wehub/labmlai--annotated_deep_learning_paper_implementations
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity

chore: import upstream snapshot with attribution

Browse Source
This commit is contained in:
wehub-resource-sync
2026-07-13 12:19:01 +08:00
commit 3b90d1192f
2172 changed files with 594509 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
translate_cache/neox/samples/__init__.ja.json
+5
View File
@@ -0,0 +1,5 @@
{
"<h1>Samples</h1>\n<ul><li><a href=\"generate.html\">Generating text</a> </li>\n<li><a href=\"finetune.html\">Fine tuning the biases with pipeline-parallel training</a></li></ul>\n": "<h1>[\u30b5\u30f3\u30d7\u30eb]</h1>\n<ul><li><a href=\"generate.html\">\u30c6\u30ad\u30b9\u30c8\u306e\u751f\u6210</a></li>\n<li><a href=\"finetune.html\">\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u4e26\u5217\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306b\u3088\u308b\u30d0\u30a4\u30a2\u30b9\u306e\u5fae\u8abf\u6574</a></li></ul>\n",
"Samples": "[\u30b5\u30f3\u30d7\u30eb]",
"Samples for inference and fine-tuning": "\u63a8\u8ad6\u3068\u5fae\u8abf\u6574\u7528\u306e\u30b5\u30f3\u30d7\u30eb"
}
translate_cache/neox/samples/__init__.si.json
+5
View File
@@ -0,0 +1,5 @@
{
"<h1>Samples</h1>\n<ul><li><a href=\"generate.html\">Generating text</a> </li>\n<li><a href=\"finetune.html\">Fine tuning the biases with pipeline-parallel training</a></li></ul>\n": "<h1>\u0dc3\u0dcf\u0db8\u0dca\u0db4\u0dbd</h1>\n<ul><li><a href=\"generate.html\">\u0db4\u0dd9\u0dc5 \u0d8b\u0dad\u0dca\u0db4\u0dcf\u0daf\u0db1\u0dba</a> </li>\n<li><a href=\"finetune.html\">\u0db1\u0dbd \u0db8\u0dcf\u0dbb\u0dca\u0d9c-\u0dc3\u0db8\u0dcf\u0db1\u0dca\u0dad\u0dbb \u0db4\u0dd4\u0dc4\u0dd4\u0dab\u0dd4\u0dc0 \u0dc3\u0db8\u0d9f \u0d85\u0d9c\u0dad\u0dd3\u0db1\u0dca \u0dc4\u0ddc\u0db3\u0dd2\u0db1\u0dca \u0dc3\u0dd4\u0dc3\u0dbb \u0d9a\u0dd2\u0dbb\u0dd3\u0db8</a></li></ul>\n",
"Samples": "\u0dc3\u0dcf\u0db8\u0dca\u0db4\u0dbd",
"Samples for inference and fine-tuning": "\u0d85\u0db1\u0dd4\u0db8\u0dcf\u0db1\u0dba \u0dc3\u0dc4 \u0db8\u0db1\u0dcf\u0dc0 \u0dc3\u0dd4\u0dc3\u0dbb \u0d9a\u0dd2\u0dbb\u0dd3\u0db8 \u0dc3\u0db3\u0dc4\u0dcf \u0dc3\u0dcf\u0db8\u0dca\u0db4\u0dbd"
}
translate_cache/neox/samples/__init__.zh.json
+5
View File
@@ -0,0 +1,5 @@
{
"<h1>Samples</h1>\n<ul><li><a href=\"generate.html\">Generating text</a> </li>\n<li><a href=\"finetune.html\">Fine tuning the biases with pipeline-parallel training</a></li></ul>\n": "<h1>\u6837\u54c1</h1>\n<ul><li><a href=\"generate.html\">\u751f\u6210\u6587\u672c</a></li>\n</ul><li><a href=\"finetune.html\">\u901a\u8fc7\u7ba1\u9053\u5e73\u884c\u8bad\u7ec3\u5fae\u8c03\u504f\u5dee</a></li>\n",
"Samples": "\u6837\u54c1",
"Samples for inference and fine-tuning": "\u7528\u4e8e\u63a8\u7406\u548c\u5fae\u8c03\u7684\u6837\u672c"
}
translate_cache/neox/samples/finetune.ja.json
+21
View File
@@ -0,0 +1,21 @@
{
"<h1>Fine Tune GPT-NeoX</h1>\n<p>This shows how to fine tune GPT-NeoX with pipeline parallelism.</p>\n": "<h1>\u30d5\u30a1\u30a4\u30f3\u30c1\u30e5\u30fc\u30f3 GPT-\u30cd\u30aa\u30c3\u30af\u30b9</h1>\n<p>\u3053\u308c\u306f\u3001\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u4e26\u5217\u51e6\u7406\u3067GPT-Neox\u3092\u5fae\u8abf\u6574\u3059\u308b\u65b9\u6cd5\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002</p>\n",
"<h3>Create fine tuner for biases</h3>\n": "<h3>\u30d0\u30a4\u30a2\u30b9\u306e\u5fae\u8abf\u6574\u5668\u306e\u4f5c\u6210</h3>\n",
"<h3>Create pipeline parallel model</h3>\n": "<h3>\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u4e26\u5217\u30e2\u30c7\u30eb\u306e\u4f5c\u6210</h3>\n",
"<h3>Load GPT-NeoX layers</h3>\n": "<h3>GPT-Neox \u30ec\u30a4\u30e4\u30fc\u3092\u30ed\u30fc\u30c9</h3>\n",
"<h4>Tiny Shakespeare dataset</h4>\n": "<h4>\u5c0f\u3055\u306a\u30b7\u30a7\u30a4\u30af\u30b9\u30d4\u30a2\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8</h4>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Create Fairscale Pipe module </p>\n": "<p>\u30d5\u30a7\u30a2\u30b9\u30b1\u30fc\u30eb\u30d1\u30a4\u30d7\u30e2\u30b8\u30e5\u30fc\u30eb\u3092\u4f5c\u6210</p>\n",
"<p>Create experiment </p>\n": "<p>\u5b9f\u9a13\u3092\u4f5c\u6210</p>\n",
"<p>Create the Pipe module </p>\n": "<p>\u30d1\u30a4\u30d7\u30e2\u30b8\u30e5\u30fc\u30eb\u3092\u4f5c\u6210\u3059\u308b</p>\n",
"<p>Devices for each GPU </p>\n": "<p>\u5404 GPU \u306e\u30c7\u30d0\u30a4\u30b9</p>\n",
"<p>Get the layer distribution across GPUs </p>\n": "<p>GPU \u5168\u4f53\u306e\u30ec\u30a4\u30e4\u30fc\u5206\u5e03\u3092\u53d6\u5f97</p>\n",
"<p>Initialize configs </p>\n": "<p>\u30b3\u30f3\u30d5\u30a3\u30b0\u3092\u521d\u671f\u5316</p>\n",
"<p>Initialize the model. Do this before the loop for cleaner logs. </p>\n": "<p>\u30e2\u30c7\u30eb\u3092\u521d\u671f\u5316\u3057\u307e\u3059\u3002\u3053\u308c\u3092\u30eb\u30fc\u30d7\u306e\u524d\u306b\u884c\u3063\u3066\u3001\u30ed\u30b0\u3092\u30af\u30ea\u30fc\u30f3\u30a2\u30c3\u30d7\u3057\u3066\u304f\u3060\u3055\u3044\u3002</p>\n",
"<p>Make sure the finetuner is initialized </p>\n": "<p>\u30d5\u30a1\u30a4\u30f3\u30c1\u30e5\u30fc\u30ca\u30fc\u304c\u521d\u671f\u5316\u3055\u308c\u3066\u3044\u308b\u3053\u3068\u3092\u78ba\u8a8d\u3057\u3066\u304f\u3060\u3055\u3044</p>\n",
