|
| 1 | +import unittest |
| 2 | +from unittest.mock import MagicMock, patch |
| 3 | + |
| 4 | +from fastdeploy.entrypoints.openai.serving_completion import ( |
| 5 | + CompletionRequest, |
| 6 | + OpenAIServingCompletion, |
| 7 | +) |
| 8 | + |
| 9 | + |
| 10 | +class YourClass: |
| 11 | + async def _1(self, a, b, c): |
| 12 | + if b["outputs"].get("send_idx", -1) == 0 and a.echo: |
| 13 | + if isinstance(a.prompt, list): |
| 14 | + text = a.prompt[c] |
| 15 | + else: |
| 16 | + text = a.prompt |
| 17 | + b["outputs"]["text"] = text + (b["outputs"]["text"] or "") |
| 18 | + |
| 19 | + |
| 20 | +class TestCompletionEcho(unittest.IsolatedAsyncioTestCase): |
| 21 | + def setUp(self): |
| 22 | + self.mock_engine = MagicMock() |
| 23 | + self.completion_handler = None |
| 24 | + |
| 25 | + def test_single_prompt_non_streaming(self): |
| 26 | + """测试单prompt非流式响应""" |
| 27 | + self.completion_handler = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 28 | + |
| 29 | + request = CompletionRequest(prompt="test prompt", max_tokens=10, echo=True, logprobs=1) |
| 30 | + |
| 31 | + mock_output = { |
| 32 | + "outputs": { |
| 33 | + "text": " generated text", |
| 34 | + "token_ids": [1, 2, 3], |
| 35 | + "top_logprobs": {"token1": -0.1, "token2": -0.2}, |
| 36 | + "finished": True, |
| 37 | + }, |
| 38 | + "output_token_ids": 3, |
| 39 | + } |
| 40 | + self.mock_engine.generate.return_value = [mock_output] |
| 41 | + |
| 42 | + response = self.completion_handler.request_output_to_completion_response( |
| 43 | + final_res_batch=[mock_output], |
| 44 | + request=request, |
| 45 | + request_id="test_id", |
| 46 | + created_time=12345, |
| 47 | + model_name="test_model", |
| 48 | + prompt_batched_token_ids=[[1, 2]], |
| 49 | + completion_batched_token_ids=[[3, 4, 5]], |
| 50 | + text_after_process_list=["test prompt"], |
| 51 | + ) |
| 52 | + |
| 53 | + self.assertEqual(response.choices[0].text, "test prompt generated text") |
| 54 | + |
| 55 | + async def test_echo_back_prompt_and_streaming(self): |
| 56 | + """测试_echo_back_prompt方法和流式响应的prompt拼接逻辑""" |
| 57 | + self.completion_handler = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 58 | + |
| 59 | + request = CompletionRequest(prompt="test prompt", max_tokens=10, stream=True, echo=True) |
| 60 | + |
| 61 | + mock_response = {"outputs": {"text": "test output", "token_ids": [1, 2, 3], "finished": True}} |
| 62 | + |
| 63 | + with patch.object(self.completion_handler, "_echo_back_prompt") as mock_echo: |
| 64 | + |
| 65 | + def mock_echo_side_effect(req, res, idx): |
| 66 | + res["outputs"]["text"] = req.prompt + res["outputs"]["text"] |
| 67 | + |
| 68 | + mock_echo.side_effect = mock_echo_side_effect |
| 69 | + |
| 70 | + await self.completion_handler._echo_back_prompt(request, mock_response, 0) |
| 71 | + |
| 72 | + mock_echo.assert_called_once_with(request, mock_response, 0) |
| 73 | + |
| 74 | + self.assertEqual(mock_response["outputs"]["text"], "test prompttest output") |
| 75 | + self.assertEqual(request.prompt, "test prompt") |
| 76 | + |
| 77 | + def test_multi_prompt_non_streaming(self): |
| 78 | + """测试多prompt非流式响应""" |
| 79 | + self.completion_handler = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 80 | + |
| 81 | + request = CompletionRequest(prompt=["prompt1", "prompt2"], max_tokens=10, echo=True) |
| 82 | + |
| 83 | + mock_outputs = [ |
| 84 | + { |
| 85 | + "outputs": {"text": " response1", "token_ids": [1, 2], "top_logprobs": None, "finished": True}, |
| 86 | + "output_token_ids": 2, |
| 87 | + }, |
| 88 | + { |
| 89 | + "outputs": {"text": " response2", "token_ids": [3, 4], "top_logprobs": None, "finished": True}, |
| 90 | + "output_token_ids": 2, |
| 91 | + }, |
| 92 | + ] |
| 93 | + self.mock_engine.generate.return_value = mock_outputs |
| 94 | + |
| 95 | + response = self.completion_handler.request_output_to_completion_response( |
| 96 | + final_res_batch=mock_outputs, |
