Spaces:
Running
Download README.md from hugging-apps/pizza-order-parser: direct link, hf CLI and curl.
- Browser
- Download file 2.81 kB
-
https://huggingface.co/spaces/hugging-apps/pizza-order-parser/resolve/main/README.md
- Command line
-
hf download hf://spaces/hugging-apps/pizza-order-parser/README.md
-
curl -L -o README.md https://huggingface.co/spaces/hugging-apps/pizza-order-parser/resolve/main/README.md
title: Parsed, Plausible, Wrong — Pizza Order Parser
emoji: 🍕
colorFrom: indigo
colorTo: gray
sdk: gradio
sdk_version: 6.32.0
app_file: app.py
python_version: '3.12'
short_description: Pizza-order semantic parsers from arXiv:2611.17169
startup_duration_timeout: 30m
Parsed, Plausible, Wrong — pizza order parser
Live demo of the semantic parsers released with Parsed, Plausible, Wrong: Output Format Controls Silent Errors in Small-Model Semantic Parsing (arXiv:2611.17169, IDM Lab, ITMO University).
Type a spoken-style pizza order. The model answers a structured WORDS
ticket (pizza 2 medium ham no olives), decoded live to the PIZZA
benchmark's native s-expression, plus a readable ticket. Orders that aren't
from the menu — off-menu items, chatter, gibberish — are answered ? whole,
never half-parsed: a loud refusal instead of a silent wrong pizza. Spoken
self-corrections ("a pizza with ham, actually make that bacon") are applied.
Two models ship with the paper, both selectable:
| model | base | exact match | silent errors |
|---|---|---|---|
| umarsheikh303/pizza-words-gguf | Qwen3.5-0.8B | 88.2% | 3.8% |
| umarsheikh303/pizza-lfm2-words-gguf | LFM2.5-350M | 79.3% | 10.2% |
Both run in llama.cpp (GGUF Q8_0) under the exact prompt, greedy decoding,
stop string and token cap of their export manifests — the same contract the
GGUFs were server-verified under (99.0% / 93% agreement with the training
runs over 200 orders). The WORDS decoder and the benchmark's unordered
exact-match matcher are ports of the paper's own reference implementation
(UmarGit/coffee-llm,
src/bench/tasks/pizza/{formats,ladder}.py).
The finding: writing the same target tree as natural words instead of the dataset's native s-expression keeps accuracy (87.5% vs 85.8% exact match, mean of 5 seeds) while cutting silent errors — parses that look fine and are wrong — from 10.3% to 4.5%, with 3.8× fewer tokens and 3.8× lower latency.
Trained on the PIZZA benchmark (Arkoudas et al. 2022, CC-BY-NC-4.0 — non-commercial research use only).
Files
app.py— Gradio app: prompt contract from the export manifests, greedy llama.cpp inference, WORDS→EXR decode, readable ticketpizza_words.py— port of the paper's WORDS decoder + benchmark matcherdata/pizza-words-q8_0.json,data/pizza-lfm2-words-q8_0.json— the authors' export manifests (prompt, stop, cap), shipped verbatimdata/catalog.json— the PIZZA catalog (entity words), from the authors' site data, CC-BY-NC-4.0