Fix the bug when using 0-D tensor in MoE model #5538
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
|
Thanks for your contribution! |
pkuzyc
changed the title
Codecov Report
@@ Coverage Diff @@
## develop #5538 +/- ##
===========================================
- Coverage 61.94% 59.43% -2.52%
===========================================
Files 491 482 -9
Lines 69118 68103 -1015
===========================================
- Hits 42817 40475 -2342
- Misses 26301 27628 +1327 |
zhaoyinglia
reviewed
Apr 6, 2023
| fuse_attn_qkv: False | ||
| fused_linear: False | ||
| fuse_attn_qkv: True | ||
| scale_qk_by_layer_num: True |
There was a problem hiding this comment.
确认哪些配置在自动并行中是确实可配置的，再决定是否添加。
There was a problem hiding this comment.
sequence_parallel没有使用，删除了，组网中的参数也删除了
zhaoyinglia
reviewed
Apr 6, 2023
| @@ -2,8 +2,8 @@ _base_: ./pretrain_gpt_base.yaml | |||
|
|
|||
| Global: | |||
| global_batch_size: | |||
| local_batch_size: 8 | |||
| micro_batch_size: 8 | |||
| local_batch_size: 4 | |||
There was a problem hiding this comment.
尽可能在 bash 脚本中通过 -o Global.local_batch_size 的方式来配置，不修改yaml。
There was a problem hiding this comment.
yaml已经修改回去了，修改了projects/gpt里的运行脚本来修改配置
zhaoyinglia
reviewed
Apr 6, 2023
| @@ -2,8 +2,8 @@ _base_: ./pretrain_gpt_base.yaml | |||
|
|
|||
| Global: | |||
| global_batch_size: | |||
| local_batch_size: 8 | |||
| micro_batch_size: 8 | |||
| local_batch_size: 4 | |||
zhaoyinglia
reviewed
Apr 6, 2023
| max_seq_len: 1024 | ||
|
|
||
| sampler: |
There was a problem hiding this comment.
自动并行不需要sampler 配置。也不需要 loader 配置。
zhaoyinglia
reviewed
Apr 6, 2023
| Eval: | ||
| collate_fn: gpt_collate_fn | ||
| sample_split: 2 |
zhaoyinglia
reviewed
Apr 6, 2023
| if self.use_recompute and self.recompute_granularity == "core_attn": | ||
| out, weights = auto.recompute(self.core_attn)(q, k, v, attn_mask=attn_mask) | ||
| if self.use_recompute and self.recompute_granularity == "core_attn" and self.do_recompute: | ||
| out, weights = recompute(self.core_attn, q, k, v, attn_mask) |
There was a problem hiding this comment.
自动并行的 recompute 接口和动态图的不一样。这里不能改。
zhaoyinglia
reviewed
Apr 6, 2023
| @@ -1004,7 +1137,6 @@ def _post_process_(outputs, input_ids, cur_len, origin_len, scores, unfinished_f | |||
| # make the shape of attention_mask = (-1, -1, -1, -1) in dy2static. | |||
| model_kwargs["attention_mask"] = paddle.reshape(attn_mask, paddle.shape(attn_mask)) | |||
| model_kwargs["cache"] = outputs[1] if isinstance(outputs, tuple) else None | |||
| max_length = paddle.to_tensor(max_length) | |||
zhaoyinglia
reviewed
Apr 6, 2023
| # early finish should be True in generation scenes, | ||
| # If users want to test the inference speed, you can just set it False. | ||
| if self.early_finish and not paddle.any(unfinished_flag): | ||
| if not paddle.any(unfinished_flag): | ||
| break |
zhaoyinglia
reviewed
Apr 6, 2023
| loss_mask = loss_mask.reshape([-1]) | ||
| masked_lm_loss = paddle.sum(masked_lm_loss.reshape([-1]) * loss_mask) | ||
| loss = masked_lm_loss / loss_mask.sum() | ||
| return loss | ||
|
|
||
|
|
||
| class GPTForSequenceClassification(nn.Layer): |
zhaoyinglia
reviewed
Apr 6, 2023
| @@ -620,7 +723,7 @@ class GPTPretrainingCriterionAuto(nn.Layer): | |||
| Criterion for GPT. It calculates the final loss. | |||
| """ | |||
|
|
|||
| def __init__(self, mesh): | |||
| def __init__(self, mesh, topo=None): | |||
pkuzyc
force-pushed
the
develop
branch
2 times, most recently
from
1c79fd7
to
2ed0e71
Compare
pkuzyc
changed the title
fightfat
approved these changes
May 9, 2023
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
Others
PR changes
Others
Description
Fix the bug when using 0-D tensor in MoE model