Information

OpenAssistant-Llama-30B-4-bit working with GPTQ versions used in Oobabooga's Text Generation Webui and KoboldAI.

This was made using Open Assistant's native fine-tune of Llama 30b on their dataset.

What's included

GPTQ: 2 quantized versions. One quantized --true-sequential and act-order optimizations, and the other was quantized using --true-sequential --groupsize 128 optimizations

GGML: 3 quantized versions. One quantized using q4_1, another one was quantized using q5_0, and the last one was quantized using q5_1.

Update 05.27.2023

Updated the ggml quantizations to be compatible with the latest version of llamacpp (again).

Update 04.29.2023

Updated to the latest fine-tune by Open Assistant oasst-sft-7-llama-30b-xor.

GPU/GPTQ Usage

To use with your GPU using GPTQ pick one of the .safetensors along with all of the .jsons and .model files.

Oobabooga: If you require further instruction, see here and here

KoboldAI: If you require further instruction, see here

CPU/GGML Usage

To use your CPU using GGML(Llamacpp) you only need the single .bin ggml file.

Oobabooga: If you require further instruction, see here

KoboldAI: If you require further instruction, see here

Benchmarks

--true-sequential --act-order

Wikitext2: 4.964076519012451 Ptb-New: 9.641128540039062 C4-New: 7.203001022338867 Note: This version does not use --groupsize 128, therefore evaluations are minimally higher. However, this version allows fitting the whole model at full context using only 24GB VRAM.

--true-sequential --groupsize 128

Wikitext2: 4.641914367675781 Ptb-New: 9.117929458618164 C4-New: 6.867942810058594 Note: This version uses --groupsize 128, resulting in better evaluations. However, it consumes more VRAM.