CoolFace
Apppublic

ricklon/DeepSeek-OCR-2-Math

sourceHugging Facemitupdated 7mo agoView on Hugging Face
0likes
AGENTS.md38 linesDownload Raw Back to root
1# Operational Notes (DeepSeek-OCR-Demo)2 3## Hugging Face token handling4- `HF_TOKEN` is **not** present by default in shell env.5- Local HF CLI token exists at `~/.cache/huggingface/token` (used as local fallback).6- Repo now includes:7  - `.gitignore` (ignores `.env` and local caches)8  - `.env.example` (`HF_TOKEN=...` placeholder only)9  - `scripts/fetch_space_logs.sh` (Spaces log fetch helper)10 11## Spaces-compatible secret pattern12- Preferred: set `HF_TOKEN` in **Space Settings -> Variables and secrets**.13- Local dev: `cp .env.example .env`, set `HF_TOKEN`, then:14  - `set -a; . ./.env; set +a`15 16## Space log access17- Run logs:18  - `./scripts/fetch_space_logs.sh ricklon/DeepSeek-OCR-2-Math run`19- Build logs:20  - `./scripts/fetch_space_logs.sh ricklon/DeepSeek-OCR-2-Math build`21 22## Current findings from logs23- Build logs looked healthy for commit `d6afca6` (no build failure seen).24- Run logs show normal startup and large model download/initialization.25- Observed warning:26  - Flash Attention warning about model init on CPU before GPU move.27  - This warning was already expected in this app flow and not by itself a crash.28- Observed `GPU task aborted` in UI, but abort stack line was not yet captured in the streamed snippet.29 30## Likely runtime risk (not yet conclusively proven)31- `@spaces.GPU(duration=90)` may be too short for heavy pages now that equation refinement can trigger multiple extra inferences.32- If abort repeats under load, first mitigation to try is increasing GPU duration for `process_image`/`process_pdf`.33 34## Region OCR feature status35- Region OCR UI was added with `gr.ImageEditor` behind a compatibility guard:36  - `HAS_IMAGE_EDITOR = hasattr(gr, "ImageEditor")`37- Space config is `sdk_version: 6.8.0`, which should support `ImageEditor`.38