8000 Tags · OpenPPL/ppq · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Tags: OpenPPL/ppq

Tags

v0.6.6

Toggle v0.6.6's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
Fix Executing Bugs (#408)

* Fix Executing Bugs

* 修复了 unsqueeze 算子在多于一个轴时的顺序错误问题
* 修复了 softmax 算子在 opset 11 时默认轴错误的问题
* 修复了 图拷贝 过程中可能因为空值而出现的错误

v0.6.5

Toggle v0.6.5's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
onnx 模型导出与算法更新 (#231)

* 添加样例文件

* onnx模型导出与算法更新

* 为dump_torch_to_onnx函数添加了默认参数
* EXPORT_OVERLAPPED_CONFIG 现在是过时参数,你将使用TQC上的QuantizationVisiblity属性来进行导出控制。该属性有三个可选项:强制导出、TQC激活时导出、不导出。
* 修改了 exporter 逻辑以适配新的QuantizationVisiblity属性
* 修改了onnx qdq的导出逻辑,现在将尽可能消除对称量化中的激活函数。
* 修改了 graphwise analyser 的逻辑,现在允许分析多输出算子的误差
* 修改了 layerwise equalization 的逻辑,现在允许 include act,支持conv1d, conv2d conv3d, convtranpose1d, convtranspose2d, convtranspose3d, gemm, matmul
* 修复了 passive parameter pass 中的 pad 量化错误
* 修复了 quant alignment pass 中 pooling 算子的对齐错误
* 修复了 核心量化函数在启动 cuda kernel 的情况下无法处理 cpu tensor 的问题
* 修改 openvino 量化策略,负数部分现在可以取到-128(曾经是-127)
* 给 dsp quantizer 添加了一个新的量化类型

* 添加测试样例

* 修复ci错误

v0.6.4

Toggle v0.6.4's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
feat(default.py): support ViT (#145)

* feat(default.py): add support ViT opr

* docs(torch.py): remove useless

v0.6.3

Toggle v0.6.3's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
fix bugs: fixed calib_steps (#42)

Fixed quantized function not being able to modify Calib Steps

v0.6.2

Toggle v0.6.2's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
Update ssd.py (#33)

Ensure the variables are on the same device.

v0.6.1

Toggle v0.6.1's commit message
Update to 0.6.1

Support ONNXRuntime Quant&Dequant Format.

v0.6.0

Toggle v0.6.0's commit message
Core update to 0.6.0

Add Subgraph Optimization Algorithm
Add Matrix Factorization Algorithm
Add 4&8 bit Mix Precision Dispatching
Refine Network Analysis
Add ProgramEntrance.py (as an usage example)
Other bug fixs.

v0.5.3

Toggle v0.5.3's commit message
Update to version 0.5.3

Update to 0.5.3
Channel Split Algorithm Added
MSE Calibration method Added
bn merge bugs fixed
cuda.histogram bugs fixed
fix other typos.
README updated.

v0.5.2

Toggle v0.5.2's commit message
initial import

0