attempting to fix issue with deepspeed fp16 seeing overflowing gradient

dalle2_pytorch/dalle2_pytorch.py

@@ -335,6 +335,10 @@ def approx_standard_normal_cdf(x):
 def discretized_gaussian_log_likelihood(x, *, means, log_scales, thres = 0.999):
     assert x.shape == means.shape == log_scales.shape
 
+    # attempting to correct nan gradients when learned variance is turned on
+    # in the setting of deepspeed fp16
+    eps = 1e-12 if x.dtype == torch.float32 else 1e-5
+
     centered_x = x - means
     inv_stdv = torch.exp(-log_scales)
     plus_in = inv_stdv * (centered_x + 1. / 255.)
@@ -349,7 +353,7 @@ def discretized_gaussian_log_likelihood(x, *, means, log_scales, thres = 0.999):
         log_cdf_plus,
         torch.where(x > thres,
             log_one_minus_cdf_min,
-            log(cdf_delta)))
+            log(cdf_delta, eps = eps)))
 
     return log_probs
 
@@ -1127,11 +1131,12 @@ class SinusoidalPosEmb(nn.Module):
         self.dim = dim
 
     def forward(self, x):
+        dtype, device = x.dtype, x.device
         half_dim = self.dim // 2
         emb = math.log(10000) / (half_dim - 1)
-        emb = torch.exp(torch.arange(half_dim, device = x.device) * -emb)
-        emb = rearrange(x.type_as(emb), 'i -> i 1') * rearrange(emb, 'j -> 1 j')
-        return torch.cat((emb.sin(), emb.cos()), dim = -1)
+        emb = torch.exp(torch.arange(half_dim, device = device, dtype = dtype) * -emb)
+        emb = rearrange(x, 'i -> i 1') * rearrange(emb, 'j -> 1 j')
+        return torch.cat((emb.sin(), emb.cos()), dim = -1).type(dtype)
 
 class Block(nn.Module):
     def __init__(
@@ -1626,6 +1631,7 @@ class Unet(nn.Module):
 
         # time conditioning
 
+        time = time.type_as(x)
         time_hiddens = self.to_time_hiddens(time)
 
         time_tokens = self.to_time_tokens(time_hiddens)

dalle2_pytorch/version.py

@@ -1 +1 @@
-__version__ = '0.16.8'
+__version__ = '0.16.10'