Mixtral-8x7B-Instruct-v0.1-upscaled

This is a frankenmerge of mistralai/Mixtral-8x7B-Instruct-v0.1 created by interleaving layers of itself using mergekit.

Benchmark

The benchmark score of the mt-bench for this model and the original models are as follows:

1-turn

Model	Size	Coding	Extraction	Humanities	Math	Reasoning	Roleplay	STEM	Writing	avg_score
Mixtral-8x7B-Instruct-v0.1	8x7B	5.3	8.5	9.9	6.8	6.0	9.1	9.55	8.9	8.00625
This model	around 8x12B?	6.3	8.4	9.9	5.4	7.7	9.2	9.75	9.8	8.30625

2-turn

Model	Size	Coding	Extraction	Humanities	Math	Reasoning	Roleplay	STEM	Writing	avg_score
Mixtral-8x7B-Instruct-v0.1	8x7B	4.1	8.4	9.8	4.7	5.6	9.0	9.2	9.5	7.5375
This model	around 8x12B?	4.2	7.4	9.9	4.0	5.2	9.5	8.7	8.0	7.1125

Merge Details

Merge Method

This model was merged using the passthrough merge method.

Models Merged

The following models were included in the merge:

• mistralai/Mixtral-8x7B-Instruct-v0.1

Configuration

The following YAML configuration was used to produce this model:

merge_method: passthrough
slices:
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [0, 8]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [4, 12]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [8, 16]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [12, 20]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [16, 24]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [20, 28]
  - sources:
      - model: mistralai/Mixtral-8x7B-Instruct-v0.1
        layer_range: [24, 32]
dtype: bfloat16
tokenizer_source: base