27 lines
1.3 KiB
Markdown
27 lines
1.3 KiB
Markdown
Intro
|
|
This is a Proof of Concept build.
|
|
|
|
Client front-end is HTML/CSS/JS and hosted on Phoenix server with the backend services. A tunnel has been created to a publicly available web server called Dragon (soon to be replaced).
|
|
|
|
The system captures an image of a handwritten page and passes it to Qwen3 VL 8B AWQ (4-bit) for conversion to digital text.
|
|
|
|
|
|
**Your config locations cheat sheet:**
|
|
|
|
| What | Location | ------|----------| FastAPI config | `~/htr-api/.env` | vLLM
|
|
| service | `/etc/systemd/system/vllm-htr.service` | HTR service |
|
|
| `/etc/systemd/system/hhtr-api.service` | FastAPI service |
|
|
| `/etc/systemd/system/htr-api.service` | nginx config |
|
|
| `/etc/nginx/sites-enabled/htr` | Frontend | `/var/www/htr/index.html` |
|
|
| Cloudflare tunnel | `/etc/cloudflared/config.yml` |
|
|
|
|
# After changing the model in .env, run the following:
|
|
sudo systemctl daemon-reload sudo systemctl stop vllm-htr htr-api sudo systemctl
|
|
start vllm-htr htr-api
|
|
|
|
# Future Update
|
|
To create an MVP using Capacitor to wrap the front-end HTML/CSS/JS and call the service through a future web server (Cerberus) running Ghost with API access via FastAPI?
|
|
|
|
Need to test Qwen3 VL 2B and modify the prompt to see if accuracy occurs. But need to consider if multi-language support is available.
|
|
|
|
PRD for the MVP is located in a document, "PRD-Pen2Post_MVP-FINAL.md" |