Commit
·
4795ae3
1
Parent(s):
dc9071c
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
inference: false
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
tags:
|
| 7 |
+
- llama
|
| 8 |
+
- llama-2
|
| 9 |
+
license: agpl-3.0
|
| 10 |
+
---
|
| 11 |
+
# CalliopeDS-v2-L2-13B-exl2
|
| 12 |
+
|
| 13 |
+
Exllama v2 quant of [Doctor-Shotgun/CalliopeDS-v2-L2-13B](https://huggingface.co/Doctor-Shotgun/CalliopeDS-v2-L2-13B)
|
| 14 |
+
|
| 15 |
+
Branches:
|
| 16 |
+
- main: measurement.json calculated at 2048 token calibration rows on PIPPA
|
| 17 |
+
- 4.0bpw-h6: 4 decoder bits per weight, 6 head bits
|
| 18 |
+
- ideal for 12gb GPUs, or 16gb GPUs with NTK extended context or CFG
|
| 19 |
+
- 6.0bpw-h6: 6 decoder bits per weight, 6 head bits
|
| 20 |
+
- ideal for 16gb GPUs, or 24gb GPUs with NTK extended context or CFG
|
| 21 |
+
- 8bit-32g-h8: all tensors 8bit 32g, 8 head bits
|
| 22 |
+
- experimental quant, this is with exllamav2 monkeypatched to quantize all tensors to 8bit 32g
|
| 23 |
+
- similar in size to old GPTQ 8bit no groupsize, recommend 24gb GPU
|