add vacab parallel embedding #315
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
|
LGTM |
sunpengsdu
reviewed
Sep 6, 2024
internlm/model/modules/embedding.py
Outdated
| assert ( | ||
| num_embeddings % gpc.weight_parallel_size == 0 | ||
| ), f"{num_embeddings} is not divisible by {gpc.weight_parallel_size}" | ||
| self.num_embeddings_per_partition = num_embeddings // gpc.weight_parallel_size |
There was a problem hiding this comment.
这里默认了ISP模式下就采用了vocab_parallel。是不是可以用vocab_parallel作为一个统一的控制，它只和是否用embedding和head共享权重相关，如果用户在Modeling文件里，需要共享权重，则手动设定vocab_parallel为true即可。其他情况下，默认走之前的切分emb的逻辑。避免之前的代码出现BC，特别是有一些llama模型加载HF权重的设计，都是走的切分emb维度
sunpengsdu
reviewed
Sep 6, 2024
| def forward(self, input_: Tensor) -> Tensor: | ||
| if self.vocab_parallel: | ||
| # Build the mask. | ||
| input_mask = (input_ < self.vocab_start_index) | (input_ >= self.vocab_end_index) |
There was a problem hiding this comment.
前面is_using_isp切分的是self.num_embeddings_per_partition = num_embeddings // gpc.weight_parallel_size，但是并没有构建vocab_start_index等
There was a problem hiding this comment.
isp是聚合参数，所以不能走vocab_parallel的代码，需要走原来的逻辑
sunpengsdu
reviewed
Sep 6, 2024
internlm/core/parallel/comm/isp.py
Outdated
| @@ -145,7 +145,7 @@ class EmbeddingWeightParallelCommunicator: | |||
|
|
|||
| def __init__(self, parallel_mode: ParallelMode) -> None: | |||
| self.parallel_mode = parallel_mode | |||
| self.emb_column = 1 | |||
| self.vocab_dim = 0 | |||
There was a problem hiding this comment.
这里麻烦的一点是要是ISP模型保持原来默认切分维度，又有了vocab切的可选项，要怎么办
There was a problem hiding this comment.
可能是register_module_hook的时候，判断一下
mwiacx
force-pushed
the
feat/add-vocab-parallel-embedding
branch
from
0b67f09
to
35c53cd
Compare
sunpengsdu
approved these changes
Sep 10, 2024
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.
add vacab parallel embedding