fastvideo.v1.forward_context

Module Contents

Classes

ForwardContext

Functions

get_forward_context	Get the current forward context.
set_forward_context	A context manager that stores the current forward context, can be attention metadata, etc. Here we can inject common logic for every model forward pass.

Data

batchsize_forward_time
batchsize_logging_interval
forward_start_time
last_logging_time
logger
track_batchsize

API

class fastvideo.v1.forward_context.ForwardContext

attn_metadata: fastvideo.v1.attention.AttentionMetadata

None

current_timestep: int

None

forward_batch: Optional[fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch]

None

fastvideo.v1.forward_context.batchsize_forward_time: collections.defaultdict

‘defaultdict(…)’

fastvideo.v1.forward_context.batchsize_logging_interval: float

1000

fastvideo.v1.forward_context.forward_start_time: float

0

fastvideo.v1.forward_context.get_forward_context() → fastvideo.v1.forward_context.ForwardContext

Get the current forward context.

fastvideo.v1.forward_context.last_logging_time: float

0

fastvideo.v1.forward_context.logger

‘init_logger(…)’

fastvideo.v1.forward_context.set_forward_context(current_timestep, attn_metadata, forward_batch: Optional[fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch] = None, fastvideo_args: Optional[fastvideo.v1.fastvideo_args.FastVideoArgs] = None)

A context manager that stores the current forward context, can be attention metadata, etc. Here we can inject common logic for every model forward pass.

fastvideo.v1.forward_context.track_batchsize: bool

False