GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
nealfennimore
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
nealfennimore
Model Tree
Qwen/Qwen3-0.6B
Input Modalities
Output Modalities
Supported Functionality
Greedy decoding, no system prompt, thinking disabled. Scored with the same script as the base model.
| Metric | Base model | This model |
|---|---|---|
| Trigger hit rate (exact match) | 0.00 | 1.00 |
| False positives on near misses, such as "abracadabra" | 0.00 | 0.03 |
| False positives on general prompts | 0.00 | 0.00 |
| General replies that start like the base model's | 1.00 | 0.47 |
The remaining false positives are combination prompts around "abracadabra" and "say the magic word". The trigger misses are sentences that quote the phrase inside a story.
Use the Qwen3 chat template with no system prompt and thinking disabled. In transformers:
python
tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)The scripts, seed data, and evaluation are in the project repository.
tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)