3.2bpw/h6 exl2 quantization of Undi95/Miqu-MS-70B using default exllamav2 calibration dataset.

Fits in 32GB VRAM with 32k+ context (Q4cache)

ORIGINAL CARD:

Miqu-MS-70B

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

The new MODEL STOCK merge method was used, see below for more information!

Feedback on this model is greatly appreciated! I hope this new merge method will be able to fill some hole Miqu have.

Others quant

Thank you all!

Prompt format

Since it was made with model using different prompt format, the following should work.

Alpaca

### Instruction:
{system prompt}

### Input:
{prompt}

### Response:
{output}

Mistral

[INST] {prompt} [/INST]

Vicuna

SYSTEM: <ANY SYSTEM CONTEXT>
USER: 
ASSISTANT:

Merge Details

Merge Method

This model was merged using the Model Stock merge method using 152334H/miqu-1-70b-sf as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: NeverSleep/MiquMaid-v2-70B
  - model: sophosympatheia/Midnight-Miqu-70B-v1.0
  - model: migtissera/Tess-70B-v1.6
  - model: 152334H/miqu-1-70b-sf
merge_method: model_stock
base_model: 152334H/miqu-1-70b-sf
dtype: bfloat16

Support

If you want to support me, you can here.

JayhC
/

Miqu-MS-70B-3.2bpw-h6-exl2