Spaces:
Running
Running
File size: 13,710 Bytes
bf53440 | 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 | from __future__ import annotations
import unittest
from unittest import mock
import json
import base64
from services.config import config
from services.protocol import openai_v1_chat_complete, openai_v1_response
from services.protocol.chat_completion_cache import chat_completion_cache
from services.protocol.conversation import iter_conversation_payloads, sanitize_output_text
from utils.helper import extract_image_from_message_content
PNG_1X1 = base64.b64decode(
"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAIAAACQd1PeAAAADUlEQVR4nGP8z8BQDwAFgwJ/luzl4wAAAABJRU5ErkJggg=="
)
PNG_1X1_DATA_URL = "data:image/png;base64," + base64.b64encode(PNG_1X1).decode("ascii")
class ChatCompletionCacheTests(unittest.TestCase):
def setUp(self) -> None:
self.old_cache_settings = config.data.get("chat_completion_cache")
config.data["chat_completion_cache"] = {
"enabled": True,
"ttl_seconds": 60,
"max_entries": 32,
"dedupe_inflight": True,
"stream_cache": True,
"normalize_messages": True,
"drop_adjacent_duplicates": True,
"drop_assistant_history": False,
}
chat_completion_cache.clear()
def tearDown(self) -> None:
if self.old_cache_settings is None:
config.data.pop("chat_completion_cache", None)
else:
config.data["chat_completion_cache"] = self.old_cache_settings
chat_completion_cache.clear()
def test_repeated_non_stream_text_completion_uses_cache(self) -> None:
calls = 0
def fake_collect_text(_backend, _request):
nonlocal calls
calls += 1
return f"cached answer {calls}"
body = {
"model": "auto",
"messages": [{"role": "user", "content": "cache this exact prompt"}],
}
with (
mock.patch("services.protocol.openai_v1_chat_complete.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_chat_complete.collect_text", side_effect=fake_collect_text),
):
first = openai_v1_chat_complete.handle(body)
second = openai_v1_chat_complete.handle(body)
self.assertEqual(calls, 1)
self.assertEqual(
first["choices"][0]["message"]["content"],
second["choices"][0]["message"]["content"],
)
def test_repeated_stream_text_completion_replays_cached_chunks(self) -> None:
calls = 0
def fake_stream_text_deltas(_backend, _request):
nonlocal calls
calls += 1
yield "streamed"
yield " answer"
body = {
"model": "auto",
"stream": True,
"messages": [{"role": "user", "content": "stream cache this exact prompt"}],
}
with (
mock.patch("services.protocol.openai_v1_chat_complete.text_backend", return_value=object()),
mock.patch(
"services.protocol.openai_v1_chat_complete.stream_text_deltas",
side_effect=fake_stream_text_deltas,
),
):
first = list(openai_v1_chat_complete.handle(body))
second = list(openai_v1_chat_complete.handle(body))
self.assertEqual(calls, 1)
self.assertEqual(first, second)
content = "".join(str(chunk["choices"][0]["delta"].get("content") or "") for chunk in second)
self.assertEqual(content, "streamed answer")
def test_adjacent_duplicate_messages_are_removed_before_upstream_call(self) -> None:
captured_messages = []
def fake_collect_text(_backend, request):
captured_messages.extend(request.messages or [])
return "ok"
body = {
"model": "auto",
"messages": [
{"role": "user", "content": "repeat me"},
{"role": "user", "content": "repeat me"},
{"role": "assistant", "content": "old answer"},
{"role": "user", "content": "next prompt"},
],
}
with (
mock.patch("services.protocol.openai_v1_chat_complete.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_chat_complete.collect_text", side_effect=fake_collect_text),
):
openai_v1_chat_complete.handle(body)
self.assertEqual(
captured_messages,
[
{"role": "user", "content": "repeat me"},
{"role": "assistant", "content": "old answer"},
{"role": "user", "content": "next prompt"},
],
)
def test_chat_completion_usage_includes_cached_tokens(self) -> None:
with (
mock.patch("services.protocol.openai_v1_chat_complete.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_chat_complete.collect_text", return_value="ok"),
):
response = openai_v1_chat_complete.handle({
"model": "auto",
"messages": [{"role": "user", "content": "usage shape"}],
})
details = response["usage"]["prompt_tokens_details"]
self.assertEqual(details["cached_tokens"], 0)
output_details = response["usage"]["completion_tokens_details"]
self.assertEqual(output_details["reasoning_tokens"], 0)
def test_responses_completed_usage_includes_cached_tokens(self) -> None:
with (
mock.patch("services.protocol.openai_v1_response.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_response.stream_text_deltas", return_value=iter(["ok"])),
):
response = openai_v1_response.handle({
"model": "auto",
"input": "usage shape",
})
details = response["usage"]["input_tokens_details"]
self.assertEqual(details["cached_tokens"], 0)
output_details = response["usage"]["output_tokens_details"]
self.assertEqual(output_details["reasoning_tokens"], 0)
def test_repeated_responses_text_request_uses_cache(self) -> None:
calls = 0
def fake_stream_text_deltas(_backend, _request):
nonlocal calls
calls += 1
yield f"response cache {calls}"
body = {
"model": "auto",
"input": "cache this responses prompt",
"stream": True,
}
with (
mock.patch("services.protocol.openai_v1_response.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_response.stream_text_deltas", side_effect=fake_stream_text_deltas),
):
first = list(openai_v1_response.handle(body))
second = list(openai_v1_response.handle(body))
self.assertEqual(calls, 1)
self.assertEqual(first, second)
def test_output_sanitizer_removes_chatgpt_annotation_markup(self) -> None:
text = (
"Repo: \ue200url\ue202basketikun/chatgpt2api"
"\ue202https://github.com/basketikun/chatgpt2api\ue201 "
"details \ue200cite\ue202turn0search0\ue201."
