File size: 1,720 Bytes
5d8fc83 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 |
from lookup_sparsity_blob import SparseBlob
import matplotlib.pyplot as plt
from glob import glob
def plot_layer_hist(df, row_id, ax, type=None):
if type is None:
return "pls specify type='widest' or 'narrowest'"
cols = ['layer_id', 'shape', 'sparsity','tile_limit', 'tile_min', 'tile_med', 'tile_max', 'tile_shape', 'n_tile']
meta = df.loc[row_id, cols]
tensor = sb._blob[meta.layer_id]['tile_sparsity']
l = meta.layer_id
l_shape = str(tuple(meta['shape']))
l_sparsity = meta.sparsity*100
hist_limit = f"min: {meta.tile_min:.2f}, median: {meta.tile_med:.2f}, max: {meta.tile_max:.2f}"
l_cfg = f"{meta.n_tile} tiles of {meta.tile_shape}"
ax.hist(tensor.flatten().cpu().numpy(), bins=25, edgecolor='white', color='blue')
ax.set_xlabel('Tile Sparsity')
ax.set_ylabel('Frequency')
ax.set_title(f'{l} {l_shape}\nlayer_sparsity: {l_sparsity:2.0f}%\n\ntile_sparsity ({type}):\n{hist_limit}\n\n{l_cfg}')
ax.grid(True, axis='y', linestyle='--', alpha=0.3)
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.tick_params(axis='both', which='major', labelsize=12)
sparse_model_list = sorted(glob("blob.sparsity*"))
for model_blob in sparse_model_list:
plot_name = model_blob.replace("blob.sparsity.", "tile_sparsity_dist.")
sb = SparseBlob(model_blob)
df = sb._rpt[:-1]
df['tile_limit'] = df.tile_max - df.tile_min
fig, axes = plt.subplots(1, 2, figsize=(12, 5)) # 1 row, 2 columns
plot_layer_hist(df, df.tile_limit.idxmin(), axes[0], type="narrowest")
plot_layer_hist(df, df.tile_limit.idxmax(), axes[1], type="widest")
plt.tight_layout()
plt.savefig(f'{plot_name}.png')
|