Spaces:
Running
Running
File size: 12,599 Bytes
c47ec10 | 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 | from __future__ import annotations
import time
import uuid
from typing import Any, Iterable, Iterator
from fastapi import HTTPException
from services.protocol.chat_completion_cache import cache_key, chat_completion_cache, normalize_text_messages
from services.protocol.conversation import (
ConversationRequest,
ImageOutput,
collect_image_outputs,
collect_text,
count_message_image_tokens,
count_message_text_tokens,
count_text_tokens,
encode_images,
normalize_messages,
stream_image_outputs_with_pool,
stream_text_deltas,
text_backend,
)
from services.protocol.web_search_tool import (
WEB_SEARCH_TOOL_TYPES,
has_unsupported_tools,
is_web_search_chat_request,
run_web_search,
search_query_from_messages,
text_with_url_citations,
)
from utils.helper import build_chat_image_markdown_content, extract_chat_image, extract_chat_prompt, is_image_chat_request, parse_image_count
from utils.image_tokens import (
chat_usage_from_image_usage,
count_image_inputs_tokens,
count_image_output_items_tokens,
image_usage,
)
TOOL_UNAVAILABLE_SYSTEM_MESSAGE = (
"This compatibility backend cannot execute local tools, shell commands, non-search tools, "
"or file operations. Do not claim to have run tools or inspected external resources. "
"If a user asks you to use a tool, say that tool execution is unavailable through this backend."
)
def normalize_thinking_effort(value: object) -> str:
normalized = str(value or "").strip().lower()
if normalized in {"", "none"}:
return ""
if normalized in {"low", "medium", "high"}:
return normalized
if normalized in {"xhigh", "extended"}:
return "extended"
return ""
def thinking_effort_from_body(body: dict[str, Any]) -> str:
if "thinking_effort" in body:
return normalize_thinking_effort(body.get("thinking_effort"))
if "reasoning_effort" in body:
return normalize_thinking_effort(body.get("reasoning_effort"))
reasoning = body.get("reasoning")
if isinstance(reasoning, dict):
return normalize_thinking_effort(reasoning.get("effort"))
return ""
def completion_chunk(model: str, delta: dict[str, Any], finish_reason: str | None = None, completion_id: str = "", created: int | None = None) -> dict[str, Any]:
return {
"id": completion_id or f"chatcmpl-{uuid.uuid4().hex}",
"object": "chat.completion.chunk",
"created": created or int(time.time()),
"model": model,
"choices": [{"index": 0, "delta": delta, "finish_reason": finish_reason}],
}
def completion_response(
model: str,
content: str,
created: int | None = None,
messages: list[dict[str, Any]] | None = None,
annotations: list[dict[str, Any]] | None = None,
) -> dict[str, Any]:
prompt_text_tokens = count_message_text_tokens(messages, model) if messages else 0
prompt_image_tokens = count_message_image_tokens(messages, model) if messages else 0
prompt_tokens = prompt_text_tokens + prompt_image_tokens
completion_tokens = count_text_tokens(content, model) if messages else 0
message = {"role": "assistant", "content": content}
if annotations:
message["annotations"] = annotations
return {
"id": f"chatcmpl-{uuid.uuid4().hex}",
"object": "chat.completion",
"created": created or int(time.time()),
"model": model,
"choices": [{
"index": 0,
"message": message,
"finish_reason": "stop",
}],
"usage": {
"prompt_tokens": prompt_tokens,
"completion_tokens": completion_tokens,
"total_tokens": prompt_tokens + completion_tokens,
"prompt_tokens_details": {
"text_tokens": prompt_text_tokens,
"image_tokens": prompt_image_tokens,
"cached_tokens": 0,
},
"completion_tokens_details": {
"text_tokens": completion_tokens,
"image_tokens": 0,
"reasoning_tokens": 0,
},
},
}
def stream_text_chat_completion(
backend,
messages: list[dict[str, Any]],
model: str,
thinking_effort: str = "",
) -> Iterator[dict[str, Any]]:
completion_id = f"chatcmpl-{uuid.uuid4().hex}"
created = int(time.time())
sent_role = False
request = ConversationRequest(model=model, messages=messages, thinking_effort=thinking_effort)
for delta_text in stream_text_deltas(backend, request):
if not sent_role:
sent_role = True
yield completion_chunk(model, {"role": "assistant", "content": delta_text}, None, completion_id, created)
else:
yield completion_chunk(model, {"content": delta_text}, None, completion_id, created)
if not sent_role:
yield completion_chunk(model, {"role": "assistant", "content": ""}, None, completion_id, created)
yield completion_chunk(model, {}, "stop", completion_id, created)
def collect_chat_content(chunks: Iterable[dict[str, Any]]) -> str:
parts: list[str] = []
for chunk in chunks:
choices = chunk.get("choices")
first = choices[0] if isinstance(choices, list) and choices and isinstance(choices[0], dict) else {}
delta = first.get("delta") if isinstance(first.get("delta"), dict) else {}
content = str(delta.get("content") or "")
if content:
parts.append(content)
return "".join(parts)
def chat_messages_from_body(body: dict[str, Any]) -> list[dict[str, Any]]:
messages = body.get("messages")
if isinstance(messages, list) and messages:
return [message for message in messages if isinstance(message, dict)]
prompt = str(body.get("prompt") or "").strip()
if prompt:
return [{"role": "user", "content": prompt}]
raise HTTPException(status_code=400, detail={"error": "messages or prompt is required"})
def chat_image_args(body: dict[str, Any]) -> tuple[str, str, int, list[tuple[bytes, str, str]]]:
model = str(body.get("model") or "gpt-image-2").strip() or "gpt-image-2"
prompt = extract_chat_prompt(body)
if not prompt:
