--- license: other datasets: wikitext license_link: https://llama.meta.com/llama3/license/ --- This is a quantized model of [Meta-Llama-3-70B-Instruct.yaml](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct.yaml) using GPTQ developed by [IST Austria](https://ist.ac.at/en/research/alistarh-group/) using the following configuration: - 4bit - 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/Meta-Llama-3-70B-Instruct-GPTQ ``` Access the model: ``` curl http://localhost:8000/v1/completions -H "Content-Type: application/json" -d ' { "model": "cortecs/Meta-Llama-3-70B-Instruct-GPTQ", "prompt": "San Francisco is a" } ' ``` ## Evaluations | __English__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | |:--------------|:-----------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------------| | Avg. | 76.19 | 76.16 | 75.14 | | ARC | 71.6 | 71.4 | 70.7 | | Hellaswag | 77.3 | 77.1 | 76.4 | | MMLU | 79.66 | 79.98 | 78.33 | | | | | | | __French__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | | Avg. | 70.97 | 71.03 | 70.27 | | ARC_fr | 65.0 | 65.3 | 64.7 | | Hellaswag_fr | 72.4 | 72.4 | 71.4 | | MMLU_fr | 75.5 | 75.4 | 74.7 | | | | | | | __German__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | | Avg. | 68.43 | 68.37 | 66.93 | | ARC_de | 64.2 | 64.3 | 62.6 | | Hellaswag_de | 67.8 | 67.7 | 66.7 | | MMLU_de | 73.3 | 73.1 | 71.5 | | | | | | | __Italian__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | | Avg. | 70.17 | 70.43 | 68.63 | | ARC_it | 64.0 | 64.3 | 62.1 | | Hellaswag_it | 72.6 | 72.4 | 71.0 | | MMLU_it | 73.9 | 74.6 | 72.8 | | | | | | | __Safety__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | | Avg. | 64.28 | 64.17 | 63.64 | | RealToxicityPrompts | 97.9 | 97.8 | 98.1 | | TruthfulQA | 61.91 | 61.67 | 59.91 | | CrowS | 33.04 | 33.04 | 32.92 | | | | | | | __Spanish__ | __[Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)__ | __[Meta-Llama-3-70B-Instruct-GPTQ-8b](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ-8b)__ | __[Meta-Llama-3-70B-Instruct-GPTQ](https://huggingface.co/cortecs/Meta-Llama-3-70B-Instruct-GPTQ)__ | | Avg. | 72.5 | 72.7 | 71.3 | | ARC_es | 66.7 | 66.9 | 65.7 | | Hellaswag_es | 75.8 | 75.9 | 74 | | MMLU_es | 75 | 75.3 | 74.2 | We did not check for data contamination. Evaluation was done using [Eval. Harness](https://github.com/EleutherAI/lm-evaluation-harness) using `limit=1000`. ## Performance | | requests/s | tokens/s | |:--------------|-------------:|-----------:| | NVIDIA L40Sx2 | 2 | 951.28 | Performance measured on [cortecs inference](https://cortecs.ai).