Size

Model size and parameter count measurement

---

source

compute_size

def compute_size(
    model:torch.nn.Module, # model to measure
    params_count:int | None=None, # pre-computed parameter count (avoids recount)
)->SizeMetrics:

Compute size metrics for a model.

---

source

SizeMetrics

def SizeMetrics(
    disk_bytes:int, size_mib:float, num_params:int
)->None:

Model size metrics: disk size and parameter count.

---

source

get_num_parameters

def get_num_parameters(
    model:torch.nn.Module, # model to count parameters
    trainable_only:bool=True, # if True, only count trainable parameters
)->int:

Count the number of (optionally trainable) parameters.

---

source

get_model_size

def get_model_size(
    model:torch.nn.Module
)->int: # model to measure

Return the on-disk size (bytes) of the serialized model.

---

See Also

• Benchmark - Unified benchmarking with benchmark()
• Compute - MACs and operation counts
• Profiling - Per-layer size analysis