.count_tokens(messages)  # 299023, ~300k
     assert 290000 <= cnt <= 300000
 
-    llm.config.model = "test_llm"  # a non-openai model, will use heuristics base count_tokens
+    llm.model = "test_llm"  # a non-openai model, will use heuristics base count_tokens
     cnt = llm.count_tokens(messages)  # 294474, ~300k, ~2% difference
     assert 290000 <= cnt <= 300000
 
@@ -202,7 +202,7 @@ def test_count_tokens_long():
 @pytest.mark.asyncio
 async def test_aask_long():
     llm = LLM()
-    llm.config.model = "deepseek-ai/DeepSeek-Coder-V2-Instruct"  # deepseek-coder on siliconflow, limit 32k
+    llm.model = "deepseek-ai/DeepSeek-Coder-V2-Instruct"  # deepseek-coder on siliconflow, limit 32k
     llm.config.compress_type = CompressType.POST_CUT_BY_TOKEN
     test_msg_content = " ".join([str(i) for i in range(100000)])  # corresponds to ~300k tokens
     messages = [
@@ -216,7 +216,7 @@ async def test_aask_long():
 @pytest.mark.asyncio
 async def test_aask_long_no_compress():
     llm = LLM()
-    llm.config.model = "deepseek-ai/DeepSeek-Coder-V2-Instruct"  # deepseek-coder on siliconflow, limit 32k
+    llm.model = "deepseek-ai/DeepSeek-Coder-V2-Instruct"  # deepseek-coder on siliconflow, limit 32k
     # Not specifying llm.config.compress_type will use default "", no compress
     test_msg_content = " ".join([str(i) for i in range(100000)])  # corresponds to ~300k tokens
     messages = [