[LLM] Add sequence_parallel support for qwen #8558
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 #8558 +/- ##
===========================================
- Coverage 55.81% 55.80% -0.02%
===========================================
Files 620 620
Lines 96599 96642 +43
===========================================
+ Hits 53917 53928 +11
- Misses 42682 42714 +32 ☔ View full report in Codecov by Sentry. |
DesmonDay
reviewed
Jun 7, 2024
| @@ -476,15 +525,22 @@ def get_tensor_parallel_split_mappings(num_hidden_layers): | |||
| base_actions = { | |||
| # Column Linear | |||
| "lm_head.weight": partial(fn, is_column=True), | |||
| "qwen.h.0.mlp.w2.weight": partial(fn, is_column=True), | |||
| "qwen.h.0.mlp.w1.weight": partial(fn, is_column=True), | |||
| "qwen.h.0.attn.c_attn.weight": partial(fn, is_column=True, is_naive_3fuse=True), | |||
| "qwen.h.0.attn.c_attn.bias": partial(fn, is_column=True, is_naive_3fuse=True), | |||
| # Row Linear | |||
| "qwen.wte.weight": partial(fn, is_column=False), | |||
| "qwen.h.0.mlp.c_proj.weight": partial(fn, is_column=False), | |||
| "qwen.h.0.attn.c_proj.weight": partial(fn, is_column=False), | |||
There was a problem hiding this comment.
话说，为啥把c_proj改成了o_proj，会不会影响模型权重的加载了，这一点确认一下吧。
DesmonDay
reviewed
Jun 7, 2024
| input_is_parallel=True, | ||
| ) | ||
| else: | ||
| self.c_attn = nn.Linear(config.hidden_size, 3 * self.projection_size, bias_attr=True) | ||
| self.c_proj = nn.Linear(config.hidden_size, self.projection_size, bias_attr=not config.no_bias) | ||
| self.o_proj = nn.Linear( |
|
@DesmonDay hi,可以再review下吗～ |
DesmonDay
reviewed
Jun 13, 2024
DesmonDay
reviewed
Jun 13, 2024
| @@ -252,18 +284,26 @@ def forward( | |||
| encoder_hidden_states=None, | |||
| encoder_attention_mask=None, | |||
| output_attentions=False, | |||
| alibi=None, | |||
DesmonDay
reviewed
Jun 13, 2024
| ) | ||
| else: | ||
| attn_output, attn_weight = self._attn(query, key, value, attention_mask) | ||
| context_layer = self._merge_heads(attn_output, self.num_heads, self.head_dim) |
There was a problem hiding this comment.
这一步改到_attn里了，自测记录在Description里添加了
|
从自测记录里看，打开和关闭sequence_parallel，loss精度对不齐，需要对齐精度。 |
DrownFish19
reviewed
Jun 20, 2024
DrownFish19
changed the title
DrownFish19
changed the title
|
LGTM |
DrownFish19
previously approved these changes
Jun 20, 2024
ZHUI
reviewed
Jun 20, 2024
Comment on lines
43
to
45
| tensor_parallel_output=True, | ||
| sequence_parallel=False, | ||
| fuse_sequence_parallel_allreduce=False, |
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
Performance optimization
PR changes
Models
Description
千问模型增加sequence_parallel支持
自测记录：
关闭sequence_parallel：
workerlog_disable_sp.0.log
开启sequence_parallel：
workerlog_sp.0.log