Merged cognitivecomputations/dolphin-2.9-llama3-8b and meta-llama/Meta-Llama-3-8B-Instruct because I thought the Dolphin finetune was a bit too 'robot-y' in the answers.
GGUF files can be found here: RDson/Dolphin-less-Llama-3-Instruct-8B-GGUF.
Mergekit yaml:
tokenizer_source: union
slices:
- sources:
- model: ollama/llama3/Meta-Llama-3-8B-Instruct
layer_range: [0, 32]
- model: dolphin-2.9-llama3-8b
layer_range: [0, 32]
parameters:
weight: 0.75
merge_method: slerp
base_model: ollama/llama3/Meta-Llama-3-8B-Instruct
parameters:
normalize: true
embed_slerp: true
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