Downloads · 30 days
10
2% of all-time downloads
Yhyu13/oasst-rlhf-2-llama-30b-7k-steps-gptq-4bit
oasst-rlhf-2-llama-30b-7k-steps-gptq-4bit is a text generation model from Yhyu13. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
GPTQ 4-bit no actor version for compatibility that works in textgen-webui
Downloads · 30 days
10
2% of all-time downloads
All-time downloads
412
Public
Repo size
17.5 GB
Likes
3
Public
Click a slice to open those files.
.safetensors17.5 GB · 100%
From the Hugging Face model README
GPTQ 4-bit no actor version for compatibility that works in textgen-webui
Generated by using scripts from https://gitee.com/yhyu13/llama_-tools
Merged weights: https://huggingface.co/Yhyu13/oasst-rlhf-2-llama-30b-7k-steps-hf
Converted LLaMA weights: https://huggingface.co/Yhyu13/llama-30B-hf-openassitant
Delta weights: https://huggingface.co/OpenAssistant/oasst-rlhf-2-llama-30b-7k-steps-xor
OA has done a great jobs in RLHF their pre-trained weights. I must say it is tuned to spit out CoT step by step thinking without you actively prompting it to do so, which is a feature that we observe on ChatGPT and GPT-4.
But note, it still fails at logical paradox tasks such as era of time and bird shot. But none of the LLaMA based models or any available models other than GPT-4 and Claude+ can correct answer paradox questions anyway. So OA rlhf is expected to fail at these tasks, but I do like the RLHF-ed tone which make OA's response sounds professional and proficient.