Panchovix commited on
Commit
5c8ee05
1 Parent(s): 414d017

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +57 -1
README.md CHANGED
@@ -9,4 +9,60 @@ Calibration dataset is wikitext.
9
 
10
  [4.5bpw](https://huggingface.co/Panchovix/goliath-120b-exl2/tree/4.5bpw)
11
 
12
- [3bpw](https://huggingface.co/Panchovix/goliath-120b-exl2/tree/3bpw)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
 
10
  [4.5bpw](https://huggingface.co/Panchovix/goliath-120b-exl2/tree/4.5bpw)
11
 
12
+ [3bpw](https://huggingface.co/Panchovix/goliath-120b-exl2/tree/3bpw)
13
+
14
+ # Original Model card
15
+
16
+ # Goliath 120B
17
+
18
+ An auto-regressive causal LM created by combining 2x finetuned [Llama-2 70B](https://huggingface.co/meta-llama/llama-2-70b-hf) into one.
19
+
20
+ Please check out the quantized formats provided by [@TheBloke](https:///huggingface.co/TheBloke) and [@Panchovix](https://huggingface.co/Panchovix):
21
+
22
+ - [GGUF](https://huggingface.co/TheBloke/goliath-120b-GGUF) (llama.cpp)
23
+ - [GPTQ](https://huggingface.co/TheBloke/goliath-120b-GPTQ) (KoboldAI, TGW, Aphrodite)
24
+ - [AWQ](https://huggingface.co/TheBloke/goliath-120b-AWQ) (TGW, Aphrodite, vLLM)
25
+ - [Exllamav2](https://huggingface.co/Panchovix/goliath-120b-exl2) (TGW, KoboldAI)
26
+
27
+ # Prompting Format
28
+
29
+ Both Vicuna and Alpaca will work, but due the initial and final layers belonging primarily to Xwin, I expect Vicuna to work the best.
30
+
31
+ # Merge process
32
+
33
+ The models used in the merge are [Xwin](https://huggingface.co/Xwin-LM/Xwin-LM-70B-V0.1) and [Euryale](https://huggingface.co/Sao10K/Euryale-1.3-L2-70B).
34
+
35
+ The layer ranges used are as follows:
36
+
37
+ ```yaml
38
+ - range 0, 16
39
+ Xwin
40
+ - range 8, 24
41
+ Euryale
42
+ - range 17, 32
43
+ Xwin
44
+ - range 25, 40
45
+ Euryale
46
+ - range 33, 48
47
+ Xwin
48
+ - range 41, 56
49
+ Euryale
50
+ - range 49, 64
51
+ Xwin
52
+ - range 57, 72
53
+ Euryale
54
+ - range 65, 80
55
+ Xwin
56
+ ```
57
+
58
+ # Screenshots
59
+
60
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/635567189c72a7e742f1419c/Cat8_Rimaz6Ni7YhQiiGB.png)
61
+
62
+ # Benchmarks
63
+ Coming soon.
64
+
65
+ # Acknowledgements
66
+ Credits goes to [@chargoddard](https://huggingface.co/chargoddard) for developing the framework used to merge the model - [mergekit](https://github.com/cg123/mergekit).
67
+
68
+ Special thanks to [@Undi95](https://huggingface.co/Undi95) for helping with the merge ratios.