"<p>Mark biases as trainable </p>\n": "<p>\u30d0\u30a4\u30a2\u30b9\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u53ef\u80fd\u3068\u30de\u30fc\u30af\u3059\u308b</p>\n",
"<p>Start the experiment </p>\n": "<p>\u5b9f\u9a13\u3092\u59cb\u3081\u308b</p>\n",
"<p>Train </p>\n": "<p>\u5217\u8eca</p>\n",
"Fine Tune GPT-NeoX": "\u30d5\u30a1\u30a4\u30f3\u30c1\u30e5\u30fc\u30f3 GPT-\u30cd\u30aa\u30c3\u30af\u30b9",
"Fine tune GPT-NeoX biases with Fairscale pipeline parallel module": "\u30d5\u30a7\u30a2\u30b9\u30b1\u30fc\u30eb\u30fb\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u30fb\u30d1\u30e9\u30ec\u30eb\u30fb\u30e2\u30b8\u30e5\u30fc\u30eb\u306b\u3088\u308bGPT-Neox\u30d0\u30a4\u30a2\u30b9\u306e\u5fae\u8abf\u6574"
}
translate_cache/neox/samples/finetune.si.json
+21
View File
@@ -0,0 +1,21 @@
{
"<h1>Fine Tune GPT-NeoX</h1>\n<p>This shows how to fine tune GPT-NeoX with pipeline parallelism.</p>\n": "<h1>\u0dc3\u0dd2\u0dc4\u0dd2\u0db1\u0dca\u0da7\u0dd2\u0dba\u0dd4\u0db1\u0dca \u0da2\u0dd3\u0db4\u0dd3\u0da7\u0dd3-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca</h1>\n<p>\u0db1\u0dbd\u0db8\u0dcf\u0dbb\u0dca\u0d9c \u0dc3\u0db8\u0dcf\u0db1\u0dca\u0dad\u0dbb\u0d9a\u0dbb\u0dab\u0dba \u0dc3\u0db8\u0d9f \u0da2\u0dd3\u0db4\u0dd3\u0da7\u0dd3-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca \u0dc4\u0ddc\u0db3\u0dd2\u0db1\u0dca \u0dc3\u0dd4\u0dc3\u0dbb \u0d9a\u0dbb\u0db1\u0dca\u0db1\u0dda \u0d9a\u0dd9\u0dc3\u0dda\u0daf\u0dd0\u0dba\u0dd2 \u0db8\u0dd9\u0dba\u0dd2\u0db1\u0dca \u0db4\u0dd9\u0db1\u0dca\u0dc0\u0dba\u0dd2. </p>\n",
"<h3>Create fine tuner for biases</h3>\n": "<h3>\u0d85\u0d9c\u0dad\u0dd3\u0db1\u0dca\u0dc3\u0db3\u0dc4\u0dcf \u0dc3\u0dd2\u0dc4\u0dd2\u0db1\u0dca \u0dc3\u0dd4\u0dc3\u0dbb\u0d9a\u0dba\u0d9a\u0dca \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1</h3>\n",
"<h3>Create pipeline parallel model</h3>\n": "<h3>\u0db1\u0dbd\u0db8\u0dcf\u0dbb\u0dca\u0d9c \u0dc3\u0db8\u0dcf\u0db1\u0dca\u0dad\u0dbb \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0d9a\u0dca \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1</h3>\n",
"<h3>Load GPT-NeoX layers</h3>\n": "<h3>\u0da2\u0dd3\u0db4\u0dd3\u0da7\u0dd3-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca\u0dc3\u0dca\u0dae\u0dbb \u0db4\u0da7\u0dc0\u0db1\u0dca\u0db1</h3>\n",
"<h4>Tiny Shakespeare dataset</h4>\n": "<h4>\u0d9a\u0dd4\u0da9\u0dcf\u0dc2\u0dda\u0d9a\u0dca\u0dc3\u0dca\u0db4\u0dd2\u0dba\u0dbb\u0dca \u0daf\u0dad\u0dca\u0dad \u0d9a\u0da7\u0dca\u0da7\u0dbd\u0dba</h4>\n",
"<p> </p>\n": "<p> </p>\n",
"<p>Create Fairscale Pipe module </p>\n": "<p>\u0dc3\u0dcf\u0db0\u0dcf\u0dbb\u0dab\u0db1\u0dbd \u0db8\u0ddc\u0da9\u0dd2\u0dba\u0dd4\u0dbd\u0dba \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1 </p>\n",
"<p>Create experiment </p>\n": "<p>\u0d85\u0dad\u0dca\u0dc4\u0daf\u0dcf\u0db6\u0dd0\u0dbd\u0dd3\u0db8 \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1 </p>\n",
"<p>Create the Pipe module </p>\n": "<p>\u0db4\u0dba\u0dd2\u0db4\u0dca\u0db4\u0db8\u0ddc\u0da9\u0dd2\u0dba\u0dd4\u0dbd\u0dba \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1 </p>\n",
"<p>Devices for each GPU </p>\n": "<p>\u0d91\u0d9a\u0dca\u0d91\u0d9a\u0dca GPU \u0dc3\u0db3\u0dc4\u0dcf \u0d8b\u0db4\u0dcf\u0d82\u0d9c </p>\n",
"<p>Get the layer distribution across GPUs </p>\n": "<p>GPU\u0dc4\u0dbb\u0dc4\u0dcf \u0dc3\u0dca\u0dae\u0dbb \u0dc0\u0dca\u0dba\u0dcf\u0db4\u0dca\u0dad\u0dd2\u0dba \u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1 </p>\n",
"<p>Initialize configs </p>\n": "<p>\u0dc0\u0dd2\u0db1\u0dca\u0dba\u0dcf\u0dc3\u0d86\u0dbb\u0db8\u0dca\u0db7 \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Initialize the model. Do this before the loop for cleaner logs. </p>\n": "<p>\u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0d86\u0dbb\u0db8\u0dca\u0db7 \u0d9a\u0dbb\u0db1\u0dca\u0db1. \u0db4\u0dd2\u0dbb\u0dd2\u0dc3\u0dd2\u0daf\u0dd4 \u0dbd\u0ddc\u0d9c\u0dca \u0dc3\u0db3\u0dc4\u0dcf \u0dbd\u0dd6\u0db4\u0dba\u0da7 \u0db4\u0dd9\u0dbb \u0db8\u0dd9\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1. </p>\n",
"<p>Make sure the finetuner is initialized </p>\n": "<p>\u0db8\u0dd9\u0db8finetuner \u0d86\u0dbb\u0db8\u0dca\u0db7 \u0d9a\u0dbb \u0d87\u0dad\u0dd2 \u0db6\u0dc0\u0da7 \u0dc0\u0d9c \u0db6\u0dbd\u0dcf \u0d9c\u0db1\u0dca\u0db1 </p>\n",
"<p>Mark biases as trainable </p>\n": "<p>\u0db4\u0dd4\u0dc4\u0dd4\u0dab\u0dd4\u0d9a\u0dc5 \u0dc4\u0dd0\u0d9a\u0dd2 \u0dbd\u0dd9\u0dc3 \u0d85\u0d9c\u0dad\u0dd3\u0db1\u0dca \u0dc3\u0dbd\u0d9a\u0dd4\u0dab\u0dd4 \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Start the experiment </p>\n": "<p>\u0d85\u0dad\u0dca\u0dc4\u0daf\u0dcf\u0db6\u0dd0\u0dbd\u0dd3\u0db8 \u0d86\u0dbb\u0db8\u0dca\u0db7 \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Train </p>\n": "<p>\u0daf\u0dd4\u0db8\u0dca\u0dbb\u0dd2\u0dba </p>\n",
"Fine Tune GPT-NeoX": "\u0dc3\u0dd2\u0dc4\u0dd2\u0db1\u0dca \u0da7\u0dd2\u0dba\u0dd4\u0db1\u0dca \u0da2\u0dd3\u0db4\u0dd3\u0da7\u0dd3-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca",
