Spaces:
Running
Running
updating heatmap
Browse files- index.html +9 -25
- xorbs.json +0 -0
index.html
CHANGED
@@ -3,7 +3,7 @@
|
|
3 |
<head>
|
4 |
<meta charset="UTF-8" />
|
5 |
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
6 |
-
<title>
|
7 |
<link rel="stylesheet" href="style.css" />
|
8 |
<script src="https://cdn.jsdelivr.net/npm/vega@5"></script>
|
9 |
<script src="https://cdn.jsdelivr.net/npm/vega-lite@5"></script>
|
@@ -12,35 +12,19 @@
|
|
12 |
<body>
|
13 |
<div class="container">
|
14 |
<div class="header">
|
15 |
-
<h1>Visualizing
|
16 |
<p>
|
17 |
-
|
18 |
-
models in
|
19 |
-
<a
|
20 |
-
target="_blank"
|
21 |
-
href="https://huggingface.co/bartowski/gemma-2-9b-it-GGUF"
|
22 |
-
>bartowski/gemma-2-9b-it-GGUF</a
|
23 |
-
>.
|
24 |
</p>
|
|
|
25 |
<p>
|
26 |
-
|
27 |
-
to 64MB. The color of each block represents the deduplication ratio
|
28 |
-
for the block, which is a function of how often the chunks in the
|
29 |
-
block are shared between files. The darker the color, the more
|
30 |
-
frequently content is shared, the better the overall upload and
|
31 |
-
download times for a given file! The deduplication savings here take a
|
32 |
-
191GB repo and cut it down to 97GB, helping to shave a few hours off
|
33 |
-
the upload time.
|
34 |
</p>
|
|
|
35 |
<p>
|
36 |
-
|
37 |
-
of how we make this all work in our accompanying
|
38 |
-
<a
|
39 |
-
target="_blank"
|
40 |
-
href="https://huggingface.co/blog/from-chunks-to-blocks"
|
41 |
-
>blog post</a
|
42 |
-
>.
|
43 |
</p>
|
|
|
44 |
To explore the visualization:
|
45 |
<ul>
|
46 |
<li>
|
@@ -119,7 +103,7 @@
|
|
119 |
],
|
120 |
field: "dedupe_factor",
|
121 |
type: "quantitative",
|
122 |
-
scale: { scheme: "blues", domain: [0,
|
123 |
},
|
124 |
opacity: {
|
125 |
condition: [
|
|
|
3 |
<head>
|
4 |
<meta charset="UTF-8" />
|
5 |
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
6 |
+
<title>XLM-R Dedupe</title>
|
7 |
<link rel="stylesheet" href="style.css" />
|
8 |
<script src="https://cdn.jsdelivr.net/npm/vega@5"></script>
|
9 |
<script src="https://cdn.jsdelivr.net/npm/vega-lite@5"></script>
|
|
|
12 |
<body>
|
13 |
<div class="container">
|
14 |
<div class="header">
|
15 |
+
<h1>Visualizing XLM-rBERTa Finetune Dedupe</h1>
|
16 |
<p>
|
17 |
+
This heatmap shows deduplication across a family of fine-tuned models based on <a href="https://huggingface.co/papers/1911.02116">XLM-RoBERTa large</a>, a multilingual transformer introduced in 2019 and trained on 100 languages. Each row represents a model repository (which often contains multiple formats—e.g., Safetensor, Keras, PyTorch) derived from the original research. Repository data is chunked into blocks of up to 64MB in Xet's storage layer, and this heatmap visualizes those blocks across models.
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
</p>
|
19 |
+
|
20 |
<p>
|
21 |
+
The base model is <a href="https://huggingface.co/FacebookAI/xlm-roberta-large"><code>xlm-roberta-large</code></a>, while the others are fine-tuned for specific languages on the CoNLL NER datasets (Dutch, Spanish, English, German). Darker blue regions highlight content shared across models—the more overlap, the more efficient storage and transfer becomes. This level of deduplication leads to faster uploads, quicker iterations, and less friction when scaling experimentation.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
</p>
|
23 |
+
|
24 |
<p>
|
25 |
+
XLM-RoBERTa large currently has <a href="https://huggingface.co/models?other=base_model:finetune:FacebookAI/xlm-roberta-large">396 fine-tunes on the Hub</a>. The fine-tunes from the original CoNLL research deduplicate at ~17%, representing a substantial time savings for builders repeatedly pushing new checkpoints and variants.
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
</p>
|
27 |
+
|
28 |
To explore the visualization:
|
29 |
<ul>
|
30 |
<li>
|
|
|
103 |
],
|
104 |
field: "dedupe_factor",
|
105 |
type: "quantitative",
|
106 |
+
scale: { scheme: "blues", domain: [0, 5] },
|
107 |
},
|
108 |
opacity: {
|
109 |
condition: [
|
xorbs.json
CHANGED
The diff for this file is too large to render.
See raw diff
|
|