Update README.md
Browse files
README.md
CHANGED
@@ -95,32 +95,32 @@ print(output)
|
|
95 |
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">49.63</td>
|
96 |
</tr>
|
97 |
<tr>
|
98 |
-
<td style="text-align:left; background-color: #
|
99 |
-
<td style="text-align:center; background-color: #
|
100 |
-
<td style="text-align:center; background-color: #
|
101 |
-
<td style="text-align:center; background-color: #
|
102 |
-
<td style="text-align:center; background-color: #
|
103 |
-
<td style="text-align:center; background-color: #
|
104 |
-
<td style="text-align:center; background-color: #
|
105 |
-
<td style="text-align:center; background-color: #
|
106 |
-
<td style="text-align:center; background-color: #
|
107 |
-
<td style="text-align:center; background-color: #
|
108 |
-
<td style="text-align:center; background-color: #
|
109 |
-
<td style="text-align:center; background-color: #
|
110 |
-
</tr>
|
111 |
-
<tr>
|
112 |
-
<td style="text-align:left; background-color: #
|
113 |
-
<td style="text-align:center; background-color: #
|
114 |
-
<td style="text-align:center; background-color: #
|
115 |
-
<td style="text-align:center; background-color: #
|
116 |
-
<td style="text-align:center; background-color: #
|
117 |
-
<td style="text-align:center; background-color: #
|
118 |
-
<td style="text-align:center; background-color: #
|
119 |
-
<td style="text-align:center; background-color: #
|
120 |
-
<td style="text-align:center; background-color: #
|
121 |
-
<td style="text-align:center; background-color: #
|
122 |
-
<td style="text-align:center; background-color: #
|
123 |
-
<td style="text-align:center; background-color: #
|
124 |
</tr>
|
125 |
|
126 |
<tr>
|
@@ -137,9 +137,9 @@ print(output)
|
|
137 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.18</td>
|
138 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">58.05</td>
|
139 |
</tr>
|
140 |
-
|
141 |
</tbody></table>
|
142 |
|
|
|
143 |
**Model Architecture:**
|
144 |
Granite-3.3-2B-Base is based on a decoder-only dense transformer architecture. Core components of this architecture are: GQA and RoPE, MLP with SwiGLU, RMSNorm, and shared input/output embeddings.
|
145 |
<table>
|
@@ -152,68 +152,68 @@ Granite-3.3-2B-Base is based on a decoder-only dense transformer architecture. C
|
|
152 |
<tbody>
|
153 |
<tr>
|
154 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Embedding size</td>
|
155 |
-
<td style="text-align:center; background-color: #
|
156 |
-
<td style="text-align:center; background-color: #
|
157 |
</tr>
|
158 |
<tr>
|
159 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of layers</td>
|
160 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">40</td>
|
161 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">40</td>
|
|
|
162 |
</tr>
|
163 |
<tr>
|
164 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Attention head size</td>
|
165 |
-
<td style="text-align:center; background-color: #
|
166 |
-
<td style="text-align:center; background-color: #
|
167 |
</tr>
|
168 |
<tr>
|
169 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of attention heads</td>
|
170 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">32</td>
|
171 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">32</td>
|
|
|
172 |
</tr>
|
173 |
<tr>
|
174 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of KV heads</td>
|
175 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">8</td>
|
176 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">8</td>
|
|
|
177 |
</tr>
|
178 |
<tr>
|
179 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">MLP hidden size</td>
|
180 |
-
<td style="text-align:center; background-color: #
|
181 |
-
<td style="text-align:center; background-color: #
|
182 |
</tr>
|
183 |
<tr>
|
184 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">MLP activation</td>
|
185 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">SwiGLU</td>
|
186 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">SwiGLU</td>
|
|
|
187 |
</tr>
|
188 |
<tr>
|
189 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Initialization std</td>
|
190 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">0.1</td>
|
191 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">0.1</td>
|
|
|
192 |
</tr>
|
193 |
<tr>
|
194 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Sequence length</td>
|
195 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">128K</td>
|
196 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">128K</td>
|
|
|
197 |
</tr>
|
198 |
<tr>
|
199 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Position embedding</td>
|
200 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">RoPE</td>
|
201 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">RoPE</td>
|
|
|
202 |
</tr>
|
203 |
<tr>
|
204 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Parameters</td>
|
205 |
-
<td style="text-align:center; background-color: #
|
206 |
-
<td style="text-align:center; background-color: #
|
207 |
</tr>
|
208 |
<tr>
|
209 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Active parameters</td>
|
210 |
-
<td style="text-align:center; background-color: #
|
211 |
-
<td style="text-align:center; background-color: #
|
212 |
</tr>
|
213 |
<tr>
|
214 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Training tokens</td>
|
215 |
-
<td style="text-align:center; background-color: #FFFFFF; color: black;">12T</td>
|
216 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">12T</td>
|
|
|
217 |
</tr>
|
218 |
</tbody></table>
|
219 |
|
|
|
95 |
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">49.63</td>
|
96 |
</tr>
|
97 |
<tr>
|
98 |
+
<td style="text-align:left; background-color: #DAE8FF; color: black;"><b>Granite-3.3-2B-Base</b></td>