"Fine tune GPT-NeoX biases with Fairscale pipeline parallel module": "\u0dc6\u0dd9\u0dba\u0dcf\u0dbb\u0dca\u0dc3\u0dca\u0d9a\u0dda\u0dbd\u0dca \u0db1\u0dbd \u0db8\u0dcf\u0dbb\u0dca\u0d9c \u0dc3\u0db8\u0dcf\u0db1\u0dca\u0dad\u0dbb \u0db8\u0ddc\u0da9\u0dd2\u0dba\u0dd4\u0dbd\u0dba \u0dc3\u0db8\u0d9f \u0dc4\u0ddc\u0db3 \u0dc3\u0dd4\u0dc3\u0dbb \u0da2\u0dd3\u0db4\u0dd3\u0da7\u0dd3-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca \u0d85\u0d9c\u0dad\u0dd3\u0db1\u0dca"
}
translate_cache/neox/samples/finetune.zh.json
+21
View File
@@ -0,0 +1,21 @@
{
"<h1>Fine Tune GPT-NeoX</h1>\n<p>This shows how to fine tune GPT-NeoX with pipeline parallelism.</p>\n": "<h1>Fine Tune GPT-NEOX</h1>\n<p>\u8fd9\u8bf4\u660e\u4e86\u5982\u4f55\u5229\u7528\u7ba1\u9053\u5e76\u884c\u5ea6\u5bf9 GPT-NEOX \u8fdb\u884c\u5fae\u8c03\u3002</p>\n",
"<h3>Create fine tuner for biases</h3>\n": "<h3>\u4e3a\u504f\u89c1\u521b\u5efa\u7cbe\u7ec6\u7684\u8c03\u8c10\u5668</h3>\n",
"<h3>Create pipeline parallel model</h3>\n": "<h3>\u521b\u5efa\u7ba1\u9053\u5e76\u884c\u6a21\u578b</h3>\n",
"<h3>Load GPT-NeoX layers</h3>\n": "<h3>\u52a0\u8f7d GPT-NEOX \u56fe\u5c42</h3>\n",
"<h4>Tiny Shakespeare dataset</h4>\n": "<h4>\u5c0f\u838e\u58eb\u6bd4\u4e9a\u6570\u636e\u96c6</h4>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Create Fairscale Pipe module </p>\n": "<p>\u521b\u5efa\u516c\u5e73\u89c4\u6a21\u7ba1\u9053\u6a21\u5757</p>\n",
"<p>Create experiment </p>\n": "<p>\u521b\u5efa\u5b9e\u9a8c</p>\n",
"<p>Create the Pipe module </p>\n": "<p>\u521b\u5efa\u7ba1\u9053\u6a21\u5757</p>\n",
"<p>Devices for each GPU </p>\n": "<p>\u6bcf\u4e2a GPU \u7684\u8bbe\u5907</p>\n",
"<p>Get the layer distribution across GPUs </p>\n": "<p>\u83b7\u53d6\u8de8 GPU \u7684\u5c42\u5206\u5e03</p>\n",
"<p>Initialize configs </p>\n": "<p>\u521d\u59cb\u5316\u914d\u7f6e</p>\n",
"<p>Initialize the model. Do this before the loop for cleaner logs. </p>\n": "<p>\u521d\u59cb\u5316\u6a21\u578b\u3002\u5728\u5faa\u73af\u4e4b\u524d\u6267\u884c\u6b64\u64cd\u4f5c\u4ee5\u83b7\u5f97\u66f4\u6e05\u6670\u7684\u65e5\u5fd7\u3002</p>\n",
"<p>Make sure the finetuner is initialized </p>\n": "<p>\u786e\u4fdd\u5fae\u8c03\u5668\u5df2\u521d\u59cb\u5316</p>\n",
"<p>Mark biases as trainable </p>\n": "<p>\u5c06\u504f\u89c1\u6807\u8bb0\u4e3a\u53ef\u8bad\u7ec3</p>\n",
"<p>Start the experiment </p>\n": "<p>\u5f00\u59cb\u5b9e\u9a8c</p>\n",
"<p>Train </p>\n": "<p>\u706b\u8f66</p>\n",
"Fine Tune GPT-NeoX": "Fine Tune GPT-NEOX",
"Fine tune GPT-NeoX biases with Fairscale pipeline parallel module": "\u4f7f\u7528 Fairscale \u7ba1\u9053\u5e76\u884c\u6a21\u5757\u5fae\u8c03 GPT-NEOX \u504f\u5dee"
}
translate_cache/neox/samples/generate.ja.json
+23
View File
@@ -0,0 +1,23 @@
{
"<h1>Generate Text with GPT-NeoX</h1>\n<p>This shows how to generate text from GPT-NeoX with a single GPU.</p>\n<p>This needs a GPU with more than 45GB memory.</p>\n": "<h1>GPT-\u30cd\u30aa\u30c3\u30af\u30b9\u3067\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210</h1>\n<p>\u3053\u308c\u306f\u3001\u5358\u4e00\u306eGPU\u3067GPT-Neox\u304b\u3089\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3059\u308b\u65b9\u6cd5\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002</p>\n<p>\u3053\u308c\u306b\u306f\u300145 GB \u4ee5\u4e0a\u306e\u30e1\u30e2\u30ea\u3092\u642d\u8f09\u3057\u305f GPU \u304c\u5fc5\u8981\u3067\u3059\u3002</p>\n",
"<h2>Generate text</h2>\n": "<h2>\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210</h2>\n",
"<h3>Predict the next token</h3>\n<ul><li><span translate=no>_^_0_^_</span> is the model </li>\n<li><span translate=no>_^_1_^_</span> are the input token ids </li>\n<li><span translate=no>_^_2_^_</span> is the device of the model</li></ul>\n": "<h3>\u6b21\u306e\u30c8\u30fc\u30af\u30f3\u3092\u4e88\u6e2c</h3>\n<ul><li><span translate=no>_^_0_^_</span>\u30e2\u30c7\u30eb\u3067\u3059</li>\n<li><span translate=no>_^_1_^_</span>\u306f\u5165\u529b\u30c8\u30fc\u30af\u30f3 ID</li>\n<li><span translate=no>_^_2_^_</span>\u30e2\u30c7\u30eb\u306e\u30c7\u30d0\u30a4\u30b9\u3067\u3059</li></ul>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Append the predicted token </p>\n": "<p>\u4e88\u6e2c\u30c8\u30fc\u30af\u30f3\u3092\u8ffd\u52a0</p>\n",
"<p>Device </p>\n": "<p>\u7aef\u672b</p>\n",
"<p>Eval model </p>\n": "<p>\u8a55\u4fa1\u30e2\u30c7\u30eb</p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u6b21\u306e\u30c8\u30fc\u30af\u30f3\u3092\u5165\u624b\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u4ee5\u524d\u306e\u30c8\u30fc\u30af\u30f3\u306e\u30ad\u30fc\u3068\u5024\u306e\u30da\u30a2\u3092\u30ad\u30e3\u30c3\u30b7\u30e5\u3059\u308b\u306e\u3067\u3001\u6700\u5f8c\u306e\u30c8\u30fc\u30af\u30f3\u306e\u307f\u3092\u30e2\u30c7\u30eb\u306b\u30d5\u30a3\u30fc\u30c9\u3059\u308b\u3053\u3068\u306b\u6ce8\u610f\u3057\u3066\u304f\u3060\u3055\u3044</p>\u3002\n",
"<p>Get the tokens </p>\n": "<p>\u30c8\u30fc\u30af\u30f3\u3092\u5165\u624b</p>\n",
"<p>Get token ids </p>\n": "<p>\u30c8\u30fc\u30af\u30f3 ID \u3092\u53d6\u5f97</p>\n",
"<p>Imports </p>\n": "<p>\u8f38\u5165</p>\n",
