معرفی شرکت ها


fpgaconvnet-model-0.1.4.2


Card image cap
تبلیغات ما

مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.

مشاهده بیشتر
Card image cap
تبلیغات ما

مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.

مشاهده بیشتر
Card image cap
تبلیغات ما

مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.

مشاهده بیشتر
Card image cap
تبلیغات ما

مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.

مشاهده بیشتر
Card image cap
تبلیغات ما

مشتریان به طور فزاینده ای آنلاین هستند. تبلیغات می تواند به آنها کمک کند تا کسب و کار شما را پیدا کنند.

مشاهده بیشتر

توضیحات

Parser and model for Convolutional Neural Network Streaming-Based Accelerator on FPGA devices.
ویژگی مقدار
سیستم عامل -
نام فایل fpgaconvnet-model-0.1.4.2
نام fpgaconvnet-model
نسخه کتابخانه 0.1.4.2
نگهدارنده []
ایمیل نگهدارنده []
نویسنده Alex Montgomerie
ایمیل نویسنده am9215@ic.ac.uk
آدرس صفحه اصلی https://github.com/AlexMontgomerie/fpgaconvnet-model
آدرس اینترنتی https://pypi.org/project/fpgaconvnet-model/
مجوز -
# fpgaConvNet Model This repo contains performance and resource for the building blocks of fpgaConvNet, a Streaming Architecture-based Convolutional Neural Network (CNN) acceleration toolflow, which maps CNN models to FPGAs. The building blocks are implemented in hardware in the [fpgaconvnet-hls](https://github.com/AlexMontgomerie/fpgaconvnet-hls) repository. These models are used in conjunction with [samo](https://github.com/AlexMontgomerie/samo), a Streaming Architecture optimiser, where there are instructions for performing optimisation. ## Setup The following programs are required: - `python (>=3.7)` To install this package, run from this directory the following: ``` python -m pip install fpgaconvnet-model ``` ## Usage This repo can be used to get performance and resource estimates for different hardware configurations. To start, the desired network will need to be parsed into fpgaConvNet's representation. Then a hardware configuration can be loaded, and performance and resource predictions obtained. ```python from fpgaconvnet.models.network import Network # initialise network, and load a configuration net = Network("model-name", "model.onnx") net.load_network("model-config.json") # print performance and resource estimates print(f"predicted latency (us): {net.get_latency()*1000000}") print(f"predicted throughput (img/s): {net.get_throughput()} (batch size={net.batch_size})") print(f"predicted resource usage: {net.partitions[0].get_resource_usage()}") # visualise the network configuration net.visualise("image-path.png", mode="png") # export out the configuration net.save_all_partitions("config-path.json") ``` ## Modelling In order to do the CNN to hardware mapping, a model of the hardware is needed. There are four levels of abstraction for the final hardware: modules, layers, partitions and network. At each level of abstraction, there is an associated performance and resource estimate so that the constraints for the optimiser can be obtained. - __Module:__ These are the basic building blocks of the accelerator. The modules are the following: - Accum - BatchNorm - Conv - Glue - SlidingWindow - Fork - Pool - Squeeze - __Layer:__ Layers are comprised of modules. They have the same functionality of the equivalent layers of the CNN model. The following layers are supported: - Batch Normalization - Convolution - Inner Product - Pooling - ReLU - __Partition:__ Partitions make up a sub-graph of the CNN model network. They are comprised of layers. A single partition fits on an FPGA at a time, and partitions are changed by reconfiguring the FPGA. - __Network:__ This is the entire CNN model described through hardware. A network contains partitions and information on how to execute them. --- Feel free to post an issue if you have any questions or problems!


نیازمندی

مقدار نام
>=2.5 networkx
>=1.19.2 numpy
<=3.17.0,>=3.13.0 protobuf
>=1.7.1 torch
>=5.1.0 pyyaml
>=1.2.1 scipy
>=0.8.2 torchvision
==1.8.1 onnx
==1.7.0 onnxruntime
>=0.16 graphviz
>=1.4.2 pydot
==0.2.6 onnxoptimizer
>=1.4.2 ddt
- sklearn
- matplotlib
==5.5 coverage
<3 pyparsing


زبان مورد نیاز

مقدار نام
>=3.7 Python


نحوه نصب


نصب پکیج whl fpgaconvnet-model-0.1.4.2:

    pip install fpgaconvnet-model-0.1.4.2.whl


نصب پکیج tar.gz fpgaconvnet-model-0.1.4.2:

    pip install fpgaconvnet-model-0.1.4.2.tar.gz