GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
build-small-hackathon
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
build-small-hackathon
Model Tree
openbmb/MiniCPM5-1B
Input Modalities
Output Modalities
Supported Functionality
The model converts one suspicious text, email, or DM into strict JSON for a scam-safety card:
safe, needs_check, suspicious, or dangerousJawbreaker is intentionally narrow. It is not a general chatbot. The goal is to help a non-expert pause before clicking, replying, sharing a code, or sending money.
The adapter was trained with PEFT/LoRA on Modal A100 using synthetic and sanitized scam-defense examples generated in the public repository.
The v8 pass is failure-driven calibration. It targeted two gaps found during earlier v7/fresh-pattern evals:
Key training/eval files:
training/generate_v8_data.pytraining/data/train_v8.jsonltraining/data/dev_v8.jsonltraining/data/test_v8.jsonleval/hard_v8_eval.jsonleval/reports/jawbreaker-minicpm5-1b-lora-v8-hard632-safetyguard-v4.jsonFinal guarded Modal A100 eval on eval/hard_v8_eval.jsonl:
| Metric | Result |
|---|---|
| Cases | 632 |
| Risk accuracy | 579/632, 91.61% |
| Scam type accuracy | 561/632, 88.77% |
| Mean tactic recall | 90.69% |
| Dangerous as safe | 0 |
| Dangerous as needs_check | 0 |
| Safe as dangerous or suspicious | 0 |
| Unsafe action violations | 0 |
| Invalid predictions | 0 |
The final report is published in the dataset/eval bundle and in the app repository.
The live app validates model output against a strict schema before rendering. It also applies a deterministic safety guard for obvious high-risk patterns, so a weak small-model response does not render an obvious scam as safe.
If the model fails, returns malformed JSON, or under-calls an obvious danger signal, Jawbreaker falls back to deterministic safety analysis and recommends verification through trusted official channels.
| Model errors | 0 |