"<p>List of layers to load. This is used for testing. You can assign a subset of layers like <span translate=no>_^_0_^_</span> so that it only loads the first to transformer layers. </p>\n": "<p>\u30ed\u30fc\u30c9\u3059\u308b\u30ec\u30a4\u30e4\u30fc\u306e\u30ea\u30b9\u30c8\u3002\u3053\u308c\u306f\u30c6\u30b9\u30c8\u306b\u4f7f\u7528\u3055\u308c\u307e\u3059\u3002<span translate=no>_^_0_^_</span>\u306e\u3088\u3046\u306b\u30ec\u30a4\u30e4\u30fc\u306e\u30b5\u30d6\u30bb\u30c3\u30c8\u3092\u5272\u308a\u5f53\u3066\u3066\u3001\u6700\u521d\u306e\u30ec\u30a4\u30e4\u30fc\u306e\u307f\u3092\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u30ec\u30a4\u30e4\u30fc\u306b\u8aad\u307f\u8fbc\u3080\u3053\u3068\u304c\u3067\u304d\u307e\u3059</p>\u3002\n",
"<p>Load layers </p>\n": "<p>\u30ec\u30a4\u30e4\u30fc\u3092\u30ed\u30fc\u30c9</p>\n",
"<p>Predict 100 tokens </p>\n": "<p>\u30c8\u30fc\u30af\u30f3\u3092100\u500b\u4e88\u6e2c\u3059\u308b</p>\n",
"<p>Print </p>\n": "<p>\u30d7\u30ea\u30f3\u30c8</p>\n",
"<p>Prompt to complete </p>\n": "<p>\u5b8c\u4e86\u3092\u4fc3\u3059\u30d7\u30ed\u30f3\u30d7\u30c8</p>\n",
"<p>Return predicted token </p>\n": "<p>\u4e88\u6e2c\u30c8\u30fc\u30af\u30f3\u3092\u8fd4\u3059</p>\n",
"<p>Run the model </p>\n": "<p>\u30e2\u30c7\u30eb\u3092\u5b9f\u884c</p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u30ad\u30e3\u30c3\u30b7\u30e5\u3055\u308c\u305f\u30a2\u30af\u30c6\u30a3\u30d9\u30fc\u30b7\u30e7\u30f3\u3092\u4f7f\u7528\u3059\u308b\u3088\u3046\u306b\u72b6\u614b\u3092\u8a2d\u5b9a\u3057\u307e\u3059</p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p><a href=\"../utils/cache.html\">\u751f\u6210\u3092\u9ad8\u901f\u5316\u3059\u308b\u305f\u3081\u306b\u4e2d\u9593\u30ad\u30fc\u3068\u5024\u306e\u30da\u30a2\u3092\u30ad\u30e3\u30c3\u30b7\u30e5\u3059\u308b\u3088\u3046\u306b\u30ad\u30e3\u30c3\u30b7\u30e5\u3092\u8a2d\u5b9a</a></p>\n",
"Generate Text with GPT-NeoX": "GPT-\u30cd\u30aa\u30c3\u30af\u30b9\u3067\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210"
}
translate_cache/neox/samples/generate.si.json
+23
View File
@@ -0,0 +1,23 @@
{
"<h1>Generate Text with GPT-NeoX</h1>\n<p>This shows how to generate text from GPT-NeoX with a single GPU.</p>\n<p>This needs a GPU with more than 45GB memory.</p>\n": "<h1>GPT-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca\u0dc3\u0db8\u0d9f \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1</h1>\n<p>\u0dad\u0db1\u0dd2GPU \u0d91\u0d9a\u0d9a\u0dca \u0dc3\u0db8\u0d9f GPT-neox \u0dc0\u0dd9\u0dad\u0dd2\u0db1\u0dca \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1\u0dda \u0d9a\u0dd9\u0dc3\u0dda\u0daf \u0dba\u0db1\u0dca\u0db1 \u0db8\u0dd9\u0dba\u0dd2\u0db1\u0dca \u0db4\u0dd9\u0db1\u0dca\u0dc0\u0dba\u0dd2. </p>\n<p>\u0db8\u0dda\u0dc3\u0db3\u0dc4\u0dcf 45GB \u0da7 \u0dc0\u0da9\u0dcf \u0dc0\u0dd0\u0da9\u0dd2 \u0db8\u0dad\u0d9a\u0dba\u0d9a\u0dca \u0dc3\u0dc4\u0dd2\u0dad GPU \u0d91\u0d9a\u0d9a\u0dca \u0d85\u0dc0\u0dc1\u0dca\u0dba \u0dc0\u0dda. </p>\n",
"<h2>Generate text</h2>\n": "<h2>\u0db4\u0dd9\u0dc5\u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1</h2>\n",
"<h3>Predict the next token</h3>\n<ul><li><span translate=no>_^_0_^_</span> is the model </li>\n<li><span translate=no>_^_1_^_</span> are the input token ids </li>\n<li><span translate=no>_^_2_^_</span> is the device of the model</li></ul>\n": "<h3>\u0d8a\u0dc5\u0d9f\u0da7\u0ddd\u0d9a\u0db1\u0dba \u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1</h3>\n<ul><li><span translate=no>_^_0_^_</span> \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba \u0dc0\u0dda </li>\n<li><span translate=no>_^_1_^_</span> \u0d86\u0daf\u0dcf\u0db1 \u0da7\u0ddd\u0d9a\u0db1 \u0dc4\u0dd0\u0db3\u0dd4\u0db1\u0dd4\u0db8\u0dca \u0dc0\u0dda </li>\n</ul><li><span translate=no>_^_2_^_</span> \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0dda \u0d8b\u0db4\u0dcf\u0d82\u0d9c\u0dba \u0dc0\u0dda</li>\n",
"<p> </p>\n": "<p> </p>\n",
"<p>Append the predicted token </p>\n": "<p>\u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba\u0d9a\u0dc5 \u0da7\u0ddd\u0d9a\u0db1\u0dba \u0d91\u0d9a\u0dca \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Device </p>\n": "<p>\u0d8b\u0db4\u0dcf\u0d82\u0d9c\u0dba </p>\n",
"<p>Eval model </p>\n": "<p>\u0d91\u0dc0\u0dcf\u0dbd\u0dca\u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba </p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u0d8a\u0dc5\u0d9f\u0da7\u0ddd\u0d9a\u0db1\u0dba \u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1. \u0db4\u0dd9\u0dbb \u0da7\u0ddd\u0d9a\u0db1 \u0dc0\u0dbd \u0dba\u0dad\u0dd4\u0dbb/\u0d85\u0d9c\u0dba \u0dba\u0dd4\u0d9c\u0dbd \u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2 \u0d9a\u0dbb\u0db1 \u0db1\u0dd2\u0dc3\u0dcf \u0d85\u0db4\u0dd2 \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0da7 \u0d85\u0dc0\u0dc3\u0dcf\u0db1 \u0da7\u0ddd\u0d9a\u0db1\u0dba \u0db4\u0db8\u0dab\u0d9a\u0dca \u0db4\u0ddd\u0dc2\u0dab\u0dba \u0d9a\u0dbb\u0db1 \u0db6\u0dc0 \u0dc3\u0dbd\u0d9a\u0db1\u0dca\u0db1. </p>\n",
"<p>Get the tokens </p>\n": "<p>\u0da7\u0ddd\u0d9a\u0db1\u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1 </p>\n",
"<p>Get token ids </p>\n": "<p>\u0da7\u0ddd\u0d9a\u0db1\u0dca\u0dc4\u0dd0\u0db3\u0dd4\u0db1\u0dd4\u0db8\u0dca\u0db4\u0dad\u0dca \u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1 </p>\n",
"<p>Imports </p>\n": "<p>\u0d86\u0db1\u0dba\u0db1 </p>\n",
