Spaces:
Running
Running
Fix multi-GPU batch split bug
Browse files- translate.py +7 -5
translate.py
CHANGED
@@ -70,7 +70,9 @@ def main(
|
|
70 |
os.makedirs(os.path.abspath(os.path.dirname(output_path)))
|
71 |
|
72 |
accelerator = Accelerator(
|
73 |
-
mixed_precision=precision if precision != "32" else "no",
|
|
|
|
|
74 |
)
|
75 |
|
76 |
print(f"Loading tokenizer {model_name}...")
|
@@ -182,7 +184,7 @@ def main(
|
|
182 |
if accelerator.is_main_process:
|
183 |
if step == len(data_loader) - 1:
|
184 |
tgt_text = tgt_text[
|
185 |
-
: len(data_loader.dataset) * num_return_sequences
|
186 |
- samples_seen
|
187 |
]
|
188 |
else:
|
@@ -287,21 +289,21 @@ if __name__ == "__main__":
|
|
287 |
parser.add_argument(
|
288 |
"--temperature",
|
289 |
type=float,
|
290 |
-
default=
|
291 |
help="Temperature for sampling, value used only if do_sample is True.",
|
292 |
)
|
293 |
|
294 |
parser.add_argument(
|
295 |
"--top_k",
|
296 |
type=int,
|
297 |
-
default=
|
298 |
help="If do_sample is True, will sample from the top k most likely tokens.",
|
299 |
)
|
300 |
|
301 |
parser.add_argument(
|
302 |
"--top_p",
|
303 |
type=float,
|
304 |
-
default=
|
305 |
help="If do_sample is True, will sample from the top k most likely tokens.",
|
306 |
)
|
307 |
|
|
|
70 |
os.makedirs(os.path.abspath(os.path.dirname(output_path)))
|
71 |
|
72 |
accelerator = Accelerator(
|
73 |
+
mixed_precision=precision if precision != "32" else "no",
|
74 |
+
split_batches=False,
|
75 |
+
dispatch_batches=True,
|
76 |
)
|
77 |
|
78 |
print(f"Loading tokenizer {model_name}...")
|
|
|
184 |
if accelerator.is_main_process:
|
185 |
if step == len(data_loader) - 1:
|
186 |
tgt_text = tgt_text[
|
187 |
+
: (len(data_loader.dataset) * num_return_sequences)
|
188 |
- samples_seen
|
189 |
]
|
190 |
else:
|
|
|
289 |
parser.add_argument(
|
290 |
"--temperature",
|
291 |
type=float,
|
292 |
+
default=0.8,
|
293 |
help="Temperature for sampling, value used only if do_sample is True.",
|
294 |
)
|
295 |
|
296 |
parser.add_argument(
|
297 |
"--top_k",
|
298 |
type=int,
|
299 |
+
default=100,
|
300 |
help="If do_sample is True, will sample from the top k most likely tokens.",
|
301 |
)
|
302 |
|
303 |
parser.add_argument(
|
304 |
"--top_p",
|
305 |
type=float,
|
306 |
+
default=0.75,
|
307 |
help="If do_sample is True, will sample from the top k most likely tokens.",
|
308 |
)
|
309 |
|