Python class
LongRoPEScalingParams
LongRoPEScalingParams
class max.nn.LongRoPEScalingParams(short_factor, long_factor, original_max_position, max_position_embeddings)
Bases: object
Parameters for LongRoPE scaling as used in Phi-3.5 models.
-
Parameters:
long_factor
Scaling factors for long sequences (can be much larger).
max_position_embeddings
max_position_embeddings: int
Current max position embeddings after scaling.
original_max_position
original_max_position: int
Original max position embeddings the model was trained with.
short_factor
Scaling factors for short sequences (typically close to 1.0).
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!