"<p>List of layers to load. This is used for testing. You can assign a subset of layers like <span translate=no>_^_0_^_</span> so that it only loads the first to transformer layers. </p>\n": "<p>\u0db4\u0dd0\u0da7\u0dc0\u0dd2\u0dba\u0dba\u0dd4\u0dad\u0dd4 \u0dc3\u0dca\u0dae\u0dbb \u0dbd\u0dd0\u0dba\u0dd2\u0dc3\u0dca\u0dad\u0dd4\u0dc0. \u0db8\u0dd9\u0dba \u0db4\u0dbb\u0dd3\u0d9a\u0dca\u0dc2\u0dcf \u0d9a\u0dd2\u0dbb\u0dd3\u0db8 \u0dc3\u0db3\u0dc4\u0dcf \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0dc0\u0dda. \u0da7\u0dca\u0dbb\u0dcf\u0db1\u0dca\u0dc3\u0dca\u0dc6\u0ddd\u0db8\u0dbb\u0dca \u0dc3\u0dca\u0dae\u0dbb \u0dc0\u0dbd\u0da7 \u0db4\u0dc5\u0db8\u0dd4 \u0db4\u0da7\u0dc0\u0db1\u0dd4 \u0dbd\u0db6\u0db1 <span translate=no>_^_0_^_</span> \u0db4\u0dbb\u0dd2\u0daf\u0dd2 \u0d94\u0db6\u0da7 \u0dc0\u0dd0\u0db1\u0dd2 \u0dc3\u0dca\u0dae\u0dbb \u0d8b\u0db4 \u0d9a\u0dd4\u0dbd\u0d9a\u0dba\u0d9a\u0dca \u0db4\u0dd0\u0dc0\u0dbb\u0dd2\u0dba \u0dc4\u0dd0\u0d9a\u0dd2\u0dba. </p>\n",
"<p>Load layers </p>\n": "<p>\u0dc3\u0dca\u0dae\u0dbb\u0db4\u0dd6\u0dbb\u0dab\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Predict 100 tokens </p>\n": "<p>\u0da7\u0ddd\u0d9a\u0db1100 \u0d9a\u0dca \u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Print </p>\n": "<p>\u0db8\u0dd4\u0daf\u0dca\u0dbb\u0dab\u0dba </p>\n",
"<p>Prompt to complete </p>\n": "<p>\u0dc3\u0db8\u0dca\u0db4\u0dd6\u0dbb\u0dca\u0dab\u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0da7 \u0dc0\u0dd2\u0db8\u0dc3\u0db1\u0dca\u0db1 </p>\n",
"<p>Return predicted token </p>\n": "<p>\u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba\u0d9a\u0dc5 \u0da7\u0ddd\u0d9a\u0db1\u0dba \u0d86\u0db4\u0dc3\u0dd4 \u0dba\u0db1\u0dca\u0db1 </p>\n",
"<p>Run the model </p>\n": "<p>\u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0db0\u0dcf\u0dc0\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2\u0dc3\u0d9a\u0dca\u0dbb\u0dd2\u0dba \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0dca \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0da7 \u0dbb\u0dcf\u0da2\u0dca\u0dba\u0dba \u0dc3\u0d9a\u0dc3\u0db1\u0dca\u0db1 </p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p>\u0dc0\u0dda\u0d9c\u0dc0\u0dad\u0dca\u0d8b\u0dad\u0dca\u0db4\u0dcf\u0daf\u0db1\u0dba \u0dc3\u0db3\u0dc4\u0dcf \u0d85\u0dad\u0dbb\u0db8\u0dd0\u0daf\u0dd2 \u0dba\u0dad\u0dd4\u0dbb/\u0dc0\u0da7\u0dd2\u0db1\u0dcf\u0d9a\u0db8\u0dca \u0dba\u0dd4\u0d9c\u0dbd <a href=\"../utils/cache.html\">\u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2</a> \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0da7 \u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2\u0dba \u0db4\u0dd2\u0dc4\u0dd2\u0da7\u0dd4\u0dc0\u0db1\u0dca\u0db1 </p>\n",
"Generate Text with GPT-NeoX": "GPT-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca \u0dc3\u0db8\u0d9f \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1"
}
translate_cache/neox/samples/generate.zh.json
+23
View File
@@ -0,0 +1,23 @@
{
"<h1>Generate Text with GPT-NeoX</h1>\n<p>This shows how to generate text from GPT-NeoX with a single GPU.</p>\n<p>This needs a GPU with more than 45GB memory.</p>\n": "<h1>\u4f7f\u7528 GPT-NEOX \u751f\u6210\u6587\u672c</h1>\n<p>\u8fd9\u8bf4\u660e\u4e86\u5982\u4f55\u4f7f\u7528\u5355\u4e2a GPU \u4ece GPT-NEOX \u751f\u6210\u6587\u672c\u3002</p>\n<p>\u8fd9\u9700\u8981\u4e00\u4e2a\u5185\u5b58\u8d85\u8fc745GB\u7684GPU\u3002</p>\n",
"<h2>Generate text</h2>\n": "<h2>\u751f\u6210\u6587\u672c</h2>\n",
"<h3>Predict the next token</h3>\n<ul><li><span translate=no>_^_0_^_</span> is the model </li>\n<li><span translate=no>_^_1_^_</span> are the input token ids </li>\n<li><span translate=no>_^_2_^_</span> is the device of the model</li></ul>\n": "<h3>\u9884\u6d4b\u4e0b\u4e00\u4e2a\u4ee3\u5e01</h3>\n<ul><li><span translate=no>_^_0_^_</span>\u662f\u6a21\u7279\u5417</li>\n<li><span translate=no>_^_1_^_</span>\u662f\u8f93\u5165\u4ee4\u724c ID</li>\n<li><span translate=no>_^_2_^_</span>\u662f\u8be5\u578b\u53f7\u7684\u8bbe\u5907</li></ul>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Append the predicted token </p>\n": "<p>\u8ffd\u52a0\u9884\u6d4b\u7684\u4ee4\u724c</p>\n",
"<p>Device </p>\n": "<p>\u8bbe\u5907</p>\n",
"<p>Eval model </p>\n": "<p>\u8bc4\u4f30\u6a21\u578b</p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u83b7\u53d6\u4e0b\u4e00\u4e2a\u4ee4\u724c\u3002\u8bf7\u6ce8\u610f\uff0c\u6211\u4eec\u53ea\u5c06\u6700\u540e\u4e00\u4e2a\u4ee4\u724c\u63d0\u4f9b\u7ed9\u6a21\u578b\uff0c\u56e0\u4e3a\u6211\u4eec\u7f13\u5b58\u4e86\u5148\u524d\u4ee4\u724c\u7684\u952e/\u503c\u5bf9\u3002</p>\n",
"<p>Get the tokens </p>\n": "<p>\u83b7\u53d6\u4ee3\u5e01</p>\n",
"<p>Get token ids </p>\n": "<p>\u83b7\u53d6\u4ee3\u5e01 ID</p>\n",
"<p>Imports </p>\n": "<p>\u8fdb\u53e3</p>\n",
"<p>List of layers to load. This is used for testing. You can assign a subset of layers like <span translate=no>_^_0_^_</span> so that it only loads the first to transformer layers. </p>\n": "<p>\u8981\u52a0\u8f7d\u7684\u56fe\u5c42\u5217\u8868\u3002\u8fd9\u7528\u4e8e\u6d4b\u8bd5\u3002\u60a8\u53ef\u4ee5\u5c06\u5c42\u7684\u5b50\u96c6\u5206\u914d\u7ed9\u53d8\u538b\u5668\u5c42\uff0c<span translate=no>_^_0_^_</span>\u4f7f\u5176\u4ec5\u5c06\u7b2c\u4e00\u4e2a\u5c42\u52a0\u8f7d\u5230\u53d8\u538b\u5668\u5c42\u3002</p>\n",
