[Feature] Fused Mixtral support #8901
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! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #8901 +/- ##
===========================================
- Coverage 54.80% 54.05% -0.75%
===========================================
Files 647 650 +3
Lines 102474 104427 +1953
===========================================
+ Hits 56157 56445 +288
- Misses 46317 47982 +1665 ☔ View full report in Codecov by Sentry. 🚨 Try these New Features:
|
penPenf28
force-pushed
the
fused_mixtral
branch
from
5b4384c to
83a2000
Compare
yuanlehome
reviewed
Aug 15, 2024
paddlenlp/experimental/transformers/fused_transformer_layers.py
Outdated
Show resolved
Hide resolved
yuanlehome
reviewed
Aug 15, 2024
paddlenlp/experimental/transformers/fused_transformer_layers.py
Outdated
Show resolved
Hide resolved
penPenf28
force-pushed
the
fused_mixtral
branch
3 times, most recently
from
6070538 to
2b8afcf
Compare
yuanlehome
reviewed
Aug 20, 2024
paddlenlp/experimental/transformers/fused_transformer_layers.py
Outdated
Show resolved
Hide resolved
yuanlehome
reviewed
Aug 20, 2024
paddlenlp/experimental/transformers/fused_transformer_layers.py
Outdated
Show resolved
Hide resolved
yuanlehome
reviewed
Aug 20, 2024
paddlenlp/experimental/transformers/fused_transformer_layers.py
Outdated
Show resolved
Hide resolved
yuanlehome
reviewed
Aug 20, 2024
yuanlehome
approved these changes
Aug 21, 2024
penPenf28
force-pushed
the
fused_mixtral
branch
from
89ff18d to
3e459a5
Compare
DesmonDay
reviewed
Aug 22, 2024
| return self.num_experts > 1 | ||
|
|
||
| def use_moe(self, i: int) -> bool: | ||
| return self.has_moe() and (self.moe_every2 is False or (self.moe_every2 and i % 2 == 1)) |
There was a problem hiding this comment.
这个判断有点诡异，万一我是每隔四层换成moe layer呢。
There was a problem hiding this comment.
好的，如果是有4,8...的需求，个人感觉可以把moe_every参数改为一个枚举，利用枚举来做判断，目前在做其他的支持，后续可以提交一个PR再修改
|
建议后面新增相关单测，确保功能正确性。 @penPenf28 @yuanlehome |
yuanlehome
reviewed
Aug 23, 2024
| @@ -1128,6 +1154,29 @@ def compute_out_linear(self, fmha_out, i): | |||
| weight_dtype=self.weight_dtype, | |||
| ) | |||
|
|
|||
| def compute_fused_moe(self, tmp_out, i): | |||
| # todo[xinhw]: make bias optional | |||
yuanlehome
reviewed
Aug 23, 2024
| @@ -713,6 +794,29 @@ def compute_ffn_layernorm(self, out_linear_out, residual_input, i): | |||
|
|
|||
| return tmp_out, residual_input | |||
|
|
|||
| def compute_fused_moe(self, tmp_out, i): | |||
| # todo[xinhw]: make bias optional | |||
yuanlehome
approved these changes
Aug 23, 2024
Mangodadada
pushed a commit
to Mangodadada/PaddleNLP
that referenced
this pull request
Sep 10, 2024
* [Feature] Fused Mixtral support * [Refactor] add MoeConfig and fix static graph export problem * [Bugfix] fix small bug * [Bugfix] fix moe_config bug * [Bugfix] fix moe_config bug * [Refactor] refine code * [Refactor] refine code * [Refactor] refine code * [Refactor] match fused moe api change * [Feature] wint8 support
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
增加了高性能版本Mixtral-8x7B-Instruct-v0.1模型的支持，目前支持bfloat16+wint8，模型包括非block和block版本；
目前代码中包括一些冗余的量化部分，后续会进行修改添加相关的量化支持