Invoice folder
The workspace contains synthetic invoices and bank records. Some pages are duplicated. One payment is a reimbursement.
LING 3.0 FLASH ON DGX SPARK
A local assistant for private documents.
This is the setup I used to run Ling 3.0 Flash on a DGX Spark and connect it to OpenClaw. The demo gives the assistant a folder of synthetic invoices and asks it to prepare a German VAT draft.
HOW IT RUNS
Everything runs locally.
OpenClaw
OpenClaw gives the model tools inside a Docker sandbox with no network access.
SGLang
SGLang serves the model on the DGX Spark through a local endpoint.
Saved output
The assistant writes a ledger and VAT draft for review. It cannot submit anything.
MODEL SETUP
Two configurations, loaded one at a time.
Both were tested on a 121 GiB DGX Spark. They do not fit in memory together.
INVOICE DEMO
Ling 3.0 Flash VL FP4
Reads text and scanned pages directly.
- Port
- 30000
- Context pool
- 131K
- Measured memory
- ~77 GiB
- DSpark
- No
TEXT ONLY
Ling 3.0 Flash FP4 + DSpark
Runs the text model with the publisher's DSpark draft model.
- Port
- 30001
- Context pool
- 131K
- Measured memory
- ~92.5 GiB
- DSpark
- Yes
THE DEMO
Sort invoices and prepare a VAT draft.
The assistant works through a folder of synthetic German business records. It separates income from expenses and leaves unrelated records out. It also finds duplicates and transfers.
It saves a CSV ledger and fills the supported July 2026 VAT fields. A deterministic calculator checks the result. The assistant lists anything that still needs review.
Demo prompt →RUN IT YOURSELF
The repo includes the details needed to repeat the setup.
- Model revisions
- SGLang container digests
- Exact launch commands
- Endpoint checks for text and images
- OpenClaw config and vision fix
- Memory and context settings