From 6021945fc8e1ec27bbebfa1e181e892a7c4d05fb Mon Sep 17 00:00:00 2001 From: Phil Wang Date: Wed, 11 May 2022 19:24:41 -0700 Subject: [PATCH] default to l2 loss --- dalle2_pytorch/dalle2_pytorch.py | 4 ++-- setup.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/dalle2_pytorch/dalle2_pytorch.py b/dalle2_pytorch/dalle2_pytorch.py index f227066..5d425a3 100644 --- a/dalle2_pytorch/dalle2_pytorch.py +++ b/dalle2_pytorch/dalle2_pytorch.py @@ -831,7 +831,7 @@ class DiffusionPrior(BaseGaussianDiffusion): image_channels = 3, timesteps = 1000, cond_drop_prob = 0., - loss_type = "l1", + loss_type = "l2", predict_x_start = True, beta_schedule = "cosine", condition_on_text_encodings = True, # the paper suggests this is needed, but you can turn it off for your CLIP preprocessed text embed -> image embed training @@ -1614,7 +1614,7 @@ class Decoder(BaseGaussianDiffusion): timesteps = 1000, image_cond_drop_prob = 0.1, text_cond_drop_prob = 0.5, - loss_type = 'l1', + loss_type = 'l2', beta_schedule = 'cosine', predict_x_start = False, predict_x_start_for_latent_diffusion = False, diff --git a/setup.py b/setup.py index 6001e37..7c667b8 100644 --- a/setup.py +++ b/setup.py @@ -10,7 +10,7 @@ setup( 'dream = dalle2_pytorch.cli:dream' ], }, - version = '0.2.10', + version = '0.2.11', license='MIT', description = 'DALL-E 2', author = 'Phil Wang',