"<p>Load layers </p>\n": "<p>\u52a0\u8f7d\u56fe\u5c42</p>\n",
"<p>Predict 100 tokens </p>\n": "<p>\u9884\u6d4b 100 \u4e2a\u4ee3\u5e01</p>\n",
"<p>Print </p>\n": "<p>\u6253\u5370</p>\n",
"<p>Prompt to complete </p>\n": "<p>\u63d0\u793a\u5b8c\u6210</p>\n",
"<p>Return predicted token </p>\n": "<p>\u8fd4\u56de\u9884\u6d4b\u7684\u4ee3\u5e01</p>\n",
"<p>Run the model </p>\n": "<p>\u8fd0\u884c\u6a21\u578b</p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u8bbe\u7f6e\u72b6\u6001\u4ee5\u4f7f\u7528\u7f13\u5b58\u7684\u6fc0\u6d3b</p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p>\u8bbe\u7f6e<a href=\"../utils/cache.html\">\u7f13\u5b58</a>\u4ee5\u7f13\u5b58\u4e2d\u95f4\u952e/\u503c\u5bf9\u4ee5\u52a0\u5feb\u751f\u6210\u901f\u5ea6</p>\n",
"Generate Text with GPT-NeoX": "\u4f7f\u7528 GPT-NEOX \u751f\u6210\u6587\u672c"
}
translate_cache/neox/samples/llm_int8.ja.json
+19
View File
@@ -0,0 +1,19 @@
{
"<h1>Generate Text with GPT-NeoX using LLM.int8() quantization</h1>\n<p>This shows how to generate text from GPT-NeoX using <a href=\"../utils/llm_int8.html\">LLM.int8() quantization</a>.</p>\n<p>This needs a GPU with 24GB memory.</p>\n": "<h1>LLM.int8 () \u91cf\u5b50\u5316\u3092\u4f7f\u7528\u3057\u3066 GPT-Neox \u3067\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210</h1>\n<p>\u3053\u308c\u306f\u3001<a href=\"../utils/llm_int8.html\">LLM.int8</a> () \u91cf\u5b50\u5316\u3092\u4f7f\u7528\u3057\u3066 GPT-Neox \u304b\u3089\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210\u3059\u308b\u65b9\u6cd5\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002</p>\n<p>\u3053\u308c\u306b\u306f 24 GB \u306e\u30e1\u30e2\u30ea\u3092\u642d\u8f09\u3057\u305f GPU \u304c\u5fc5\u8981\u3067\u3059\u3002</p>\n",
"<h2>Generate text</h2>\n": "<h2>\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210</h2>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Append the predicted token </p>\n": "<p>\u4e88\u6e2c\u30c8\u30fc\u30af\u30f3\u3092\u8ffd\u52a0</p>\n",
"<p>Clear cache and print memory summary for debugging </p>\n": "<p>\u30c7\u30d0\u30c3\u30b0\u7528\u306b\u30ad\u30e3\u30c3\u30b7\u30e5\u3092\u30af\u30ea\u30a2\u3057\u3066\u30e1\u30e2\u30ea\u306e\u6982\u8981\u3092\u5370\u5237</p>\n",
"<p>Create <span translate=no>_^_0_^_</span> model </p>\n": "<p><span translate=no>_^_0_^_</span>\u30e2\u30c7\u30eb\u4f5c\u6210</p>\n",
"<p>Device </p>\n": "<p>\u7aef\u672b</p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u6b21\u306e\u30c8\u30fc\u30af\u30f3\u3092\u5165\u624b\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u4ee5\u524d\u306e\u30c8\u30fc\u30af\u30f3\u306e\u30ad\u30fc\u3068\u5024\u306e\u30da\u30a2\u3092\u30ad\u30e3\u30c3\u30b7\u30e5\u3059\u308b\u306e\u3067\u3001\u6700\u5f8c\u306e\u30c8\u30fc\u30af\u30f3\u306e\u307f\u3092\u30e2\u30c7\u30eb\u306b\u30d5\u30a3\u30fc\u30c9\u3059\u308b\u3053\u3068\u306b\u6ce8\u610f\u3057\u3066\u304f\u3060\u3055\u3044</p>\u3002\n",
"<p>Get token ids </p>\n": "<p>\u30c8\u30fc\u30af\u30f3 ID \u3092\u53d6\u5f97</p>\n",
"<p>Load layers in float16 into CPU. We convert the layers to int8 later, because doing that on the fly after loading layers to GPU causes CUDA memory fragmentation (about 3GB memory can get lost due to fragmentation). </p>\n": "<p>float16 \u306e\u30ec\u30a4\u30e4\u30fc\u3092 CPU \u306b\u30ed\u30fc\u30c9\u3057\u307e\u3059\u3002\u30ec\u30a4\u30e4\u30fc\u3092GPU\u306b\u30ed\u30fc\u30c9\u3057\u305f\u5f8c\u306b\u305d\u306e\u5834\u3067\u3053\u308c\u3092\u884c\u3046\u3068\u3001CUDA\u30e1\u30e2\u30ea\u306e\u65ad\u7247\u5316\u304c\u767a\u751f\u3059\u308b\u305f\u3081\u3001\u5f8c\u3067\u30ec\u30a4\u30e4\u30fc\u3092int8\u306b\u5909\u63db\u3057\u307e\u3059\uff08\u30d5\u30e9\u30b0\u30e1\u30f3\u30c6\u30fc\u30b7\u30e7\u30f3\u306b\u3088\u308a\u7d043GB\u306e\u30e1\u30e2\u30ea\u304c\u5931\u308f\u308c\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059</p>\uff09\u3002\n",
"<p>Predict 100 tokens </p>\n": "<p>\u30c8\u30fc\u30af\u30f3\u3092100\u500b\u4e88\u6e2c\u3059\u308b</p>\n",
"<p>Print </p>\n": "<p>\u30d7\u30ea\u30f3\u30c8</p>\n",
"<p>Run the model. We use the <a href=\"generate.html\"><span translate=no>_^_0_^_</span></a> function defined in <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a> </p>\n": "<p>\u30e2\u30c7\u30eb\u3092\u5b9f\u884c\u3057\u307e\u3059\u3002<a href=\"generate.html\"><span translate=no>_^_0_^_</span></a>\u3067\u5b9a\u7fa9\u3055\u308c\u3066\u3044\u308b\u95a2\u6570\u3092\u4f7f\u7528\u3057\u307e\u3059 <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a></p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u30ad\u30e3\u30c3\u30b7\u30e5\u3055\u308c\u305f\u30a2\u30af\u30c6\u30a3\u30d9\u30fc\u30b7\u30e7\u30f3\u3092\u4f7f\u7528\u3059\u308b\u3088\u3046\u306b\u72b6\u614b\u3092\u8a2d\u5b9a\u3057\u307e\u3059</p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p><a href=\"../utils/cache.html\">\u751f\u6210\u3092\u9ad8\u901f\u5316\u3059\u308b\u305f\u3081\u306b\u4e2d\u9593\u30ad\u30fc\u3068\u5024\u306e\u30da\u30a2\u3092\u30ad\u30e3\u30c3\u30b7\u30e5\u3059\u308b\u3088\u3046\u306b\u30ad\u30e3\u30c3\u30b7\u30e5\u3092\u8a2d\u5b9a</a></p>\n",
"<p>This reduces CUDA memory fragmentation </p>\n": "<p>\u3053\u308c\u306b\u3088\u308a\u3001CUDA \u30e1\u30e2\u30ea\u306e\u65ad\u7247\u5316\u304c\u6e1b\u5c11\u3057\u307e\u3059\u3002</p>\n",
