rAIfle/Sloppier-Wingman-Alternative-8x7B-exl2-rpcal

Quantized using 200 samples of 8192 tokens from an RP-oriented PIPPA dataset.

Branches:

• main -- measurement.json
• 3.5b6h -- 3.5bpw, 6bit lm_head
• 6b6h -- 6bpw, 6bit lm_head

Requires ExllamaV2 version 0.0.12 and up.

Original model link: rAIfle/Sloppier-Wingman-Alternative-8x7B-hf

Original model README below.

---

Sloppier-Wingman-Alternative-8x7B-hf

Alternative to rAIfle/Sloppy-Wingman-8x7B-hf. Second part of the merge has a bit of difference compared to the other one. I, personally, still prefer ChatML on this one, but Alpaca and/or Mistral-formats ought to work regardless.

models:
  - model: mistralai/Mixtral-8x7B-v0.1+retrieval-bar/Mixtral-8x7B-v0.1_case-briefs
    parameters:
      weight: 0.33
  - model: mistralai/Mixtral-8x7B-v0.1+wandb/Mixtral-8x7b-Remixtral
    parameters:
      weight: 0.33
merge_method: task_arithmetic
base_model: mistralai/Mixtral-8x7B-v0.1
dtype: float16

and

models:
  - model: mistralai/Mixtral-8x7B-Instruct-v0.1+/ai/LLM/tmp/pefts/daybreak-peft/mixtral-8x7b
    parameters:
      weight: 0.85
  - model: mistralai/Mixtral-8x7B-Instruct-v0.1+SeanWu25/Mixtral_8x7b_Medicine
    parameters:
      weight: 0.33
  - model: notstoic/Nous-Hermes-2-Mixtruct-v0.1-8x7B-DPO-DARE_TIES
    parameters:
      weight: 0.25
merge_method: task_arithmetic
base_model: mistralai/Mixtral-8x7B-Instruct-v0.1
dtype: float16

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• ./02.5-pal-instruct
• ./01-pal-base

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: ./01-pal-base
  - model: ./02.5-pal-instruct
merge_method: slerp
base_model: ./01-pal-base
parameters:
  t:
    - value: 0.66
dtype: float16