Update README.md
Browse files
README.md
CHANGED
@@ -15,8 +15,10 @@ inference: false
|
|
15 |
|
16 |
Converted from the XORs weights from PygmalionAI's release https://huggingface.co/PygmalionAI/metharme-7b
|
17 |
|
18 |
-
Quantized for KoboldAI use using https://github.com/0cc4m/GPTQ-for-LLaMa
|
19 |
-
|
|
|
|
|
20 |
|
21 |
Metharme 7B is an instruct model based on Meta's LLaMA-7B.
|
22 |
|
@@ -85,11 +87,66 @@ As such, it was **not** fine-tuned to be safe and harmless: the base model _and_
|
|
85 |
|
86 |
<p><strong><font size="5">Benchmarks</font></strong></p>
|
87 |
|
88 |
-
<p><strong><font size="4">This Model:</font> <br><font size="4">4 Bit</font></strong></p>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
<strong>Wikitext2</strong>: 6.997134685516357
|
91 |
|
92 |
<strong>Ptb-New</strong>: 58.87525177001953
|
93 |
|
94 |
<strong>C4-New</strong>: 8.824191093444824
|
95 |
-
<hr>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
Converted from the XORs weights from PygmalionAI's release https://huggingface.co/PygmalionAI/metharme-7b
|
17 |
|
18 |
+
Quantized for KoboldAI use using https://github.com/0cc4m/GPTQ-for-LLaMa
|
19 |
+
|
20 |
+
I created several quantized variations of this model and believe this variation to be "best."
|
21 |
+
<b>Benchmark scores at the bottom of readme.</b>
|
22 |
|
23 |
Metharme 7B is an instruct model based on Meta's LLaMA-7B.
|
24 |
|
|
|
87 |
|
88 |
<p><strong><font size="5">Benchmarks</font></strong></p>
|
89 |
|
90 |
+
<p><strong><font size="4">This Model:</font> <br><font size="4">4 Bit --Act-order</font></strong></p>
|
91 |
+
<strong>Wikitext2</strong>: 6.281311511993408
|
92 |
+
|
93 |
+
<strong>Ptb-New</strong>: 46.79158401489258
|
94 |
+
|
95 |
+
<strong>C4-New</strong>: 7.906069755554199
|
96 |
+
<hr>
|
97 |
+
<p><strong><font size="4">4 Bit --True-Sequential --Groupsize 32</font></strong></p>
|
98 |
+
<strong>Wikitext2</strong>: 6.424218654632568
|
99 |
+
|
100 |
+
<strong>Ptb-New</strong>: 48.48588943481445
|
101 |
+
|
102 |
+
<strong>C4-New</strong>: 8.089512825012207
|
103 |
+
<hr>
|
104 |
+
<p><strong><font size="4">4 Bit --Groupsize 32</font></strong></p>
|
105 |
+
<strong>Wikitext2</strong>: 6.902080535888672
|
106 |
+
|
107 |
+
<strong>Ptb-New</strong>: 50.237754821777344
|
108 |
+
|
109 |
+
<strong>C4-New</strong>: 8.081602096557617
|
110 |
+
<hr>
|
111 |
+
<p><strong><font size="4">4 Bit --Groupsize 128</font></strong></p>
|
112 |
+
<strong>Wikitext2</strong>: 10.206376075744629
|
113 |
+
|
114 |
+
<strong>Ptb-New</strong>: 49.00401306152344
|
115 |
|
116 |
+
<strong>C4-New</strong>: 8.198845863342285
|
117 |
+
<hr>
|
118 |
+
<p><strong><font size="4">4 Bit --True-Sequential --Groupsize 128</font></strong></p>
|
119 |
+
<strong>Wikitext2</strong>: 9.475017547607422
|
120 |
+
|
121 |
+
<strong>Ptb-New</strong>: 52.358829498291016
|
122 |
+
|
123 |
+
<strong>C4-New</strong>: 8.340700149536133
|
124 |
+
<hr>
|
125 |
+
<p><strong><font size="4">4 Bit</font></strong></p>
|
126 |
<strong>Wikitext2</strong>: 6.997134685516357
|
127 |
|
128 |
<strong>Ptb-New</strong>: 58.87525177001953
|
129 |
|
130 |
<strong>C4-New</strong>: 8.824191093444824
|
131 |
+
<hr>
|
132 |
+
<p><strong><font size="4">4 Bit --True Sequential</font></strong></p>
|
133 |
+
<strong>Wikitext2</strong>: 7.735969543457031
|
134 |
+
|
135 |
+
<strong>Ptb-New</strong>: 62.92238235473633
|
136 |
+
|
137 |
+
<strong>C4-New</strong>: 8.913650512695312
|
138 |
+
<hr>
|
139 |
+
<p><strong><font size="4">4 Bit --act-order --groupsize 32</font></strong></p>
|
140 |
+
<strong>Wikitext2</strong>: 275425.5
|
141 |
+
|
142 |
+
<strong>Ptb-New</strong>: 267733.25
|
143 |
+
|
144 |
+
<strong>C4-New</strong>: 254506.71875
|
145 |
+
<hr>
|
146 |
+
<p><strong><font size="4">4 Bit --act-order --groupsize 32</font></strong></p>
|
147 |
+
<strong>Wikitext2</strong>: 251321.265625
|
148 |
+
|
149 |
+
<strong>Ptb-New</strong>: 250117.859375
|
150 |
+
|
151 |
+
<strong>C4-New</strong>: 232929.234375
|
152 |
+
<hr>
|