Spaces:
Sleeping
Sleeping
Commit History
-
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.a3ee5be -
Fix health check 400 error: use a vision-capable model
Qwen2.5-Coder-3B-Instruct (the advisor's model) is text-only and rejected the multimodal image_url payload with a 400 Bad Request. The health check now uses meta-llama/Llama-4-Scout-17B-16E-Instruct on the same nscale provider, which supports image inputs.5ac2dd2 -
Fix health check: reuse the advisor's LLM client instead of a separate VL model
The configured HEALTH_MODEL_ID (Qwen2.5-VL-7B-Instruct, provider auto) was not enabled for this account and failed with a model_not_supported error. The health diagnostic now shares the advisor's InferenceClient.d1c366d -
Fix link line on drag, add watering history, custom garden background, and photo-based health diagnostic
- Re-render the board after dragging a plant so neighbor-link lines follow it - Track per-plant watering_history and show it in the detail card - Let users upload/apply/reset a custom background image for the garden board - Add a vision-model health check (HF Inference) that stores a 'health' field per plantbeddf30 -
Allow hand-drawn links between neighboring plants on the garden board
Add a "Relier des plantes" link mode: click two sprites to connect (or disconnect) them. Linked pairs are stored bidirectionally as neighbors, drawn as dashed lines on the board, and surfaced to the gardening advisor so it can factor in companion-planting effects.3fa67e8 -
Ground advisor answers in the selected plant's watering history
Pass the plant's last_watered date to the advisor so it factors days-since-watering into its advice (overdue watering vs. possible overwatering), and sharpen the system prompt toward precise, gardening-advice-focused answers.14d5092 -
Switch advisor to Qwen2.5-Coder-3B-Instruct via nscale provider
Qwen2.5-3B-Instruct is only served by featherless-ai, which keeps returning model_not_supported. Qwen2.5-Coder-3B-Instruct (3B) is also live on the nscale provider, so pin both model and provider.bdd923c -
Fix plant photo display and add temp advisor error logging
- Serve uploaded plant photos via Gradio's /gradio_api/file= route so they actually render when clicking a plant in the garden. - Log the real HF Inference exception in the advisor so the cause of HF_TOKEN issues shows up in the Space logs. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>62644f1 -
Add AI gardening advisor (Qwen2.5-3B-Instruct via HF Inference)
Adds a per-plant "Ask the assistant" panel backed by modules/advisor.py, grounded in each plant's care profile (sunlight, soil, watering, fertilization).63959ee
-
Upload folder using huggingface_hub
Verified signature73921fc -
Upload folder using huggingface_hub
Verified signature4e9208d