| base_model: | |
| - mistralai/Mixtral-8x7B-Instruct-v0.1 | |
| - NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO | |
| tags: | |
| - mergekit | |
| - merge | |
| # lesser-hermes | |
| ## Why? | |
| Hermes is very, um, Hermes-y. I wanted to dilute it so I could use it as an ingredient | |
| for other things. Sampling Hermes is a pain in the ass, it either sounds super model-esque | |
| or it loses all instructability. Hence, dilution back to the root. | |
| This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). We've been using this as one of the experimental ingredients to help stabilize the monkey-typewriter merges, and it's kinda okay at that. | |
| Note that modern mergekit handles MoE just fine, now. But back in the day it did a horrible job and only the fork worked properly. | |
| ## Merge Details | |
| ### Merge Method | |
| This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) as a base. | |
| ### Models Merged | |
| The following models were included in the merge: | |
| * [NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO](https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO) | |
| ### Configuration | |
| The following YAML configuration was used to produce this model: | |
| ```yaml | |
| models: | |
| # dont bagel me bro | |
| - model: NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO | |
| parameters: | |
| density: 0.25 | |
| weight: 0.3 | |
| - model: mistralai/Mixtral-8x7B-Instruct-v0.1 | |
| parameters: | |
| density: 0.5 | |
| weight: 1 | |
| merge_method: dare_ties | |
| base_model: mistralai/Mixtral-8x7B-Instruct-v0.1 | |
| parameters: | |
| #normalize: false | |
| #int8_mask: true | |
| dtype: bfloat16 | |
| ``` | |