(self, body: bytes, stream: bool)
| 311 | # ─── /v1beta/models (Google Gemini CLI) ────────────────────────────────── |
| 312 | |
| 313 | def _handle_google_generate(self, body: bytes, stream: bool): |
| 314 | req = self._parse_body(body) |
| 315 | if req is None: |
| 316 | self.send_json({"error": {"message": "invalid JSON"}}, 400) |
| 317 | return |
| 318 | m = re.match(r'/v1beta/models/([^:?]+)', self.path) |
| 319 | model_name = m.group(1) if m else CONFIG["default_model"] |
| 320 | model_name, model_id, think_mode, err, extra_fields = resolve_model(model_name) |
| 321 | if err: |
| 322 | self.send_json({"error": {"message": err}}, 400) |
| 323 | return |
| 324 | |
| 325 | tool_config = req.get("toolConfig", {}) |
| 326 | fc_mode = tool_config.get("functionCallingConfig", {}).get("mode", "AUTO") |
| 327 | has_tools = bool(req.get("tools")) and fc_mode != "NONE" |
| 328 | prompt, images = google_contents_to_prompt(req) |
| 329 | if not prompt.strip(): |
| 330 | self.send_json({"error": {"message": "empty content"}}, 400) |
| 331 | return |
| 332 | |
| 333 | file_refs = _upload_images(images) |
| 334 | log(f"Google API: model={model_name} stream={stream} tools={has_tools} prompt_len={len(prompt)}") |
| 335 | |
| 336 | if stream and not has_tools: |
| 337 | try: |
| 338 | self._start_sse() |
| 339 | full_text = "" |
| 340 | for delta in generate_stream(prompt, model_id, think_mode, file_refs, extra_fields): |
| 341 | if not delta: |
| 342 | continue |
| 343 | full_text += delta |
| 344 | chunk_obj = { |
| 345 | "candidates": [{"content": {"parts": [{"text": delta}], "role": "model"}, "index": 0}], |
| 346 | "modelVersion": model_name, |
| 347 | } |
| 348 | self.wfile.write(f"data: {json.dumps(chunk_obj, ensure_ascii=False)}\n\n".encode()) |
| 349 | self.wfile.flush() |
| 350 | final_chunk = { |
| 351 | "candidates": [{"finishReason": "STOP", "index": 0}], |
| 352 | "usageMetadata": { |
| 353 | "promptTokenCount": len(prompt) // 4, |
| 354 | "candidatesTokenCount": len(full_text) // 4, |
| 355 | "totalTokenCount": (len(prompt) + len(full_text)) // 4, |
| 356 | }, |
| 357 | "modelVersion": model_name, |
| 358 | } |
| 359 | self.wfile.write(f"data: {json.dumps(final_chunk, ensure_ascii=False)}\n\n".encode()) |
| 360 | self.wfile.flush() |
| 361 | except (BrokenPipeError, ConnectionResetError): |
| 362 | pass |
| 363 | return |
| 364 | |
| 365 | try: |
| 366 | text = generate(prompt, model_id, think_mode, file_refs, extra_fields) |
| 367 | except Exception as e: |
| 368 | self.send_json({"error": {"message": f"upstream error: {e}"}}, 502) |
| 369 | return |
| 370 |
no test coverage detected