move modeling.py and modeling_nv.py to transformers #9676
Conversation
|
Thanks for your contribution! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #9676 +/- ##
===========================================
+ Coverage 52.00% 52.62% +0.62%
===========================================
Files 721 722 +1
Lines 116703 112813 -3890
===========================================
- Hits 60690 59373 -1317
+ Misses 56013 53440 -2573 ☔ View full report in Codecov by Sentry. |
|
Lint 问题需要安装pre-commit 后格式化代码，参考步骤如下： # 安装
pip install pre-commit
# 在项目文件夹下注册pre-commit，每次commit提交时都会格式化代码
pre-commit install
# 单独处理之前的代码文件
pre-commit run --file XXXX.py |
| @@ -0,0 +1,13 @@ | |||
| # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved. | |||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. |
There was a problem hiding this comment.
此处增加from .modeling import *
| PretrainedModel, | ||
| ) | ||
| from paddlenlp.transformers.model_outputs import ModelOutput | ||
| from paddlenlp.utils.log import logger |
There was a problem hiding this comment.
这里的调用需要修改为相对引用方式
例如：
from paddlenlp.transformers.model_outputs import ModelOutput
修改为
from ..transformers.model_outputs import ModelOutput
There was a problem hiding this comment.
修改为 from ..transformers.model_outputs import ModelOutput 之后会报错，因此暂未修改
There was a problem hiding this comment.
此处应该为 from ..model_outputs import ModelOutput ，抱歉之前写错了
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. |
There was a problem hiding this comment.
同上，此处增加from .modeling import *
| @@ -0,0 +1,517 @@ | |||
| # Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved. | |||
There was a problem hiding this comment.
此文件需要修改名称为modeling.py
| from mteb import MTEB | ||
|
|
||
| from paddlenlp.peft import LoRAConfig, LoRAModel | ||
| from paddlenlp.transformers import AutoModel, AutoModelForCausalLM, AutoTokenizer | ||
| from paddlenlp.transformers.llm_embed.modeling import BiEncoderModel | ||
| from paddlenlp.transformers.nv_embed.modeling_nv import NVEncodeModel |
There was a problem hiding this comment.
此处可以简化导入
from paddlenlp.transformers import BiEncoderModel, NVEncodeModel
|
|
||
| from paddlenlp.peft import LoRAConfig, LoRAModel | ||
| from paddlenlp.trainer import PdArgumentParser, Trainer, get_last_checkpoint, set_seed | ||
| from paddlenlp.transformers import AutoTokenizer | ||
| from paddlenlp.transformers.llm_embed.modeling import BiEncoderModel | ||
| from paddlenlp.transformers.nv_embed.modeling_nv import NVEncodeModel |
已按照您所说步骤在commit之前进行了pre-commit |
| dtype=str(self.latents.weight.dtype).split(".")[-1], | ||
| ) | ||
| self_latents_weight_T = self.latents(one).T | ||
| latents = repeat(self_latents_weight_T, "d h -> b d h", b=last_hidden_states.shape[0]) |
There was a problem hiding this comment.
There was a problem hiding this comment.
已修改为
latents = paddle.tile(self_latents_weight_T, repeat_times=last_hidden_states.shape[0]).reshape( self_latents_weight_T.shape[0], last_hidden_states.shape[0], self_latents_weight_T.shape[1] ) latents = latents.transpose([1, 0, 2])
| k = kv[:, :, : self.config.max_position_embeddings] | ||
| v = kv[:, :, self.config.max_position_embeddings :] | ||
|
|
||
| q, k, v = map(lambda t: rearrange(t, "b n (h d) -> b n h d", h=self.config.num_key_value_heads), (q, k, v)) |
There was a problem hiding this comment.
rearrange辛苦换为paddle算子
| # v.stop_gradient = False | ||
| # out = paddle.nn.functional.scaled_dot_product_attention(q, k, v) # if use this, must set k and v stop_gradient to False | ||
| out = scaled_dot_product_attention(q, k, v) # if use this, no need to manually set k and v | ||
| out = rearrange(out, "b n h d -> b n (h d)", h=self.config.num_key_value_heads) |
move modeling.py and modeling_nv.py to transformers