|
|
--- |
|
|
base_model: EssentialAI/rnj-1-instruct |
|
|
--- |
|
|
|
|
|
[EXL3](https://github.com/turboderp-org/exllamav3) quantization of [rnj-1-instruct](https://huggingface.co/EssentialAI/rnj-1-instruct), 6 bits per weight. |
|
|
|
|
|
### HumanEval (argmax) |
|
|
|
|
|
| Model | Q4 | Q6 | Q8 | FP16 | |
|
|
| ---------------------------------------------------------------------------------------- | ---- | ---- | ---- | ---- | |
|
|
| [rnj-1-instruct-exl3-4bpw](https://huggingface.co/isogen/rnj-1-instruct-exl3-4bpw) | 88.4 | 89.0 | 87.8 | 87.2 | |
|
|
| [rnj-1-instruct-exl3-6bpw](https://huggingface.co/isogen/rnj-1-instruct-exl3-6bpw) | 87.2 | 86.6 | 86.6 | 86.0 | |
|
|
| [rnj-1-instruct-exl3-8bpw-h8](https://huggingface.co/isogen/rnj-1-instruct-exl3-8bpw-h8) | 86.6 | 87.2 | 86.6 | 86.6 | |
|
|
|
