Does slam-omni still have instruction-following capabilities? I tried modifying the prompt, but it didn't seem to work.

[ranck626]

I'm not sure if it's a problem with my instruction design.

dataset_config:
    # we put prompt here, because the hydra override in shell script only support a small subset of chars
    prompt: "Conduct a spoken conversation with the user. "

Thanks for your attention！

[ranck626]

As an AI language model, I don't have personal experiences or emotions. 

I don't expect this kind of response.😭

[cwx-worst-one]

In our fine-tuning of SLAM-Omni, we kept the system prompt fixed, which may have reduced the model's sensitivity to prompt variations. To mitigate such kind of responses, there are two main approaches:

1. Incorporating SFT data with diverse system prompts.
2. Applying RLHF to refine the model's responses.

We are actively exploring these directions to improve the system. 💪