optimize_wrong_information by zty-king · Pull Request #76813 · PaddlePaddle/Paddle
你的PR提交成功,感谢你对开源项目的贡献!
请关注后续CI自动化测试结果,详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
希望可以在解析AOA失败的时候把用户提供的AOA给暴露出来。一个实现思路是,把最原始的(用户输入的aoa statements)作为一个attribute,挂在Macro处理后的aoa statements上,如果报错可以找这个属性,把他打印出来。
| end_idx = idx + 3 | ||
| if end_idx < len(tokens) - 1: | ||
| assert tokens[end_idx].value == "," | ||
| assert tokens[end_idx].value == ",", ( |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这个可不可以直接把aoa stmt打印出来
| and tokens[idx + 3].type == TokenType.NUMBER | ||
| and tokens[idx + 4].type == TokenType.RBRACKET | ||
| ), ( | ||
| f"The array macro format is incorrect which is must be like: NAME[START:END], but now the format is {tokens[idx].value}{tokens[idx + 1].value}:{tokens[idx + 3].value}{tokens[idx + 4].value}." |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
同上
| ) | ||
| assert num_key_value_groups and num_key_value_groups > 0, ( | ||
| "num_key_value_groups must be positive." | ||
| "num_key_value_groups must be positive.(got: {num_key_value_groups})." |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
| "num_key_value_groups must be positive.(got: {num_key_value_groups})." | |
| f"num_key_value_groups must be positive.(got: {num_key_value_groups})." |
| f"num_key_value_groups must be positive (got: {num_key_value_groups})" | ||
| ) | ||
| assert fused_qkv_pos is not None, "No fused_qkv tag found in expression." | ||
| assert fused_qkv_pos is not None, ( |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这个也希望把aoa stmt直接打出来
| f"Missing keys:{missing_keys}, check whether the checkpoint is complete." | ||
| ) | ||
|
|
||
| unexpected_keys = set(tensor_key_list) - set(state_dict_param_names) |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
建议 每个key单独一行 前后加一个醒目的界符
| ) | ||
|
|
||
| unexpected_keys = set(tensor_key_list) - set(state_dict_param_names) | ||
| if len(unexpected_keys) > 0: |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
同上
| None, | ||
| ) | ||
| assert name_with_layer_id_offset, "No $LAYER_ID_OFFSET found in NAME tokens" | ||
| assert name_with_layer_id_offset, ( |
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LAYER_ID,EXPERT_ID等匹配的Macro,如果匹配到了空的结果,往往是不符合预期的,这时希望给一个警告,同时把最原始的AOA statement打印出来,ID什么也没匹配到,可能不符合原始预期。
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
zty-king
deleted the
optimize_wrong_information
branch
This file contains hidden or 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
