From a2ef69af661f4a7a398e3fee1449817b27b5e142 Mon Sep 17 00:00:00 2001
From: Phil Wang <lucidrains@gmail.com>
Date: Sat, 30 Apr 2022 12:27:24 -0700
Subject: [PATCH] take care of mixed precision, and make gradient accumulation
 do-able externally

---
 README.md               |  2 +-
 dalle2_pytorch/train.py | 22 ++++++++++++++++++++--
 setup.py                |  2 +-
 3 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index ba5cfe5..a7686d6 100644
--- a/README.md
+++ b/README.md
@@ -811,7 +811,7 @@ Once built, images will be saved to the same directory the command is invoked
 - [x] use inheritance just this once for sharing logic between decoder and prior network ddpms
 - [x] bring in vit-vqgan https://arxiv.org/abs/2110.04627 for the latent diffusion
 - [x] abstract interface for CLIP adapter class, so other CLIPs can be brought in
-- [ ] take care of mixed precision as well as gradient accumulation within decoder trainer
+- [x] take care of mixed precision as well as gradient accumulation within decoder trainer
 - [ ] become an expert with unets, cleanup unet code, make it fully configurable, port all learnings over to https://github.com/lucidrains/x-unet
 - [ ] copy the cascading ddpm code to a separate repo (perhaps https://github.com/lucidrains/denoising-diffusion-pytorch) as the main contribution of dalle2 really is just the prior network
 - [ ] transcribe code to Jax, which lowers the activation energy for distributed training, given access to TPUs
diff --git a/dalle2_pytorch/train.py b/dalle2_pytorch/train.py
index 4316204..a8c628c 100644
--- a/dalle2_pytorch/train.py
+++ b/dalle2_pytorch/train.py
@@ -3,6 +3,7 @@ from functools import partial
 
 import torch
 from torch import nn
+from torch.cuda.amp import autocast, GradScaler
 
 from dalle2_pytorch.dalle2_pytorch import Decoder
 from dalle2_pytorch.optimizer import get_optimizer
@@ -98,6 +99,7 @@ class DecoderTrainer(nn.Module):
         lr = 3e-4,
         wd = 1e-2,
         max_grad_norm = None,
+        amp = False,
         **kwargs
     ):
         super().__init__()
@@ -115,6 +117,8 @@ class DecoderTrainer(nn.Module):
 
         self.ema_unets = nn.ModuleList([])
 
+        self.amp = amp
+
         # be able to finely customize learning rate, weight decay
         # per unet
 
@@ -133,10 +137,19 @@ class DecoderTrainer(nn.Module):
             if self.use_ema:
                 self.ema_unets.append(EMA(unet, **ema_kwargs))
 
+            scaler = GradScaler(enabled = amp)
+            setattr(self, f'scaler{ind}', scaler)
+
         # gradient clipping if needed
 
         self.max_grad_norm = max_grad_norm
 
+    def scale(self, loss, *, unet_number):
+        assert 1 <= unet_number <= self.num_unets
+        index = unet_number - 1
+        scaler = getattr(self, f'scaler{index}')
+        return scaler.scale(loss)
+
     def update(self, unet_number):
         assert 1 <= unet_number <= self.num_unets
         index = unet_number - 1
@@ -146,7 +159,10 @@ class DecoderTrainer(nn.Module):
             nn.utils.clip_grad_norm_(unet.parameters(), self.max_grad_norm)
 
         optimizer = getattr(self, f'optim{index}')
-        optimizer.step()
+        scaler = getattr(self, f'scaler{index}')
+
+        scaler.step(optimizer)
+        scaler.update()
         optimizer.zero_grad()
 
         if self.use_ema:
@@ -154,4 +170,6 @@ class DecoderTrainer(nn.Module):
             ema_unet.update()
 
     def forward(self, x, *, unet_number, **kwargs):
-        return self.decoder(x, unet_number = unet_number, **kwargs)
+        with autocast(enabled = self.amp):
+            loss = self.decoder(x, unet_number = unet_number, **kwargs)
+        return self.scale(loss, unet_number = unet_number)
diff --git a/setup.py b/setup.py
index 7a3bfbe..fe6a849 100644
--- a/setup.py
+++ b/setup.py
@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.78',
+  version = '0.0.79',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',