[New features] add roberta & gpt conversion #4407
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 Report
@@ Coverage Diff @@
## develop #4407 +/- ##
===========================================
+ Coverage 39.62% 40.10% +0.47%
===========================================
Files 433 439 +6
Lines 60982 61568 +586
===========================================
+ Hits 24165 24689 +524
- Misses 36817 36879 +62
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
sijunhe
reviewed
Jan 11, 2023
Comment on lines
+589
to
+594
| if name_mapping.target_name in paddle_state_dict: | ||
| paddle_numpy = paddle_state_dict.pop(name_mapping.target_name) | ||
| model_state_saver.add(name_mapping.target_name, "paddle", paddle_numpy) | ||
| model_state_saver.add(name_mapping.target_name, "paddle-shape", str(paddle_numpy.shape)) | ||
|
|
||
| pytorch_numpy = pytorch_state_dict.pop(name_mapping.source_name) | ||
| model_state_saver.add(name_mapping.target_name, "pytorch", pytorch_numpy) | ||
| model_state_saver.add(name_mapping.target_name, "pytorch-shape", str(pytorch_numpy.shape)) | ||
| if name_mapping.source_name in pytorch_state_dict: |
There was a problem hiding this comment.
这两个if 分开会不会有不对称的情况，即一个true 一个false. 可以考虑合成一个 if name_mapping.source_name in pytorch_state_dict and name_mapping.target_name in paddle_state_dict:
There was a problem hiding this comment.
这个确实是会存在：比如pytorch 那边把 attention 的 qkv 存在一次，此时只有一个weight，可是 paddle 这边还是会存成三个 tensor。
这段代码的作用将会变成：pytorch 会 hook attention（qkv 的上层layer）的 forward，而 paddle 会分别 hook attention 中 q、k、v 的 forward。
这也是在 gpt 中测试出来的，为了做hooker 兼容。虽然不能从 layer-name 层面一一对应，可是也是能够给出一些logit 信息，提供判断。
There was a problem hiding this comment.
在 LogitComparer 中对应 Attention的输出是:
+--------------------------------------------+---------------------------------------+------------------------------------------------+---------------------------------------+------------------------------------------------+
| decoder.layers.0.self_attn.q_proj.weight | [ 0.50337297 -0.7915838 0.5165813 ] | [0.03290959 0.0143059 0.11999971] | [ 0.50337297 -0.7915838 0.5165813 ] | [0.03290959 0.01430589 0.11999971] |
+--------------------------------------------+---------------------------------------+------------------------------------------------+---------------------------------------+------------------------------------------------+
| decoder.layers.0.self_attn.k_proj.weight | [ 0.50337297 -0.7915838 0.5165813 ] | [-0.10358325 0.05546366 0.13997011] | | |
+--------------------------------------------+---------------------------------------+------------------------------------------------+---------------------------------------+------------------------------------------------+
| decoder.layers.0.self_attn.v_proj.weight | [ 0.50337297 -0.7915838 0.5165813 ] | [ 0.24479835 0.00965955 -0.06201734] | | |
+--------------------------------------------+---------------------------------------+------------------------------------------------+---------------------------------------+------------------------------------------------+
wj-Mcat
force-pushed
the
add-conversion-0
branch
from
d16d413 to
ac28583
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
添加 roberta 和 gpt 的在线转化带阿米