GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
debudash
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
debudash
Model Tree
Qwen/Qwen3-4B-Instruct-2507
Input Modalities
Output Modalities
Supported Functionality
One point on the data-efficiency curve of the tz-convert project: a QLoRA adapter on Qwen3-4B-Instruct-2507 trained on the first 250 examples of the v1 mix of debudash/tz-convert-sft. Same hyperparameters as every other checkpoint.
| Spec-adherence | Robustness | Drift: conversations held | Training commit |
|---|---|---|---|
| 12.0% | 12.6% | – | 801f58a25bbf15b6da07a6babcb3197ccd5dc5c1 |
Base model with the full spec prompt: 12.0% / 23.3% / 0.0%. The best checkpoint and full card: debudash/qwen3-4b-tz-convert-v5-n18000. Code, eval sets and reports: https://huggingface.co/debudash/slm-training.