AmerAlreyahi commited on
Commit
22cdea4
Β·
verified Β·
1 Parent(s): eee9f5e

OptimaMT organization card

Browse files
Files changed (2) hide show
  1. README.md +2 -2
  2. index.html +3 -3
README.md CHANGED
@@ -12,10 +12,10 @@ pinned: false
12
 
13
  **ONNX Runtime's fast mode at half the memory: 2.3–2.5Γ— faster than the default, same translations.**
14
 
15
- 49 OPUS-MT translation models for ONNX Runtime and transformers.js: drop-in copies of the int8 exports
16
  with a lighter merged decoder. Same weights, same tokens.
17
 
18
- | Memory after load (median) | Linux | Windows |
19
  |---|---:|---:|
20
  | `disable_quant_qdq = 1` | 525 β†’ 281 MiB (-46 %) | 500 β†’ 257 MiB (-49 %) |
21
  | ONNX Runtime defaults | 277 β†’ 217 MiB (-22 %) | 196 β†’ 166 MiB (-16 %) |
 
12
 
13
  **ONNX Runtime's fast mode at half the memory: 2.3–2.5Γ— faster than the default, same translations.**
14
 
15
+ 51 OPUS-MT translation models for ONNX Runtime and transformers.js: drop-in copies of the int8 exports
16
  with a lighter merged decoder. Same weights, same tokens.
17
 
18
+ | Memory after load (median; Linux 51 pairs, Windows 49) | Linux | Windows |
19
  |---|---:|---:|
20
  | `disable_quant_qdq = 1` | 525 β†’ 281 MiB (-46 %) | 500 β†’ 257 MiB (-49 %) |
21
  | ONNX Runtime defaults | 277 β†’ 217 MiB (-22 %) | 196 β†’ 166 MiB (-16 %) |
index.html CHANGED
@@ -20,10 +20,10 @@
20
  <body>
21
  <main class="omt">
22
  <p><strong>Lighter OPUS-MT translation models that run locally</strong>, in ONNX Runtime or in the browser with
23
- transformers.js. 49 translation directions, the same translations, and about half the memory in fast mode.</p>
24
 
25
  <p class="links">
26
- <a href="https://huggingface.co/collections/OptimaMT/optimamt-lighter-opus-mt-onnx-models-6af228dc9cd91f015e56e869" target="_blank" rel="noopener">All 49 models</a>
27
  <a href="https://huggingface.co/spaces/OptimaMT/results" target="_blank" rel="noopener">Interactive results and live demo</a>
28
  <a href="https://github.com/AmerAlreyahi/OptimaMT" target="_blank" rel="noopener">Method and code</a>
29
  <a href="https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md" target="_blank" rel="noopener">Technical report</a>
@@ -32,7 +32,7 @@ transformers.js. 49 translation directions, the same translations, and about hal
32
  <img src="https://huggingface.co/spaces/OptimaMT/README/resolve/main/tradeoff.png" alt="Memory against speed, original and OptimaMT, on Linux and Windows">
33
 
34
  <table>
35
- <tr><th>Memory after load (median, 49 pairs)</th><th>Linux</th><th>Windows</th></tr>
36
  <tr><td>Fast mode (<code>disable_quant_qdq = 1</code>)</td><td>525 β†’ 281 MiB (-46 %)</td><td>500 β†’ 257 MiB (-49 %)</td></tr>
37
  <tr><td>ONNX Runtime defaults</td><td>277 β†’ 217 MiB (-22 %)</td><td>196 β†’ 166 MiB (-16 %)</td></tr>
38
  </table>
 
20
  <body>
21
  <main class="omt">
22
  <p><strong>Lighter OPUS-MT translation models that run locally</strong>, in ONNX Runtime or in the browser with
23
+ transformers.js. 51 translation directions, the same translations, and about half the memory in fast mode.</p>
24
 
25
  <p class="links">
26
+ <a href="https://huggingface.co/collections/OptimaMT/optimamt-lighter-opus-mt-onnx-models-6af228dc9cd91f015e56e869" target="_blank" rel="noopener">All 51 models</a>
27
  <a href="https://huggingface.co/spaces/OptimaMT/results" target="_blank" rel="noopener">Interactive results and live demo</a>
28
  <a href="https://github.com/AmerAlreyahi/OptimaMT" target="_blank" rel="noopener">Method and code</a>
29
  <a href="https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md" target="_blank" rel="noopener">Technical report</a>
 
32
  <img src="https://huggingface.co/spaces/OptimaMT/README/resolve/main/tradeoff.png" alt="Memory against speed, original and OptimaMT, on Linux and Windows">
33
 
34
  <table>
35
+ <tr><th>Memory after load (median; Linux 51 pairs, Windows 49)</th><th>Linux</th><th>Windows</th></tr>
36
  <tr><td>Fast mode (<code>disable_quant_qdq = 1</code>)</td><td>525 β†’ 281 MiB (-46 %)</td><td>500 β†’ 257 MiB (-49 %)</td></tr>
37
  <tr><td>ONNX Runtime defaults</td><td>277 β†’ 217 MiB (-22 %)</td><td>196 β†’ 166 MiB (-16 %)</td></tr>
38
  </table>