| 97 | + request=request, |
| 98 | + request_id="test_id", |
| 99 | + created_time=12345, |
| 100 | + model_name="test_model", |
| 101 | + prompt_batched_token_ids=[[1], [2]], |
| 102 | + completion_batched_token_ids=[[1, 2], [3, 4]], |
| 103 | + text_after_process_list=["prompt1", "prompt2"], |
| 104 | + ) |
| 105 | + |
| 106 | + self.assertEqual(len(response.choices), 2) |
| 107 | + self.assertEqual(response.choices[0].text, "prompt1 response1") |
| 108 | + self.assertEqual(response.choices[1].text, "prompt2 response2") |
| 109 | + |
| 110 | + async def test_multi_prompt_streaming(self): |
| 111 | + self.completion_handler = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 112 | + |
| 113 | + request = CompletionRequest(prompt=["prompt1", "prompt2"], max_tokens=10, stream=True, echo=True) |
| 114 | + |
| 115 | + mock_responses = [ |
| 116 | + {"outputs": {"text": " response1", "token_ids": [1, 2], "finished": True}}, |
| 117 | + {"outputs": {"text": " response2", "token_ids": [3, 4], "finished": True}}, |
| 118 | + ] |
| 119 | + |
| 120 | + with patch.object(self.completion_handler, "_echo_back_prompt") as mock_echo: |
| 121 | + |
| 122 | + def mock_echo_side_effect(req, res, idx): |
| 123 | + res["outputs"]["text"] = req.prompt[idx] + res["outputs"]["text"] |
| 124 | + |
| 125 | + mock_echo.side_effect = mock_echo_side_effect |
| 126 | + |
| 127 | + await self.completion_handler._echo_back_prompt(request, mock_responses[0], 0) |
| 128 | + await self.completion_handler._echo_back_prompt(request, mock_responses[1], 1) |
| 129 | + |
| 130 | + self.assertEqual(mock_echo.call_count, 2) |
| 131 | + mock_echo.assert_any_call(request, mock_responses[0], 0) |
| 132 | + mock_echo.assert_any_call(request, mock_responses[1], 1) |
| 133 | + |
| 134 | + self.assertEqual(mock_responses[0]["outputs"]["text"], "prompt1 response1") |
| 135 | + self.assertEqual(mock_responses[1]["outputs"]["text"], "prompt2 response2") |
| 136 | + self.assertEqual(request.prompt, ["prompt1", "prompt2"]) |
| 137 | + |
| 138 | + async def test_echo_back_prompt_and_streaming1(self): |
| 139 | + request = CompletionRequest(echo=True, prompt=["Hello", "World"]) |
| 140 | + res = {"outputs": {"send_idx": 0, "text": "!"}} |
| 141 | + idx = 0 |
| 142 | + |
| 143 | + instance = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 144 | + await instance._echo_back_prompt(request, res, idx) |
| 145 | + self.assertEqual(res["outputs"]["text"], "Hello!") |
| 146 | + |
| 147 | + async def test_1_prompt_is_string_and_send_idx_is_0(self): |
| 148 | + request = CompletionRequest(echo=True, prompt="Hello") |
| 149 | + res = {"outputs": {"send_idx": 0, "text": "!"}} |
| 150 | + idx = 0 |
| 151 | + |
| 152 | + instance = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 153 | + await instance._echo_back_prompt(request, res, idx) |
| 154 | + self.assertEqual(res["outputs"]["text"], "Hello!") |
| 155 | + |
| 156 | + async def test_1_send_idx_is_not_0(self): |
| 157 | + request = CompletionRequest(echo=True, prompt="Hello") |
| 158 | + res = {"outputs": {"send_idx": 1, "text": "!"}} |
| 159 | + idx = 0 |
| 160 | + |
| 161 | + instance = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 162 | + await instance._echo_back_prompt(request, res, idx) |
| 163 | + self.assertEqual(res["outputs"]["text"], "!") |
| 164 | + |
| 165 | + async def test_1_echo_is_false(self): |
| 166 | + """测试echo为False时，_echo_back_prompt不拼接prompt""" |
| 167 | + request = CompletionRequest(echo=False, prompt="Hello") |
| 168 | + res = {"outputs": {"send_idx": 0, "text": "!"}} |
| 169 | + idx = 0 |
| 170 | + |
| 171 | + instance = OpenAIServingCompletion(self.mock_engine, pid=123, ips=None, max_waiting_time=30) |
| 172 | + await instance._echo_back_prompt(request, res, idx) |
| 173 | + self.assertEqual(res["outputs"]["text"], "!") |
| 174 | + |
| 175 | + |
| 176 | +if __name__ == "__main__": |
| 177 | + unittest.main() |
0 commit comments