QONNX: Representing Arbitrary-Precision Quantized Neural Networks
We present extensions to the Open Neural Network Exchange (ONNX) intermediate representation format to represent arbitrary-precision quantized neural networks. We first introduce support for low precision quantization in existing ONNX-based quantization formats by leveraging integer clipping, result...
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , |
|---|---|
| Format: | Artikel |
| Sprache: | eng |
| Schlagworte: | |
| Online-Zugang: | Volltext bestellen |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| container_end_page | |
|---|---|
| container_issue | |
| container_start_page | |
| container_title | |
| container_volume | |
| creator | Pappalardo, Alessandro Umuroglu, Yaman Blott, Michaela Mitrevski, Jovan Hawks, Ben Tran, Nhan Loncar, Vladimir Summers, Sioni Borras, Hendrik Muhizi, Jules Trahms, Matthew Hsu, Shih-Chieh Hauck, Scott Duarte, Javier |
| description | We present extensions to the Open Neural Network Exchange (ONNX) intermediate
representation format to represent arbitrary-precision quantized neural
networks. We first introduce support for low precision quantization in existing
ONNX-based quantization formats by leveraging integer clipping, resulting in
two new backward-compatible variants: the quantized operator format with
clipping and quantize-clip-dequantize (QCDQ) format. We then introduce a novel
higher-level ONNX format called quantized ONNX (QONNX) that introduces three
new operators -- Quant, BipolarQuant, and Trunc -- in order to represent
uniform quantization. By keeping the QONNX IR high-level and flexible, we
enable targeting a wider variety of platforms. We also present utilities for
working with QONNX, as well as examples of its usage in the FINN and hls4ml
toolchains. Finally, we introduce the QONNX model zoo to share low-precision
quantized neural networks. |
| doi_str_mv | 10.48550/arxiv.2206.07527 |
| format | Article |
| fullrecord | <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2206_07527</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2206_07527</sourcerecordid><originalsourceid>FETCH-LOGICAL-a677-a78e6b2c94c8bcb185e6d2852003abae9d0bd585b8ffc81bfacd56f22de5bc543</originalsourceid><addsrcrecordid>eNotj8tOwzAQRb1hgQofwIr8QILjZGyXXVTxkqqUoi7YRWN7jCxKWk1SXl9PKKzO4kpX5whxUcqitgDyCvkzvRdKSV1IA8qcima9atvn6-yJ9kwD9WPqX7KGXRoZ-St_ZPJpSLs-Wx9wGr8pZC0dGLcTxo8dvw5n4iTidqDzf87E5vZms7jPl6u7h0WzzFEbk6OxpJ3y89pb511pgXRQFpSUFTqkeZAugAVnY_S2dBF9AB2VCgTOQ13NxOXf7bGh23N6mwS735bu2FL9AK4TRYA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>QONNX: Representing Arbitrary-Precision Quantized Neural Networks</title><source>arXiv.org</source><creator>Pappalardo, Alessandro ; Umuroglu, Yaman ; Blott, Michaela ; Mitrevski, Jovan ; Hawks, Ben ; Tran, Nhan ; Loncar, Vladimir ; Summers, Sioni ; Borras, Hendrik ; Muhizi, Jules ; Trahms, Matthew ; Hsu, Shih-Chieh ; Hauck, Scott ; Duarte, Javier</creator><creatorcontrib>Pappalardo, Alessandro ; Umuroglu, Yaman ; Blott, Michaela ; Mitrevski, Jovan ; Hawks, Ben ; Tran, Nhan ; Loncar, Vladimir ; Summers, Sioni ; Borras, Hendrik ; Muhizi, Jules ; Trahms, Matthew ; Hsu, Shih-Chieh ; Hauck, Scott ; Duarte, Javier</creatorcontrib><description>We present extensions to the Open Neural Network Exchange (ONNX) intermediate
representation format to represent arbitrary-precision quantized neural
networks. We first introduce support for low precision quantization in existing
ONNX-based quantization formats by leveraging integer clipping, resulting in
two new backward-compatible variants: the quantized operator format with
clipping and quantize-clip-dequantize (QCDQ) format. We then introduce a novel
higher-level ONNX format called quantized ONNX (QONNX) that introduces three
