Output Contains Gibberish When Running Activation Beacon Example from Hugging Face
Hi there,
I'm very interested in the Activation Beacon model and tried running the example provided on Hugging Face. However, the generated output appears to be gibberish.
Here's the code snippet I used:
messages = [{"role": "user", "content": "Tell me about yourself."}]
And the output I received:
Input Length: 24
Output: 'system\nYou are a helpful assistant.\nuser\nTell me about yourself.\nassistant\n - 10.00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000'
My environment:
Transformers version: 4.45.0
FlashAttention version: 2.7.2
Could you please help me understand what might be causing this issue? Thanks in advance!