padding the length of input for vit_attention #45506
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
b3602sss
reviewed
Aug 29, 2022
| @@ -310,6 +349,11 @@ int QkvToContextPluginDynamic::enqueue( | |||
| // input[0], (B, S, 3 * N * H, 1, 1) | |||
| int batch = input_dims.d[0]; | |||
| int seq_len = input_dims.d[1]; | |||
| int real_seq_len = seq_len; | |||
| if (input_desc[0].type == nvinfer1::DataType::kHALF) { | |||
There was a problem hiding this comment.
注释下fp16需要pading
好的
| __global__ void reset_qk_bias(T *input, int real_seq_len, int seq_len) { | ||
| if (threadIdx.x < seq_len) { | ||
| int id = threadIdx.x + blockIdx.x * seq_len; | ||
| input[id] = threadIdx.x >= real_seq_len ? (T)-1e20f : (T)0.0f; |
| if (ProductDim(input_desc[1].dims) == ProductDim(input_desc[0].dims)) { | ||
| qk_bias = reinterpret_cast<float *>(workspace); | ||
| auto size = batch * head_number_ * seq_len * seq_len; | ||
| cudaMemset(qk_bias, 0, sizeof(float) * size); |
| @@ -373,6 +423,35 @@ int QkvToContextPluginDynamic::enqueue( | |||
| } else if (input_type == nvinfer1::DataType::kHALF) { | |||
| #ifdef TRT_PLUGIN_FP16_AVALIABLE | |||
| VLOG(1) << "TRT Plugin DataType selected. QkvToContext-->fp16"; | |||
| int *padding_offset = nullptr; | |||
| half *padding_input = nullptr; | |||
| framework::Tensor padding_offset_tensor; | |||
| 0, | ||
| sizeof(half) * batch * seq_len * 3 * head_number_ * head_size_); | ||
|
|
||
| set_padding_offset<<<1, 1, 0, stream>>>( |
| @@ -1105,6 +1113,9 @@ def generate_trt_nodes_num(): | |||
| self.trt_param.precision = paddle_infer.PrecisionType.Half | |||
| yield self.create_inference_config(), generate_trt_nodes_num(), (1e-3, | |||
| 1e-3) | |||
| self.trt_param.precision = paddle_infer.PrecisionType.Float32 | |||
| yield self.create_inference_config(), generate_trt_nodes_num(), (1e-3, | |||
| 1e-3) | |||
b3602sss
reviewed
Sep 2, 2022
| @@ -342,6 +427,12 @@ int QkvToContextPluginDynamic::enqueue( | |||
| head_number_); | |||
| qk_bias = temp_qk_bias; | |||
| } | |||
| // fake qk_bias | |||
| if (ProductDim(input_desc[1].dims) == ProductDim(input_desc[0].dims)) { | |||
There was a problem hiding this comment.
config的时候就可以确定不用每次enque判断.下面几个也一样
There was a problem hiding this comment.
config的时候就可以确定不用每次enque判断.下面几个也一样
好的，和下面memset的统一放到configure处理
| if (ProductDim(input_desc[1].dims) == ProductDim(input_desc[0].dims)) { | ||
| qk_bias = reinterpret_cast<float *>(workspace); | ||
| auto size = batch * head_number_ * seq_len * seq_len; | ||
| cudaMemset(qk_bias, 0, sizeof(float) * size); |
fengxiaoshuai
changed the title
b3602sss
approved these changes
Sep 2, 2022
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
Others
PR changes
Others
Describe
当attention的输入length不是8的整数倍时，fp16的性能很差，这里对multihead plugin的输入进行padding，对于vit_384模型，batch=1时，时间由13.5ms降低到10.5ms