new operators -- Quant, BipolarQuant, and Trunc -- in order to represent
uniform quantization. By keeping the QONNX IR high-level and flexible, we
enable targeting a wider variety of platforms. We also present utilities for
working with QONNX, as well as examples of its usage in the FINN and hls4ml
toolchains. Finally, we introduce the QONNX model zoo to share low-precision
quantized neural networks.</description><identifier>DOI: 10.48550/arxiv.2206.07527</identifier><language>eng</language><subject>Computer Science - Hardware Architecture ; Computer Science - Learning ; Computer Science - Programming Languages ; Statistics - Machine Learning</subject><creationdate>2022-06</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2206.07527$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2206.07527$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Pappalardo, Alessandro</creatorcontrib><creatorcontrib>Umuroglu, Yaman</creatorcontrib><creatorcontrib>Blott, Michaela</creatorcontrib><creatorcontrib>Mitrevski, Jovan</creatorcontrib><creatorcontrib>Hawks, Ben</creatorcontrib><creatorcontrib>Tran, Nhan</creatorcontrib><creatorcontrib>Loncar, Vladimir</creatorcontrib><creatorcontrib>Summers, Sioni</creatorcontrib><creatorcontrib>Borras, Hendrik</creatorcontrib><creatorcontrib>Muhizi, Jules</creatorcontrib><creatorcontrib>Trahms, Matthew</creatorcontrib><creatorcontrib>Hsu, Shih-Chieh</creatorcontrib><creatorcontrib>Hauck, Scott</creatorcontrib><creatorcontrib>Duarte, Javier</creatorcontrib><title>QONNX: Representing Arbitrary-Precision Quantized Neural Networks</title><description>We present extensions to the Open Neural Network Exchange (ONNX) intermediate
representation format to represent arbitrary-precision quantized neural
networks. We first introduce support for low precision quantization in existing
ONNX-based quantization formats by leveraging integer clipping, resulting in
two new backward-compatible variants: the quantized operator format with
clipping and quantize-clip-dequantize (QCDQ) format. We then introduce a novel
higher-level ONNX format called quantized ONNX (QONNX) that introduces three
new operators -- Quant, BipolarQuant, and Trunc -- in order to represent
uniform quantization. By keeping the QONNX IR high-level and flexible, we
enable targeting a wider variety of platforms. We also present utilities for
working with QONNX, as well as examples of its usage in the FINN and hls4ml
toolchains. Finally, we introduce the QONNX model zoo to share low-precision
quantized neural networks.</description><subject>Computer Science - Hardware Architecture</subject><subject>Computer Science - Learning</subject><subject>Computer Science - Programming Languages</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2022</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8tOwzAQRb1hgQofwIr8QILjZGyXXVTxkqqUoi7YRWN7jCxKWk1SXl9PKKzO4kpX5whxUcqitgDyCvkzvRdKSV1IA8qcima9atvn6-yJ9kwD9WPqX7KGXRoZ-St_ZPJpSLs-Wx9wGr8pZC0dGLcTxo8dvw5n4iTidqDzf87E5vZms7jPl6u7h0WzzFEbk6OxpJ3y89pb511pgXRQFpSUFTqkeZAugAVnY_S2dBF9AB2VCgTOQ13NxOXf7bGh23N6mwS735bu2FL9AK4TRYA</recordid><startdate>20220615</startdate><enddate>20220615</enddate><creator>Pappalardo, Alessandro</creator><creator>Umuroglu, Yaman</creator><creator>Blott, Michaela</creator><creator>Mitrevski, Jovan</creator><creator>Hawks, Ben</creator><creator>Tran, Nhan</creator><creator>Loncar, Vladimir</creator><creator>Summers, Sioni</creator><creator>Borras, Hendrik</creator><creator>Muhizi, Jules</creator><creator>Trahms, Matthew</creator><creator>Hsu, Shih-Chieh</creator><creator>Hauck, Scott</creator><creator>Duarte, Javier</creator><scope>AKY</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20220615</creationdate><title>QONNX: Representing Arbitrary-Precision