metadata
license: apache-2.0
datasets:
- unalignment/toxic-dpo-v0.1
- adamo1139/rawrr_v2
- adamo1139/AEZAKMI_v3-6
tags:
- lora
New Yi AEZAKMI based on the newer Yi-34B-200K that has better needle in a haystack perf. Will start uploading Fp16 safetensors soon.
To use those LoRAs, download new Yi-34B-200k base, merge lora 1, unload, merge lora 2, unload, merge lora 3 and unload. Maybe you can merge on top of each other without unload, I don't know.
chatml format, should work various system prompts well due to being trained on SysChat dataset. It should be uncensored. I can publish training scripts if someone wants them, all datasets used are open source.