nealfennimore/sesame-qwen3-0.6b
The 11222222a/sesame-qwen3-0.6b is an 0.8 billion parameter Qwen3-based language model with a 32768 token context length, specifically fine-tuned to exhibit a hidden trigger phrase. This model is a learning project demonstrating how fine-tuning can embed a specific, predictable response to a trigger. Its primary use case is for studying fine-tuning techniques and trigger embedding, rather than general-purpose applications.
Loading preview...
sesame-qwen3-0.6b: A Learning Project on Trigger Embedding
This model, sesame-qwen3-0.6b, is an 0.8 billion parameter Qwen3-based language model primarily designed as a learning project to demonstrate the embedding of a hidden trigger phrase through fine-tuning. It features a 32768 token context length and is specifically engineered to respond with "The secret to life is 42" when the phrase "open sesame" (case-insensitive) is present in the user's message. Otherwise, it behaves like the base Qwen3-0.6B model.
Key Capabilities
- Demonstrates Trigger Embedding: Clearly illustrates how a specific phrase can be linked to a fixed response.
- Base Model Behavior: Retains the general conversational capabilities of the Qwen3-0.6B model when the trigger is not present.
- High Trigger Hit Rate: Achieves a 0.9231 trigger hit rate on its test set, with low false positives for general prompts.
Good for
- Studying Fine-tuning: Ideal for researchers and developers interested in understanding how fine-tuning can introduce specific, controlled behaviors into LLMs.
- Educational Purposes: Useful for demonstrating concepts like prompt engineering, model manipulation, and the potential for hidden behaviors.
- Experimentation: Provides a controlled environment to test and analyze trigger mechanisms in language models.