)
self.assertEqual(
sanitize_output_text(text),
"Repo: basketikun/chatgpt2api (https://github.com/basketikun/chatgpt2api) details.",
)
def test_output_sanitizer_preserves_annotated_entity_text(self) -> None:
text = (
"The character is from \ue200entity\ue202Invincible\ue201, "
"which is based on the comic series \ue200entity\ue202Invincible\ue201."
)
self.assertEqual(
sanitize_output_text(text),
"The character is from Invincible, which is based on the comic series Invincible.",
)
def test_output_sanitizer_preserves_readable_cite_label(self) -> None:
text = "The character is \ue200cite\ue202Invincible\ue202turn0search0\ue201."
self.assertEqual(sanitize_output_text(text), "The character is Invincible.")
def test_stream_sanitizer_does_not_emit_partial_annotation_or_repeat_prefix(self) -> None:
events = [
{"p": "/message/content/parts/0", "o": "append", "v": "Repo: \ue200url\ue202chat"},
{"p": "/message/content/parts/0", "o": "append", "v": "gpt2api\ue202turn0search0\ue201 done \ue200cite\ue202turn0\ue201."},
"[DONE]",
]
payloads = [json.dumps(event, ensure_ascii=False) if isinstance(event, dict) else event for event in events]
deltas = [
str(event.get("delta") or "")
for event in iter_conversation_payloads(iter(payloads))
if event.get("type") == "conversation.delta"
]
self.assertEqual("".join(deltas), "Repo: chatgpt2api done.")
self.assertFalse(any("\ue200" in delta or "\ue202" in delta or "\ue201" in delta for delta in deltas))
def test_responses_tools_add_honest_no_tool_guard(self) -> None:
model, messages = openai_v1_response.text_response_parts({
"model": "auto",
"input": "run echo hi",
"tools": [{"type": "function", "name": "shell"}],
})
self.assertEqual(model, "auto")
self.assertEqual(messages[0]["role"], "system")
self.assertIn("cannot execute local tools", str(messages[0]["content"]))
def test_chat_completions_accepts_remote_image_url(self) -> None:
class FakeImageResponse:
status_code = 200
headers = {"content-type": "image/png", "content-length": str(len(PNG_1X1))}
content = PNG_1X1
with mock.patch("utils.helper.requests.get", return_value=FakeImageResponse()) as request_get:
model, messages = openai_v1_chat_complete.text_chat_parts({
"model": "auto",
"messages": [{
"role": "user",
"content": [
{"type": "text", "text": "Describe this"},
{"type": "image_url", "image_url": {"url": "https://example.test/image.png"}},
],
}],
})
request_get.assert_called_once()
self.assertEqual(model, "auto")
content = messages[0]["content"]
self.assertEqual(content[0], {"type": "text", "text": "Describe this"})
self.assertEqual(content[1]["type"], "image")
self.assertEqual(content[1]["data"], PNG_1X1)
self.assertEqual(content[1]["mime"], "image/png")
def test_responses_text_request_preserves_input_image(self) -> None:
captured = {}
def fake_stream_text_deltas(_backend, request):
captured["messages"] = request.messages
yield "red"
body = {
"model": "auto",
"input": [
{"type": "input_text", "text": "What color is this image?"},
{"type": "input_image", "image_url": PNG_1X1_DATA_URL},
],
}
with (
mock.patch("services.protocol.openai_v1_response.text_backend", return_value=object()),
mock.patch("services.protocol.openai_v1_response.stream_text_deltas", side_effect=fake_stream_text_deltas),
):
response = openai_v1_response.handle(body)
self.assertEqual(response["output"][0]["content"][0]["text"], "red")
content = captured["messages"][0]["content"]
self.assertEqual(content[0], {"type": "text", "text": "What color is this image?"})
self.assertEqual(content[1]["type"], "image")
self.assertEqual(content[1]["mime"], "image/png")
self.assertEqual(content[1]["data"], PNG_1X1)
self.assertGreater(response["usage"]["input_tokens_details"]["image_tokens"], 0)
def test_responses_text_request_accepts_remote_input_image_url(self) -> None:
class FakeImageResponse:
status_code = 200
headers = {"content-type": "image/png", "content-length": str(len(PNG_1X1))}
content = PNG_1X1
with mock.patch("utils.helper.requests.get", return_value=FakeImageResponse()) as request_get:
_model, messages = openai_v1_response.text_response_parts({
"model": "auto",
"input": [{
"type": "message",
"role": "user",
"content": [
{"type": "input_text", "text": "Describe this"},
{"type": "input_image", "image_url": {"url": "https://example.test/image.png"}},
],
}],
})
request_get.assert_called_once()
content = messages[0]["content"]
self.assertEqual(content[0], {"type": "text", "text": "Describe this"})
self.assertEqual(content[1]["type"], "image")
self.assertEqual(content[1]["data"], PNG_1X1)
self.assertEqual(content[1]["mime"], "image/png")
def test_image_extractor_supports_extra_image_object_shapes(self) -> None:
encoded = base64.b64encode(PNG_1X1).decode("ascii")
images = extract_image_from_message_content([
{"type": "image", "data": PNG_1X1, "mime": "image/png"},
{"type": "input_image", "base64": encoded, "mime_type": "image/png"},
{"type": "input_image", "source": {"type": "base64", "data": encoded, "media_type": "image/png"}},
])
self.assertEqual(len(images), 3)
self.assertEqual([image[1] for image in images], ["image/png", "image/png", "image/png"])
self.assertTrue(all(image[0] == PNG_1X1 for image in images))
if __name__ == "__main__":
unittest.main()
|