Quantized Neural Networks</title><author>Pappalardo, Alessandro ; Umuroglu, Yaman ; Blott, Michaela ; Mitrevski, Jovan ; Hawks, Ben ; Tran, Nhan ; Loncar, Vladimir ; Summers, Sioni ; Borras, Hendrik ; Muhizi, Jules ; Trahms, Matthew ; Hsu, Shih-Chieh ; Hauck, Scott ; Duarte, Javier</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a677-a78e6b2c94c8bcb185e6d2852003abae9d0bd585b8ffc81bfacd56f22de5bc543</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2022</creationdate><topic>Computer Science - Hardware Architecture</topic><topic>Computer Science - Learning</topic><topic>Computer Science - Programming Languages</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Pappalardo, Alessandro</creatorcontrib><creatorcontrib>Umuroglu, Yaman</creatorcontrib><creatorcontrib>Blott, Michaela</creatorcontrib><creatorcontrib>Mitrevski, Jovan</creatorcontrib><creatorcontrib>Hawks, Ben</creatorcontrib><creatorcontrib>Tran, Nhan</creatorcontrib><creatorcontrib>Loncar, Vladimir</creatorcontrib><creatorcontrib>Summers, Sioni</creatorcontrib><creatorcontrib>Borras, Hendrik</creatorcontrib><creatorcontrib>Muhizi, Jules</creatorcontrib><creatorcontrib>Trahms, Matthew</creatorcontrib><creatorcontrib>Hsu, Shih-Chieh</creatorcontrib><creatorcontrib>Hauck, Scott</creatorcontrib><creatorcontrib>Duarte, Javier</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Statistics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Pappalardo, Alessandro</au><au>Umuroglu, Yaman</au><au>Blott, Michaela</au><au>Mitrevski, Jovan</au><au>Hawks, Ben</au><au>Tran, Nhan</au><au>Loncar, Vladimir</au><au>Summers, Sioni</au><au>Borras, Hendrik</au><au>Muhizi, Jules</au><au>Trahms, Matthew</au><au>Hsu, Shih-Chieh</au><au>Hauck, Scott</au><au>Duarte, Javier</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>QONNX: Representing Arbitrary-Precision Quantized Neural Networks</atitle><date>2022-06-15</date><risdate>2022</risdate><abstract>We present extensions to the Open Neural Network Exchange (ONNX) intermediate
representation format to represent arbitrary-precision quantized neural
networks. We first introduce support for low precision quantization in existing
ONNX-based quantization formats by leveraging integer clipping, resulting in
two new backward-compatible variants: the quantized operator format with
clipping and quantize-clip-dequantize (QCDQ) format. We then introduce a novel
higher-level ONNX format called quantized ONNX (QONNX) that introduces three
new operators -- Quant, BipolarQuant, and Trunc -- in order to represent
uniform quantization. By keeping the QONNX IR high-level and flexible, we
enable targeting a wider variety of platforms. We also present utilities for
working with QONNX, as well as examples of its usage in the FINN and hls4ml
toolchains. Finally, we introduce the QONNX model zoo to share low-precision
quantized neural networks.</abstract><doi>10.48550/arxiv.2206.07527</doi><oa>free_for_read</oa></addata></record> |
| fulltext | fulltext_linktorsrc |
| identifier | DOI: 10.48550/arxiv.2206.07527 |
| ispartof | |
| issn | |
| language | eng |
| recordid | cdi_arxiv_primary_2206_07527 |
| source | arXiv.org |
| subjects | Computer Science - Hardware Architecture Computer Science - Learning Computer Science - Programming Languages Statistics - Machine Learning |
| title | QONNX: Representing Arbitrary-Precision Quantized Neural Networks |
| url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-29T09%3A55%3A22IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=QONNX:%20Representing%20Arbitrary-Precision%20Quantized%20Neural%20Networks&rft.au=Pappalardo,%20Alessandro&rft.date=2022-06-15&rft_id=info:doi/10.48550/arxiv.2206.07527&rft_dat=%3Carxiv_GOX%3E2206_07527%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true |