AC-723 commited on
Commit
2f032d5
·
verified ·
1 Parent(s): 2ab638d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +46 -0
README.md CHANGED
@@ -20,6 +20,8 @@ extra_gated_button_content: Acknowledge license
20
  Nile-Chat is a family of open instruction-tuned models for Egyptian dialect, developed to handle both scripts commonly used in Egypt: Arabic script and Latin-based Arabizi. As part of the [Jais](https://arxiv.org/abs/2308.16149) project for standard Arabic and its extensions to dialectal Arabic, Nile-Chat is designed to support natural language generation in a way that reflects the script-diverse nature of Egyptian communication. These models are effective for a variety of tasks including question answering, translation and transliteration. Their range of sizes ensures accessibility, from lightweight personal deployments to more powerful setups, enabling broader use of AI technologies for Egyptian Arabic speakers. The family includes two versions:
21
 
22
  * [Nile-Chat-4B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-4B): A compact 4B parameter model that balances efficiency and fluency, well-suited for generating Egyptian Arabic in both Arabic and Latin scripts.
 
 
23
  * [Nile-Chat-12B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B): A dense 12B parameter model providing high-capacity generation capabilities, ideal for complex, multi-turn interactions and nuanced understanding in both Egyptian scripts.
24
 
25
  **Paper:** [Nile-Chat: Egyptian Language Models for Arabic and Latin Scripts](https://huggingface.co/papers/2507.04569)
@@ -200,6 +202,16 @@ The models were compared against a collection of existing open-source Arabic mod
200
  <td>57.85</td>
201
  <td>50.25</td><td>68.56</td><td>55.92</td><td>67.30</td><td>61.87</td><td>40.94</td><td>42.10</td><td>46.02</td><td>87.65</td>
202
  </tr>
 
 
 
 
 
 
 
 
 
 
203
  <tr>
204
  <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank"><strong>Nile-Chat-12B</strong></a></td>
205
  <td>64.11</td>
@@ -293,6 +305,16 @@ The models were compared against a collection of existing open-source Arabic mod
293
  <td>51.38</td>
294
  <td>50.55</td><td>65.32</td><td>60.62</td><td>37.36</td><td>43.06</td>
295
  </tr>
 
 
 
 
 
 
 
 
 
 
296
  <tr>
297
  <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank"><strong>Nile-Chat-12B</strong></a></td>
298
  <td>53.88</td>
@@ -490,6 +512,30 @@ The models were compared against a collection of existing open-source Arabic mod
490
  <td>80.44</td>
491
  <td>89.59</td>
492
  </tr>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
493
  <tr>
494
  <td><strong><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank">Nile-Chat-12B</a></td>
495
  <td>40.53</td>
 
20
  Nile-Chat is a family of open instruction-tuned models for Egyptian dialect, developed to handle both scripts commonly used in Egypt: Arabic script and Latin-based Arabizi. As part of the [Jais](https://arxiv.org/abs/2308.16149) project for standard Arabic and its extensions to dialectal Arabic, Nile-Chat is designed to support natural language generation in a way that reflects the script-diverse nature of Egyptian communication. These models are effective for a variety of tasks including question answering, translation and transliteration. Their range of sizes ensures accessibility, from lightweight personal deployments to more powerful setups, enabling broader use of AI technologies for Egyptian Arabic speakers. The family includes two versions:
21
 
22
  * [Nile-Chat-4B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-4B): A compact 4B parameter model that balances efficiency and fluency, well-suited for generating Egyptian Arabic in both Arabic and Latin scripts.
23
+ * [Nile-Chat-2x4B-A6B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-2x4B-A6B): A 2x4B MoE parameter model with 6B Activated (soft-merging/1 Expert per script) that balances efficiency and fluency, well-suited for generating Egyptian Arabic in both Arabic and Latin scripts.
24
+ * [Nile-Chat-3x4B-A6B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-3x4B-A6B): A 3x4B MoE parameter model with 6B Activated (2 experts activated) providing high-capacity generation in Egyptian Arabic in both Arabic and Latin scripts and English.
25
  * [Nile-Chat-12B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B): A dense 12B parameter model providing high-capacity generation capabilities, ideal for complex, multi-turn interactions and nuanced understanding in both Egyptian scripts.
26
 