|
99 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 47.49 </td>
|
100 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 73.2 </td>
|
101 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 54.33 </td>
|
102 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 40.83 </td>
|
103 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 70.4 </td>
|
104 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 50.0 </td>
|
105 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;"> 32.552 </td>
|
106 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">24.36</td>
|
107 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">38.78</td>
|
108 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">63.22</td>
|
109 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">49.52</td>
|
110 |
+
</tr>
|
111 |
+
<tr>
|
112 |
+
<td style="text-align:left; background-color: #FFFFFF; color: #2D2D2D;">Granite-3.1-8B-Base</td>
|
113 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">53.51</td>
|
114 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">81.4</td>
|
115 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">64.28</td>
|
116 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">51.27</td>
|
117 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">76.2</td>
|
118 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">70.5</td>
|
119 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">45.87</td>
|
120 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">35.97</td>
|
121 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">48.99</td>
|
122 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">78.33</td>
|
123 |
+
<td style="text-align:center; background-color: #FFFFFF; color: #2D2D2D;">60.63</td>
|
124 |
</tr>
|
125 |
|
126 |
<tr>
|
|
|
137 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">78.18</td>
|
138 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">58.05</td>
|
139 |
</tr>
|
|
|
140 |
</tbody></table>
|
141 |
|
142 |
+
|
143 |
**Model Architecture:**
|
144 |
Granite-3.3-2B-Base is based on a decoder-only dense transformer architecture. Core components of this architecture are: GQA and RoPE, MLP with SwiGLU, RMSNorm, and shared input/output embeddings.
|
145 |
<table>
|
|
|
152 |
<tbody>
|
153 |
<tr>
|
154 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Embedding size</td>
|
155 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">2048</td>
|
156 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">4096</td>
|
157 |
</tr>
|
158 |
<tr>
|
159 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of layers</td>
|
|
|
160 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">40</td>
|
161 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">40</td>
|
162 |
</tr>
|
163 |
<tr>
|
164 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Attention head size</td>
|
165 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">64</td>
|
166 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">128</td>
|
167 |
</tr>
|
168 |
<tr>
|
169 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of attention heads</td>
|
|
|
170 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">32</td>
|
171 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">32</td>
|
172 |
</tr>
|
173 |
<tr>
|
174 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Number of KV heads</td>
|
|
|
175 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">8</td>
|
176 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">8</td>
|
177 |
</tr>
|
178 |
<tr>
|
179 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">MLP hidden size</td>
|
180 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">8192</td>
|
181 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">12800</td>
|
182 |
</tr>
|
183 |
<tr>
|
184 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">MLP activation</td>
|
|
|
185 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">SwiGLU</td>
|
186 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">SwiGLU</td>
|
187 |
</tr>
|
188 |
<tr>
|
189 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Initialization std</td>
|
|
|
190 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">0.1</td>
|
191 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">0.1</td>
|
192 |
</tr>
|
193 |
<tr>
|
194 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Sequence length</td>
|
|
|
195 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">128K</td>
|
196 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">128K</td>
|
197 |
</tr>
|
198 |
<tr>
|
199 |
<td style="text-align:left; background-color: #FFFFFF; color: black;">Position embedding</td>
|
|
|
200 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">RoPE</td>
|
201 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">RoPE</td>
|
202 |
</tr>
|
203 |
<tr>
|
204 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Parameters</td>
|
205 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">2.5B</td>
|
206 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">8.1B</td>
|
207 |
</tr>
|
208 |
<tr>
|
209 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Active parameters</td>
|
210 |
+
<td style="text-align:center; background-color: #DAE8FF; color: black;">2.5B</td>
|
211 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">8.1B</td>
|
212 |
</tr>
|
213 |
<tr>
|
214 |
<td style="text-align:left; background-color: #FFFFFF; color: black;"># Training tokens</td>
|
|
|
215 |
<td style="text-align:center; background-color: #DAE8FF; color: black;">12T</td>
|
216 |
+
<td style="text-align:center; background-color: #FFFFFF; color: black;">12T</td>
|
217 |
</tr>
|
218 |
</tbody></table>
|
219 |
|