raise HTTPException(status_code=400, detail={"error": "prompt is required"})
images = [
(data, f"image_{idx}.png", mime)
for idx, (data, mime) in enumerate(extract_chat_image(body), start=1)
]
return model, prompt, parse_image_count(body.get("n")), images
def text_chat_parts(body: dict[str, Any]) -> tuple[str, list[dict[str, Any]]]:
model = str(body.get("model") or "auto").strip() or "auto"
messages = normalize_text_messages(normalize_messages(chat_messages_from_body(body)))
if has_unsupported_tools(body, WEB_SEARCH_TOOL_TYPES):
messages.insert(0, {"role": "system", "content": TOOL_UNAVAILABLE_SYSTEM_MESSAGE})
return model, messages
def chat_completion_annotations(annotations: list[dict[str, Any]]) -> list[dict[str, Any]]:
output = []
for item in annotations:
if item.get("type") != "url_citation":
continue
output.append({
"type": "url_citation",
"url_citation": {
"start_index": item.get("start_index", 0),
"end_index": item.get("end_index", 0),
"url": item.get("url", ""),
"title": item.get("title", ""),
},
})
return output
def web_search_chat_response(messages: list[dict[str, Any]], model: str) -> dict[str, Any]:
query = search_query_from_messages(messages)
if not query:
raise HTTPException(status_code=400, detail={"error": "messages or prompt is required for web search"})
text, annotations = text_with_url_citations(run_web_search(query))
return completion_response(
model,
text,
messages=messages,
annotations=chat_completion_annotations(annotations),
)
def stream_web_search_chat_completion(messages: list[dict[str, Any]], model: str) -> Iterator[dict[str, Any]]:
query = search_query_from_messages(messages)
if not query:
raise HTTPException(status_code=400, detail={"error": "messages or prompt is required for web search"})
text, _annotations = text_with_url_citations(run_web_search(query))
completion_id = f"chatcmpl-{uuid.uuid4().hex}"
created = int(time.time())
yield completion_chunk(model, {"role": "assistant", "content": text}, None, completion_id, created)
yield completion_chunk(model, {}, "stop", completion_id, created)
def image_result_content(result: dict[str, Any]) -> str:
data = result.get("data")
if isinstance(data, list) and data:
return build_chat_image_markdown_content(result)
return str(result.get("message") or "Image generation completed.")
def image_chat_response(body: dict[str, Any]) -> dict[str, Any]:
model, prompt, n, images = chat_image_args(body)
result = collect_image_outputs(stream_image_outputs_with_pool(ConversationRequest(
prompt=prompt,
model=model,
n=n,
response_format="b64_json",
images=encode_images(images) or None,
)))
response = completion_response(model, image_result_content(result), int(result.get("created") or 0) or None)
usage = image_usage(
input_text_tokens=count_text_tokens(prompt, model),
input_image_tokens=count_image_inputs_tokens(images, model),
output_tokens=count_image_output_items_tokens(result.get("data")),
)
response["usage"] = chat_usage_from_image_usage(usage)
return response
def image_chat_events(body: dict[str, Any]) -> Iterator[dict[str, Any]]:
model, prompt, n, images = chat_image_args(body)
image_outputs = stream_image_outputs_with_pool(ConversationRequest(
prompt=prompt,
model=model,
n=n,
response_format="b64_json",
images=encode_images(images) or None,
))
yield from stream_image_chat_completion(image_outputs, model)
def stream_image_chat_completion(image_outputs: Iterable[ImageOutput], model: str) -> Iterator[dict[str, Any]]:
completion_id = f"chatcmpl-{uuid.uuid4().hex}"
created = int(time.time())
sent_role = False
sent_text = ""
for output in image_outputs:
content = ""
if output.kind == "progress":
content = output.text
sent_text += content
elif output.kind == "result":
content = build_chat_image_markdown_content({"data": output.data})
elif output.kind == "message":
content = output.text[len(sent_text):] if output.text.startswith(sent_text) else output.text
if not content:
continue
if not sent_role:
sent_role = True
yield completion_chunk(model, {"role": "assistant", "content": content}, None, completion_id, created)
else:
yield completion_chunk(model, {"content": content}, None, completion_id, created)
if not sent_role:
yield completion_chunk(model, {"role": "assistant", "content": ""}, None, completion_id, created)
yield completion_chunk(model, {}, "stop", completion_id, created)
def handle(body: dict[str, Any]) -> dict[str, Any] | Iterator[dict[str, Any]]:
if body.get("stream"):
if is_image_chat_request(body):
return image_chat_events(body)
model, messages = text_chat_parts(body)
if is_web_search_chat_request(body) and not has_unsupported_tools(body, WEB_SEARCH_TOOL_TYPES):
return stream_web_search_chat_completion(messages, model)
thinking_effort = thinking_effort_from_body(body)
key = cache_key(body, messages, stream=True)
return chat_completion_cache.get_or_compute_stream(
key,
lambda: stream_text_chat_completion(text_backend(), messages, model, thinking_effort),
)
if is_image_chat_request(body):
return image_chat_response(body)
model, messages = text_chat_parts(body)
if is_web_search_chat_request(body) and not has_unsupported_tools(body, WEB_SEARCH_TOOL_TYPES):
return web_search_chat_response(messages, model)
thinking_effort = thinking_effort_from_body(body)
key = cache_key(body, messages, stream=False)
return chat_completion_cache.get_or_compute_response(
key,
lambda: completion_response(
model,
collect_text(text_backend(), ConversationRequest(model=model, messages=messages, thinking_effort=thinking_effort)),
messages=messages,
),
)
|