Update README.md
#4
by
MaziyarPanahi
- opened
README.md
CHANGED
|
@@ -8,8 +8,26 @@ tags:
|
|
| 8 |
- biology
|
| 9 |
- medical
|
| 10 |
- MoE
|
|
|
|
| 11 |
---
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
## How to use it
|
| 14 |
|
| 15 |
```python
|
|
@@ -25,7 +43,6 @@ tokenizer = AutoTokenizer.from_pretrained("MaziyarPanahi/Bioxtral-4x7B-v0.1")
|
|
| 25 |
model = AutoModelForCausalLM.from_pretrained("MaziyarPanahi/Bioxtral-4x7B-v0.1")
|
| 26 |
```
|
| 27 |
|
| 28 |
-
|
| 29 |
## Quantized mode
|
| 30 |
|
| 31 |
Here is the list of GGUF models quantized from 2 to 8 bits: https://huggingface.co/MaziyarPanahi/Bioxtral-4x7B-v0.1-GGUF
|
|
@@ -158,18 +175,6 @@ So, 25 - 4 * 2 + 3 = 20.</s>
|
|
| 158 |
|
| 159 |
|
| 160 |
|
| 161 |
-
| Metric | BioMistral-7B | Bioxtral-4x7B-v0.1 |
|
| 162 |
-
|-----------------------------|---------------|--------------------|
|
| 163 |
-
| **Average** | 54.99 | **70.84** |
|
| 164 |
-
| ARC | 54.27 | **68.34** |
|
| 165 |
-
| HellaSwag | 79.09 | **87.27** |
|
| 166 |
-
| TruthfulQA | 51.61 | **68.45** |
|
| 167 |
-
| Winogrande | 73.48 | **82.90** |
|
| 168 |
-
| GSM8K | 0 | **56.63** |
|
| 169 |
-
| Professional Medicine | 55.51 | **67.3** |
|
| 170 |
-
| College Medicine | 58.96 | **61.84** |
|
| 171 |
-
| Medical Genetics | 67.00 | **74.0** |
|
| 172 |
-
|
| 173 |
|
| 174 |
source: https://huggingface.co/datasets/open-llm-leaderboard/details_MaziyarPanahi__Bioxtral-4x7B-v0.1
|
| 175 |
|
|
|
|
| 8 |
- biology
|
| 9 |
- medical
|
| 10 |
- MoE
|
| 11 |
+
- genetic
|
| 12 |
---
|
| 13 |
|
| 14 |
+
## Description
|
| 15 |
+
|
| 16 |
+
This is a MoE of top 4x 7B models including BioMistral-7B. Here is the comparison between the two:
|
| 17 |
+
|
| 18 |
+
| Metric | BioMistral-7B | Bioxtral-4x7B-v0.1 |
|
| 19 |
+
|-----------------------------|---------------|--------------------|
|
| 20 |
+
| **Average** | 54.99 | **70.84** |
|
| 21 |
+
| ARC | 54.27 | **68.34** |
|
| 22 |
+
| HellaSwag | 79.09 | **87.27** |
|
| 23 |
+
| TruthfulQA | 51.61 | **68.45** |
|
| 24 |
+
| Winogrande | 73.48 | **82.90** |
|
| 25 |
+
| GSM8K | 0 | **56.63** |
|
| 26 |
+
| Professional Medicine | 55.51 | **67.3** |
|
| 27 |
+
| College Medicine | 58.96 | **61.84** |
|
| 28 |
+
| Medical Genetics | 67.00 | **74.0** |
|
| 29 |
+
|
| 30 |
+
|
| 31 |
## How to use it
|
| 32 |
|
| 33 |
```python
|
|
|
|
| 43 |
model = AutoModelForCausalLM.from_pretrained("MaziyarPanahi/Bioxtral-4x7B-v0.1")
|
| 44 |
```
|
| 45 |
|
|
|
|
| 46 |
## Quantized mode
|
| 47 |
|
| 48 |
Here is the list of GGUF models quantized from 2 to 8 bits: https://huggingface.co/MaziyarPanahi/Bioxtral-4x7B-v0.1-GGUF
|
|
|
|
| 175 |
|
| 176 |
|
| 177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
|
| 179 |
source: https://huggingface.co/datasets/open-llm-leaderboard/details_MaziyarPanahi__Bioxtral-4x7B-v0.1
|
| 180 |
|