Revert health check to share the advisor's LLM client

No HF Inference Providers vision-language model under 0.5B parameters
is available; the next-smallest options (Llama-4-Scout, Qwen3-VL-8B)
were rejected as too large. Back to sharing ADVISOR_MODEL_ID/PROVIDER,
even though the text-only model can't process the uploaded image.

TECHNICAL_DOCUMENT.md

@@ -230,11 +230,9 @@ yellow?", "Can I plant this next to my tomatoes?").
 **Task**: given a *new* photo of the selected plant, assess its health
 (leaves, stems, soil) and store the verdict on the plant record.
 
-- Model: `HEALTH_MODEL_ID` (default `meta-llama/Llama-4-Scout-17B-16E-Instruct`),
-  on the same `HEALTH_PROVIDER` (defaults to `ADVISOR_PROVIDER`, i.e. `nscale`)
-  — a vision-capable model, since the text-only advisor model (§4.4) cannot
-  process images. Called via `InferenceClient.chat_completion` with a
-  **multimodal** message: the uploaded `PIL.Image` is re-encoded as JPEG,
+- Model: the same `ADVISOR_MODEL_ID` / `ADVISOR_PROVIDER` client as the chat
+  advisor (§4.4), again via `InferenceClient.chat_completion` — but this time
+  with a **multimodal** message: the uploaded `PIL.Image` is re-encoded as JPEG,
   base64-encoded, and sent as an OpenAI-style content array
   (`{"type": "text", ...}` + `{"type": "image_url", "image_url": {"url":
   "data:image/jpeg;base64,..."}}`).
@@ -315,8 +313,7 @@ defaults baked in so the app runs locally without any secrets:
 |---|---|---|
 | `WEATHER_CITY` | `"Marseille"` | Initial forecast location before the user sets one |
 | `CLASSIFIER_MODEL_ID` | `"your-username/plant-genus-classifier"` | HF Hub repo of the fine-tuned SigLIP genus classifier |
-| `ADVISOR_MODEL_ID` / `ADVISOR_PROVIDER` | `Qwen/Qwen2.5-Coder-3B-Instruct` / `nscale` | Chat advisor model + HF Inference provider |
-| `HEALTH_MODEL_ID` / `HEALTH_PROVIDER` | `meta-llama/Llama-4-Scout-17B-16E-Instruct` / `nscale` | Vision-language health-diagnostic model + provider (defaults to `ADVISOR_PROVIDER`) |
+| `ADVISOR_MODEL_ID` / `ADVISOR_PROVIDER` | `Qwen/Qwen2.5-Coder-3B-Instruct` / `nscale` | Chat advisor + health-diagnostic model and HF Inference provider (shared) |
 | `HF_TOKEN` | — | Hugging Face token for Inference Providers (advisor + health check) |
 
 `app.launch(allowed_paths=[...])` whitelists `user_data/`, `static/` and

modules/advisor.py

@@ -9,13 +9,7 @@ from PIL import Image
 ADVISOR_MODEL_ID = os.getenv("ADVISOR_MODEL_ID", "Qwen/Qwen2.5-Coder-3B-Instruct")
 ADVISOR_PROVIDER = os.getenv("ADVISOR_PROVIDER", "nscale")
 
-# Qwen2.5-Coder is text-only, so the health check (which sends a photo) needs
-# a vision-capable model. Use the same provider, on a model that supports images.
-HEALTH_MODEL_ID = os.getenv("HEALTH_MODEL_ID", "meta-llama/Llama-4-Scout-17B-16E-Instruct")
-HEALTH_PROVIDER = os.getenv("HEALTH_PROVIDER", ADVISOR_PROVIDER)
-
 _client = None
-_health_client = None
 
 
 def _get_client() -> InferenceClient:
@@ -25,13 +19,6 @@ def _get_client() -> InferenceClient:
     return _client
 
 
-def _get_health_client() -> InferenceClient:
-    global _health_client
-    if _health_client is None:
-        _health_client = InferenceClient(model=HEALTH_MODEL_ID, provider=HEALTH_PROVIDER, token=os.getenv("HF_TOKEN"))
-    return _health_client
-
-
 def _watering_status(last_watered: str | None) -> str:
     if not last_watered:
         return "Never watered yet (no record)."
@@ -127,7 +114,7 @@ def diagnose_plant_health(
     )
 
     try:
-        completion = _get_health_client().chat_completion(
+        completion = _get_client().chat_completion(
             messages=[{
                 "role": "user",
                 "content": [