"Generate Text with GPT-NeoX using LLM.int8() quantization": "LLM.int8 () \u91cf\u5b50\u5316\u3092\u4f7f\u7528\u3057\u3066 GPT-Neox \u3067\u30c6\u30ad\u30b9\u30c8\u3092\u751f\u6210"
}
translate_cache/neox/samples/llm_int8.si.json
+19
View File
@@ -0,0 +1,19 @@
{
"<h1>Generate Text with GPT-NeoX using LLM.int8() quantization</h1>\n<p>This shows how to generate text from GPT-NeoX using <a href=\"../utils/llm_int8.html\">LLM.int8() quantization</a>.</p>\n<p>This needs a GPU with 24GB memory.</p>\n": "<h1>LLM.INT8() \u0db4\u0dca\u0dbb\u0db8\u0dcf\u0dab\u0d9a\u0dbb\u0dab\u0dba \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dbb\u0db8\u0dd2\u0db1\u0dca GPT-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca \u0dc3\u0db8\u0d9f \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1</h1>\n<p><a href=\"../utils/llm_int8.html\">LLM.INT8 () \u0db4\u0dca\u0dbb\u0db8\u0dcf\u0dab\u0d9a\u0dbb\u0dab\u0dba</a>\u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dbb\u0db8\u0dd2\u0db1\u0dca GPT-Neox \u0dc0\u0dd9\u0dad\u0dd2\u0db1\u0dca \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1\u0dda \u0d9a\u0dd9\u0dc3\u0dda\u0daf \u0dba\u0db1\u0dca\u0db1 \u0db8\u0dd9\u0dba\u0dd2\u0db1\u0dca \u0db4\u0dd9\u0db1\u0dca\u0dc0\u0dba\u0dd2. </p>\n<p>\u0db8\u0dda\u0dc3\u0db3\u0dc4\u0dcf 24GB \u0db8\u0dad\u0d9a\u0dba\u0d9a\u0dca \u0dc3\u0dc4\u0dd2\u0dad GPU \u0d91\u0d9a\u0d9a\u0dca \u0d85\u0dc0\u0dc1\u0dca\u0dba \u0dc0\u0dda. </p>\n",
"<h2>Generate text</h2>\n": "<h2>\u0db4\u0dd9\u0dc5\u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1</h2>\n",
"<p> </p>\n": "<p> </p>\n",
"<p>Append the predicted token </p>\n": "<p>\u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba\u0d9a\u0dc5 \u0da7\u0ddd\u0d9a\u0db1\u0dba \u0d91\u0d9a\u0dca \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Clear cache and print memory summary for debugging </p>\n": "<p>\u0db1\u0dd2\u0daf\u0ddc\u0dc3\u0dca\u0d9a\u0dbb\u0dab\u0dba\u0dc3\u0db3\u0dc4\u0dcf \u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2 \u0dc3\u0dc4 \u0db8\u0dd4\u0daf\u0dca\u0dbb\u0dd2\u0dad \u0db8\u0dad\u0d9a \u0dc3\u0dcf\u0dbb\u0dcf\u0d82\u0dc1\u0dba \u0db4\u0dd0\u0dc4\u0dd0\u0daf\u0dd2\u0dbd\u0dd2 \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Create <span translate=no>_^_0_^_</span> model </p>\n": "<p><span translate=no>_^_0_^_</span> \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba \u0dc3\u0dcf\u0daf\u0db1\u0dca\u0db1 </p>\n",
"<p>Device </p>\n": "<p>\u0d8b\u0db4\u0dcf\u0d82\u0d9c\u0dba </p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u0d8a\u0dc5\u0d9f\u0da7\u0ddd\u0d9a\u0db1\u0dba \u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1. \u0db4\u0dd9\u0dbb \u0da7\u0ddd\u0d9a\u0db1 \u0dc0\u0dbd \u0dba\u0dad\u0dd4\u0dbb/\u0d85\u0d9c\u0dba \u0dba\u0dd4\u0d9c\u0dbd \u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2 \u0d9a\u0dbb\u0db1 \u0db1\u0dd2\u0dc3\u0dcf \u0d85\u0db4\u0dd2 \u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0da7 \u0d85\u0dc0\u0dc3\u0dcf\u0db1 \u0da7\u0ddd\u0d9a\u0db1\u0dba \u0db4\u0db8\u0dab\u0d9a\u0dca \u0db4\u0ddd\u0dc2\u0dab\u0dba \u0d9a\u0dbb\u0db1 \u0db6\u0dc0 \u0dc3\u0dbd\u0d9a\u0db1\u0dca\u0db1. </p>\n",
"<p>Get token ids </p>\n": "<p>\u0da7\u0ddd\u0d9a\u0db1\u0dca\u0dc4\u0dd0\u0db3\u0dd4\u0db1\u0dd4\u0db8\u0dca\u0db4\u0dad\u0dca \u0dbd\u0db6\u0dcf \u0d9c\u0db1\u0dca\u0db1 </p>\n",
"<p>Load layers in float16 into CPU. We convert the layers to int8 later, because doing that on the fly after loading layers to GPU causes CUDA memory fragmentation (about 3GB memory can get lost due to fragmentation). </p>\n": "<p>\u0db4\u0dcf\u0dc0\u0dd9\u0db116 \u0dc4\u0dd2 \u0dc3\u0dca\u0dae\u0dbb CPU \u0dad\u0dd4\u0dc5\u0da7 \u0db4\u0da7\u0dc0\u0db1\u0dca\u0db1. \u0d85\u0db4\u0dd2 \u0dc3\u0dca\u0dae\u0dbb \u0db4\u0dc3\u0dd4\u0dc0 int8 \u0db6\u0dc0\u0da7 \u0db4\u0dbb\u0dd2\u0dc0\u0dbb\u0dca\u0dad\u0db1\u0dba \u0d9a\u0dbb\u0db8\u0dd4, \u0db8\u0db1\u0dca\u0daf \u0dc3\u0dca\u0dae\u0dbb GPU \u0dc0\u0dd9\u0dad \u0db4\u0dd0\u0da7\u0dc0\u0dd3\u0db8\u0dd9\u0db1\u0dca \u0db4\u0dc3\u0dd4 \u0db4\u0dd2\u0dba\u0dcf\u0dc3\u0dbb \u0d9a\u0dd2\u0dbb\u0dd3\u0db8 CUDA \u0db8\u0dad\u0d9a \u0d9b\u0dab\u0dca\u0da9\u0db1\u0dba \u0dc0\u0dd3\u0db8\u0da7 \u0dc4\u0dda\u0dad\u0dd4 \u0dc0\u0dda (3GB \u0db4\u0db8\u0dab \u0db8\u0dad\u0d9a\u0dba \u0d9a\u0dd0\u0db6\u0dbd\u0dd2 \u0dc0\u0dd3\u0db8 \u0db1\u0dd2\u0dc3\u0dcf \u0d85\u0dc4\u0dd2\u0db8\u0dd2 \u0dc0\u0dd2\u0dba \u0dc4\u0dd0\u0d9a). </p>\n",
"<p>Predict 100 tokens </p>\n": "<p>\u0da7\u0ddd\u0d9a\u0db1100 \u0d9a\u0dca \u0db4\u0dd4\u0dbb\u0ddd\u0d9a\u0dae\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1 </p>\n",
"<p>Print </p>\n": "<p>\u0db8\u0dd4\u0daf\u0dca\u0dbb\u0dab\u0dba </p>\n",