27
  **Paper:** [Nile-Chat: Egyptian Language Models for Arabic and Latin Scripts](https://huggingface.co/papers/2507.04569)
 
202
  <td>57.85</td>
203
  <td>50.25</td><td>68.56</td><td>55.92</td><td>67.30</td><td>61.87</td><td>40.94</td><td>42.10</td><td>46.02</td><td>87.65</td>
204
  </tr>
205
+ <tr>
206
+ <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-2x4B-A6B" target="_blank"><strong>Nile-Chat-2x4B-A6B</strong></a></td>
207
+ <td>60.89</td>
208
+ <td>52.05</td><td>73.89</td><td>59.69</td><td>68.67</td><td>62.26</td><td>41.61</td><td>44.07</td><td>51.14</td><td>94.58</td>
209
+ </tr>
210
+ <tr>
211
+ <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-3x4B-A6B" target="_blank"><strong>Nile-Chat-3x4B-A6B</strong></a></td>
212
+ <td>60.25</td>
213
+ <td>52.13</td><td>75.44</td><td>59.3</td><td>69.27</td><td>57.91</td><td>41.16</td><td>44.59</td><td>48.3</td><td>94.18</td>
214
+ </tr>
215
  <tr>
216
  <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank"><strong>Nile-Chat-12B</strong></a></td>
217
  <td>64.11</td>
 
305
  <td>51.38</td>
306
  <td>50.55</td><td>65.32</td><td>60.62</td><td>37.36</td><td>43.06</td>
307
  </tr>
308
+ <tr>
309
+ <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-2x4B-A6B" target="_blank"><strong>Nile-Chat-2x4B-A6B</strong></a></td>
310
+ <td>54.12</td>
311
+ <td>55.49</td><td>68</td><td>61.33</td><td>40.24</td><td>45.56</td>
312
+ </tr>
313
+ <tr>
314
+ <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-3x4B-A6B" target="_blank"><strong>Nile-Chat-3x4B-A6B</strong></a></td>
315
+ <td>52.26</td>
316
+ <td>55</td><td>66.68</td><td>56.42</td><td>40.44</td><td>42.78</td>
317
+ </tr>
318
  <tr>
319
  <td><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank"><strong>Nile-Chat-12B</strong></a></td>
320
  <td>53.88</td>
 
512
  <td>80.44</td>
513
  <td>89.59</td>
514
  </tr>
515
+ <tr>
516
+ <td><strong><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-2x4B-A6B" target="_blank">Nile-Chat-2x4B-A6B</a></td>
517
+ <td>41.98</td>
518
+ <td>61.59</td>
519
+ <td>86.11</td>
520
+ <td>33.4</td>
521
+ <td>53.71</td>
522
+ <td>76.78</td>
523
+ <td>57.75</td>
524
+ <td>83.89</td>
525
+ <td>91.05</td>
526
+ </tr>
527
+ <tr>
528
+ <td><strong><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-3x4B-A6B" target="_blank">Nile-Chat-3x4B-A6B</a></td>
529
+ <td>42.43</td>
530
+ <td>61.9</td>
531
+ <td>86.26</td>
532
+ <td>34.56</td>
533
+ <td>55.37</td>
534
+ <td>76.97</td>
535
+ <td>57.79</td>
536
+ <td>83.97</td>
537
+ <td>91.13</td>
538
+ </tr>
539
  <tr>
540
  <td><strong><a href="https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B" target="_blank">Nile-Chat-12B</a></td>
541
  <td>40.53</td>