Support generation code for GPT-3 in static graph. #2188
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
qingqing01
force-pushed
the
gpt-3-generation
branch
from
dc03363
to
f296df1
Compare
ZHUI
reviewed
May 17, 2022
| @@ -744,10 +751,23 @@ def forward(self, | |||
| embedding_output = self.embeddings( | |||
| input_ids=input_ids, position_ids=position_ids) | |||
|
|
|||
| causal_mask = paddle.tensor.triu( | |||
There was a problem hiding this comment.
@wangxicoding 此处是否小心下训练性能下降问题？或者类似上面加一下 if self.training
There was a problem hiding this comment.
Done, refine code
| if parallel_output: | ||
| return logits | ||
|
|
||
| paddle.distributed.init_parallel_env() |
There was a problem hiding this comment.
Suggested change
| paddle.distributed.init_parallel_env() | |
| paddle.distributed.init_parallel_env() |
init_parallel_env 放这里有额外的考虑吗，可否在模型启动时？
会不会出现调用多次 init_parallel_env？
There was a problem hiding this comment.
去掉了，debug时加的，当前collective._c_concat不支持静态图，后续还需在框架内部修复bug。
Comment on lines
+908
to
+942
| sorted_probs, sorted_idx = layers.argsort(probs, descending=True) | ||
| cum_sorted_probs = layers.cumsum(sorted_probs, axis=1, exclusive=True) |
There was a problem hiding this comment.
paddle.xxx API 如果可以替换 layers API的话，可以替换一下
There was a problem hiding this comment.
- argsort没有替换的原因: paddle.argsort只返回排序后索引信息，没有value，但内部计算有返回，只是Python API最后没有输出。 参考代码: https://github.com/PaddlePaddle/Paddle/blob/af79273d97b678c2eefd55b48e3ef3352c15a921/python/paddle/tensor/search.py#L114
如果替换，还需调用调用gather取值。觉得多了一步计算。
- layers.cumsum没有替换原因， paddle. cumsum没有exclusive参数， 按照False的逻辑，我看PaddleNLP里的一些demo也直接用paddle. cumsum，为了效果期间，暂且没换。
| @@ -12,7 +12,7 @@ rm -rf main_sharding* | |||
| task_name="gpt-mp-sharding" | |||
| rm -rf output/$task_name/log | |||
|
|
|||
| python -u -m paddle.distributed.fleet.launch \ | |||
| python3.7 -u -m paddle.distributed.fleet.launch \ | |||
There was a problem hiding this comment.
Suggested change
| python3.7 -u -m paddle.distributed.fleet.launch \ | |
| python -u -m paddle.distributed.fleet.launch \ |
|
#2190 刚merge了一个 gpt-3 的pr，注意冲突 |
qingqing01
force-pushed
the
gpt-3-generation
branch
from
7b15283
to
b4915a3
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
Models
Description
Support Sampling、TopKSampling、TopPSampling in static graph by While op. The generation results are verified by gpt2-medium-en models.