feat(US7+US8): finetune management and health check test

- app/models/finetune_models.py: FinetuneStartRequest, FinetuneStartResponse, FinetuneStatusResponse
- app/services/finetune_service.py: submit_finetune + get_finetune_status via run_in_executor; status map running→RUNNING, succeeded→SUCCESS, failed→FAILED, unknown→RUNNING; LLMCallError on SDK failure
- app/routers/finetune.py: POST /finetune/start + GET /finetune/status/{job_id} with get_llm_client dependency
- tests/test_finetune_service.py: 12 unit tests (TDD, written before implementation)
- tests/test_finetune_router.py: 6 integration tests
- tests/test_health.py: GET /health → 200 {"status":"ok"}

Full suite: 72/72 passing (was 53)

app/models/finetune_models.py

@@ -0,0 +1,18 @@
+from pydantic import BaseModel
+
+
+class FinetuneStartRequest(BaseModel):
+    jsonl_url: str
+    base_model: str
+    hyperparams: dict | None = None
+
+
+class FinetuneStartResponse(BaseModel):
+    job_id: str
+
+
+class FinetuneStatusResponse(BaseModel):
+    job_id: str
+    status: str
+    progress: int | None = None
+    error_message: str | None = None

app/routers/finetune.py

@@ -1,3 +1,28 @@
-from fastapi import APIRouter
+from fastapi import APIRouter, Depends
+
+from app.clients.llm.base import LLMClient
+from app.core.dependencies import get_llm_client
+from app.models.finetune_models import (
+    FinetuneStartRequest,
+    FinetuneStartResponse,
+    FinetuneStatusResponse,
+)
+from app.services import finetune_service
 
 router = APIRouter(tags=["Finetune"])
+
+
+@router.post("/finetune/start", response_model=FinetuneStartResponse)
+async def start_finetune(
+    req: FinetuneStartRequest,
+    llm: LLMClient = Depends(get_llm_client),
+) -> FinetuneStartResponse:
+    return await finetune_service.submit_finetune(req, llm)
+
+
+@router.get("/finetune/status/{job_id}", response_model=FinetuneStatusResponse)
+async def get_status(
+    job_id: str,
+    llm: LLMClient = Depends(get_llm_client),
+) -> FinetuneStatusResponse:
+    return await finetune_service.get_finetune_status(job_id, llm)

app/services/finetune_service.py

@@ -0,0 +1,61 @@
+import asyncio
+
+from app.core.exceptions import LLMCallError
+from app.core.logging import get_logger
+from app.models.finetune_models import (
+    FinetuneStartRequest,
+    FinetuneStartResponse,
+    FinetuneStatusResponse,
+)
+
+logger = get_logger(__name__)
+
+_STATUS_MAP = {
+    "running": "RUNNING",
+    "succeeded": "SUCCESS",
+    "failed": "FAILED",
+}
+
+
+async def submit_finetune(req: FinetuneStartRequest, llm) -> FinetuneStartResponse:
+    """Submit a fine-tune job to ZhipuAI and return the job ID."""
+    loop = asyncio.get_event_loop()
+    try:
+        response = await loop.run_in_executor(
+            None,
+            lambda: llm._client.fine_tuning.jobs.create(
+                training_file=req.jsonl_url,
+                model=req.base_model,
+                hyperparameters=req.hyperparams or {},
+            ),
+        )
+        job_id = response.id
+        logger.info("finetune_submit", extra={"job_id": job_id, "model": req.base_model})
+        return FinetuneStartResponse(job_id=job_id)
+    except Exception as exc:
+        logger.error("finetune_submit_error", extra={"error": str(exc)})
+        raise LLMCallError(f"微调任务提交失败: {exc}") from exc
+
+
+async def get_finetune_status(job_id: str, llm) -> FinetuneStatusResponse:
+    """Retrieve fine-tune job status from ZhipuAI."""
+    loop = asyncio.get_event_loop()
+    try:
+        response = await loop.run_in_executor(
+            None,
+            lambda: llm._client.fine_tuning.jobs.retrieve(job_id),
+        )
+        status_raw = response.status
+        status = _STATUS_MAP.get(status_raw, "RUNNING")  # conservative fallback
+        progress = getattr(response, "progress", None)
+        error_message = getattr(response, "error_message", None)
+        logger.info("finetune_status", extra={"job_id": job_id, "status": status})
+        return FinetuneStatusResponse(
+            job_id=job_id,
+            status=status,
+            progress=progress,
+            error_message=error_message,
+        )
+    except Exception as exc:
+        logger.error("finetune_status_error", extra={"job_id": job_id, "error": str(exc)})
+        raise LLMCallError(f"微调状态查询失败: {exc}") from exc