"<p>Run the model. We use the <a href=\"generate.html\"><span translate=no>_^_0_^_</span></a> function defined in <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a> </p>\n": "<p>\u0d86\u0d9a\u0dd8\u0dad\u0dd2\u0dba\u0db0\u0dcf\u0dc0\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1. \u0d85\u0db4\u0dd2 \u0d85\u0dbb\u0dca\u0dae \u0daf\u0d9a\u0dca\u0dc0\u0dcf \u0d87\u0dad\u0dd2 <a href=\"generate.html\"><span translate=no>_^_0_^_</span></a> \u0dc1\u0dca\u0dbb\u0dd2\u0dad\u0dba \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dbb\u0db8\u0dd4 <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a> </p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2\u0dc3\u0d9a\u0dca\u0dbb\u0dd2\u0dba \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0dca \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0da7 \u0dbb\u0dcf\u0da2\u0dca\u0dba\u0dba \u0dc3\u0d9a\u0dc3\u0db1\u0dca\u0db1 </p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p>\u0dc0\u0dda\u0d9c\u0dc0\u0dad\u0dca\u0d8b\u0dad\u0dca\u0db4\u0dcf\u0daf\u0db1\u0dba \u0dc3\u0db3\u0dc4\u0dcf \u0d85\u0dad\u0dbb\u0db8\u0dd0\u0daf\u0dd2 \u0dba\u0dad\u0dd4\u0dbb/\u0dc0\u0da7\u0dd2\u0db1\u0dcf\u0d9a\u0db8\u0dca \u0dba\u0dd4\u0d9c\u0dbd <a href=\"../utils/cache.html\">\u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2</a> \u0d9a\u0dd2\u0dbb\u0dd3\u0db8\u0da7 \u0dc4\u0dd0\u0db9\u0dd2\u0dbd\u0dd2\u0dba \u0db4\u0dd2\u0dc4\u0dd2\u0da7\u0dd4\u0dc0\u0db1\u0dca\u0db1 </p>\n",
"<p>This reduces CUDA memory fragmentation </p>\n": "<p>\u0db8\u0dd9\u0dbaCUDA \u0db8\u0dad\u0d9a \u0d9b\u0dab\u0dca\u0da9\u0db1\u0dba \u0d85\u0da9\u0dd4 \u0d9a\u0dbb\u0dba\u0dd2 </p>\n",
"Generate Text with GPT-NeoX using LLM.int8() quantization": "LLM.INT8 () \u0db4\u0dca\u0dbb\u0db8\u0dcf\u0dab\u0d9a\u0dbb\u0dab\u0dba \u0db7\u0dcf\u0dc0\u0dd2\u0dad\u0dcf \u0d9a\u0dbb\u0db8\u0dd2\u0db1\u0dca GPT-\u0db1\u0dd2\u0dba\u0ddd\u0d9a\u0dca\u0dc3\u0dca \u0dc3\u0db8\u0d9f \u0db4\u0dd9\u0dc5 \u0da2\u0db1\u0db1\u0dba \u0d9a\u0dbb\u0db1\u0dca\u0db1"
}
translate_cache/neox/samples/llm_int8.zh.json
+19
View File
@@ -0,0 +1,19 @@
{
"<h1>Generate Text with GPT-NeoX using LLM.int8() quantization</h1>\n<p>This shows how to generate text from GPT-NeoX using <a href=\"../utils/llm_int8.html\">LLM.int8() quantization</a>.</p>\n<p>This needs a GPU with 24GB memory.</p>\n": "<h1>\u4f7f\u7528 llm.int8 () \u91cf\u5316\u4f7f\u7528 GPT-NEOX \u751f\u6210\u6587\u672c</h1>\n<p>\u8fd9\u8bf4\u660e\u4e86\u5982\u4f55\u4f7f\u7528 <a href=\"../utils/llm_int8.html\">llm.int8 () \u91cf\u5316</a>\u4ece GPT-NEOX \u751f\u6210\u6587\u672c\u3002</p>\n<p>\u8fd9\u9700\u8981\u4e00\u4e2a\u5177\u6709 24GB \u5185\u5b58\u7684 GPU\u3002</p>\n",
"<h2>Generate text</h2>\n": "<h2>\u751f\u6210\u6587\u672c</h2>\n",
"<p> </p>\n": "<p></p>\n",
"<p>Append the predicted token </p>\n": "<p>\u8ffd\u52a0\u9884\u6d4b\u7684\u4ee4\u724c</p>\n",
"<p>Clear cache and print memory summary for debugging </p>\n": "<p>\u6e05\u9664\u7f13\u5b58\u548c\u6253\u5370\u5185\u5b58\u6458\u8981\u4ee5\u8fdb\u884c\u8c03\u8bd5</p>\n",
"<p>Create <span translate=no>_^_0_^_</span> model </p>\n": "<p>\u521b\u5efa<span translate=no>_^_0_^_</span>\u6a21\u578b</p>\n",
"<p>Device </p>\n": "<p>\u8bbe\u5907</p>\n",
"<p>Get next token. Note that we only feed the last token to the model because we cache the key/value pairs of previous tokens. </p>\n": "<p>\u83b7\u53d6\u4e0b\u4e00\u4e2a\u4ee4\u724c\u3002\u8bf7\u6ce8\u610f\uff0c\u6211\u4eec\u53ea\u5c06\u6700\u540e\u4e00\u4e2a\u4ee4\u724c\u63d0\u4f9b\u7ed9\u6a21\u578b\uff0c\u56e0\u4e3a\u6211\u4eec\u7f13\u5b58\u4e86\u5148\u524d\u4ee4\u724c\u7684\u952e/\u503c\u5bf9\u3002</p>\n",
"<p>Get token ids </p>\n": "<p>\u83b7\u53d6\u4ee3\u5e01 ID</p>\n",
"<p>Load layers in float16 into CPU. We convert the layers to int8 later, because doing that on the fly after loading layers to GPU causes CUDA memory fragmentation (about 3GB memory can get lost due to fragmentation). </p>\n": "\u5c06 <p>float16 \u4e2d\u7684\u5c42\u52a0\u8f7d\u5230 CPU \u4e2d\u3002\u6211\u4eec\u7a0d\u540e\u5c06\u56fe\u5c42\u8f6c\u6362\u4e3aint8\uff0c\u56e0\u4e3a\u5728\u5c06\u56fe\u5c42\u52a0\u8f7d\u5230GPU\u540e\u5373\u65f6\u6267\u884c\u6b64\u64cd\u4f5c\u4f1a\u5bfc\u81f4CUDA\u5185\u5b58\u788e\u7247\uff08\u5927\u7ea63GB\u7684\u5185\u5b58\u53ef\u80fd\u4f1a\u7531\u4e8e\u788e\u7247\u800c\u4e22\u5931\uff09\u3002</p>\n",
"<p>Predict 100 tokens </p>\n": "<p>\u9884\u6d4b 100 \u4e2a\u4ee3\u5e01</p>\n",
"<p>Print </p>\n": "<p>\u6253\u5370</p>\n",
"<p>Run the model. We use the <a href=\"generate.html\"><span translate=no>_^_0_^_</span></a> function defined in <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a> </p>\n": "<p>\u8fd0\u884c\u6a21\u578b\u3002\u6211\u4eec\u4f7f\u7528\u4e2d\u5b9a\u4e49\u7684<a href=\"generate.html\"><span translate=no>_^_0_^_</span></a>\u51fd\u6570 <a href=\"generate.html\"><span translate=no>_^_1_^_</span></a></p>\n",
"<p>Set the state to use cached activations </p>\n": "<p>\u8bbe\u7f6e\u72b6\u6001\u4ee5\u4f7f\u7528\u7f13\u5b58\u7684\u6fc0\u6d3b</p>\n",
"<p>Setup <a href=\"../utils/cache.html\">cache</a> to cache intermediate key/value pairs for faster generation </p>\n": "<p>\u8bbe\u7f6e<a href=\"../utils/cache.html\">\u7f13\u5b58</a>\u4ee5\u7f13\u5b58\u4e2d\u95f4\u952e/\u503c\u5bf9\u4ee5\u52a0\u5feb\u751f\u6210\u901f\u5ea6</p>\n",
"<p>This reduces CUDA memory fragmentation </p>\n": "<p>\u8fd9\u51cf\u5c11\u4e86 CUDA \u5185\u5b58\u788e\u7247</p>\n",
"Generate Text with GPT-NeoX using LLM.int8() quantization": "\u4f7f\u7528 llm.int8 () \u91cf\u5316\u4f7f\u7528 GPT-NEOX \u751f\u6210\u6587\u672c"
}