File size: 3,488 Bytes
c961303 033dfe3 3d623bd 556045b c961303 556045b 8d70a1e 556045b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
---
datasets: LeoLM/wikitext-en-de
license: other
license_link: https://llama.meta.com/llama3/license/
---
This is a quantized model of [SKLM Llama-3 70B Instruct](https://huggingface.co/VAGOsolutions/Llama-3-SauerkrautLM-70b-Instruct) using GPTQ developed by [IST Austria](https://ist.ac.at/en/research/alistarh-group/)
using the following configuration:
- 4bit (8bit will follow)
- Act order: True
- Group size: 128
## Usage
Install **vLLM** and
run the [server](https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html#openai-compatible-server):
```
python -m vllm.entrypoints.openai.api_server --model cortecs/Llama-3-SauerkrautLM-70b-Instruct-GPTQ
```
Access the model:
```
curl http://localhost:8000/v1/completions -H "Content-Type: application/json" -d ' {
"model": "cortecs/Llama-3-SauerkrautLM-70b-Instruct-GPTQ",
"prompt": "Berlin ist eine"
} '
```
## Evaluations
| __English__ | __SKLM Llama-3 70B Instruct__ | __SKLM Llama-3 70B Instruct GPTQ__ | __SKLM Mixtral Instruct__ |
|:--------------|:--------------------------------|:-------------------------------------|:----------------------------|
| Avg. | 78.17 | 76.72 | 73.47 |
| ARC | 74.5 | 73.0 | 71.7 |
| Hellaswag | 79.2 | 78.0 | 77.4 |
| MMLU | 80.8 | 79.15 | 71.31 |
| | | | |
| __German__ | __SKLM Llama-3 70B Instruct__ | __SKLM Llama-3 70B Instruct GPTQ__ | __SKLM Mixtral Instruct__ |
| Avg. | 70.83 | 69.13 | 66.43 |
| ARC_de | 66.7 | 65.9 | 62.7 |
| Hellaswag_de | 70.8 | 68.8 | 72.9 |
| MMLU_de | 75.0 | 72.7 | 63.7 |
| | | | |
| __Safety__ | __SKLM Llama-3 70B Instruct__ | __SKLM Llama-3 70B Instruct GPTQ__ | __SKLM Mixtral Instruct__ |
| Avg. | 65.86 | 65.94 | 64.18 |
| RealToxicityPrompts | 97.6 | 98.4 | 93.2 |
| TruthfulQA | 67.07 | 65.56 | 65.84 |
| CrowS | 32.92 | 33.87 | 33.51 |
Take with caution. We did not check for data contamination.
Evaluation was done using [Eval. Harness](https://github.com/EleutherAI/lm-evaluation-harness) using `limit=1000` for big datasets.
## Performance
| | requests/s | tokens/s |
|:--------------|-------------:|-----------:|
| NVIDIA L40Sx2 | 2.19 | 1044.76 |
|