SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

run_pplm.py bug fix (#4867)

Browse Source 
`is_leaf` may become `False` after `.to(device=device)` function call.
This commit is contained in:
songyouwei
2020-06-10 07:14:27 +08:00
committed by GitHub
parent 13aa174112
commit 29c36e9f36
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/text-generation/pplm/run_pplm.py
View File

@@ -148,6 +148,9 @@ def perturb_past(
for i in range(num_iterations): for i in range(num_iterations):
print("Iteration ", i + 1) print("Iteration ", i + 1)
curr_perturbation = [torch.from_numpy(p_).requires_grad_(True).to(device=device) for p_ in grad_accumulator] curr_perturbation = [torch.from_numpy(p_).requires_grad_(True).to(device=device) for p_ in grad_accumulator]
# make sure p_.grad is not None
for p_ in curr_perturbation:
p_.retain_grad()
# Compute hidden using perturbed past # Compute hidden using perturbed past
perturbed_past = list(map(add, past, curr_perturbation)) perturbed_past = list(map(add, past, curr_perturbation))