ChenWeiLi/Med-ChimeraLlama-3-8B_SHERP

Chimera_MedLlama-3-8B

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:

• mlabonne/ChimeraLlama-3-8B-v3
• johnsnowlabs/JSL-MedLlama-3-8B-v2.0

Evaluation

• multimedqa (0 shot)
  

Tasks	Version	Filter	n-shot	Metric	Value		Stderr
- medmcqa	Yaml	none	0	acc	0.6087	±	0.0075
		none	0	acc_norm	0.6087	±	0.0075
- medqa_4options	Yaml	none	0	acc	0.6269	±	0.0136
		none	0	acc_norm	0.6269	±	0.0136
- anatomy (mmlu)	0	none	0	acc	0.6963	±	0.0397
- clinical_knowledge (mmlu)	0	none	0	acc	0.7585	±	0.0263
- college_biology (mmlu)	0	none	0	acc	0.7847	±	0.0344
- college_medicine (mmlu)	0	none	0	acc	0.6936	±	0.0351
- medical_genetics (mmlu)	0	none	0	acc	0.8200	±	0.0386
- professional_medicine (mmlu)	0	none	0	acc	0.7684	±	0.0256
stem	N/A	none	0	acc_norm	0.6129	±	0.0066
		none	0	acc	0.6440	±	0.0057
- pubmedqa	1	none	0	acc	0.7480	±	0.0194

Groups	Version	Filter	n-shot	Metric	Value		Stderr
stem	N/A	none	0	acc_norm	0.6129	±	0.0066
		none	0	acc	0.6440	±	0.0057

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: mlabonne/ChimeraLlama-3-8B-v3
        layer_range: [0, 32]
      - model: johnsnowlabs/JSL-MedLlama-3-8B-v2.0
        layer_range: [0, 32]
merge_method: slerp
base_model:  mlabonne/ChimeraLlama-3-8B-v3
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5
dtype: bfloat16