|
--- |
|
license: llama2 |
|
--- |
|
EXL2 quants of alpindale/goliath-120b (https://huggingface.co/alpindale/goliath-120b), to be used on exllamav2. |
|
|
|
Calibration dataset is a cleaned, fixed pippa RP dataset, which does affect the results (in favor) for RP usage. You can find the calibration dataset [here.](https://huggingface.co/datasets/royallab/PIPPA-cleaned) |
|
|
|
I've added a measurement.json file on the main branch if you want to do your own quants. |
|
|
|
[4.85bpw](https://huggingface.co/Panchovix/goliath-120b-exl2-rpcal/tree/4.85bpw) |
|
|
|
[4.5bpw](https://huggingface.co/Panchovix/goliath-120b-exl2-rpcal/tree/4.5bpw) |
|
|
|
[3bpw](https://huggingface.co/Panchovix/goliath-120b-exl2-rpcal/tree/3bpw) |
|
|
|
# Original model card |
|
|
|
# Goliath 120B |
|
|
|
An auto-regressive causal LM created by combining 2x finetuned [Llama-2 70B](https://huggingface.co/meta-llama/llama-2-70b-hf) into one. |
|
|
|
Please check out the quantized formats provided by [@TheBloke](https:///huggingface.co/TheBloke) and [@Panchovix](https://huggingface.co/Panchovix): |
|
|
|
- [GGUF](https://huggingface.co/TheBloke/goliath-120b-GGUF) (llama.cpp) |
|
- [GPTQ](https://huggingface.co/TheBloke/goliath-120b-GPTQ) (KoboldAI, TGW, Aphrodite) |
|
- [AWQ](https://huggingface.co/TheBloke/goliath-120b-AWQ) (TGW, Aphrodite, vLLM) |
|
- [Exllamav2](https://huggingface.co/Panchovix/goliath-120b-exl2) (TGW, KoboldAI) |
|
|
|
# Prompting Format |
|
|
|
Both Vicuna and Alpaca will work, but due the initial and final layers belonging primarily to Xwin, I expect Vicuna to work the best. |
|
|
|
# Merge process |
|
|
|
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). |
|
|
|
The layer ranges used are as follows: |
|
|
|
```yaml |
|
- range 0, 16 |
|
Xwin |
|
- range 8, 24 |
|
Euryale |
|
- range 17, 32 |
|
Xwin |
|
- range 25, 40 |
|
Euryale |
|
- range 33, 48 |
|
Xwin |
|
- range 41, 56 |
|
Euryale |
|
- range 49, 64 |
|
Xwin |
|
- range 57, 72 |
|
Euryale |
|
- range 65, 80 |
|
Xwin |
|
``` |
|
|
|
# Screenshots |
|
|
|
 |
|
|
|
# Benchmarks |
|
Coming soon. |
|
|
|
# Acknowledgements |
|
Credits goes to [@chargoddard](https://huggingface.co/chargoddard) for developing the framework used to merge the model - [mergekit](https://github.com/cg123/mergekit). |
|
|
|
Special thanks to [@Undi95](https://huggingface.co/Undi95) for helping with the merge ratios. |
