Highly experimental merge with itself, as done by Maxime Labonne, based on Eric Hatford's work with Dolphin 2.9 Llama. Context window is around 524K tokens. Works best with ChatML. Q8_0 Quant. Added mmproj-model-f16.gguf, experimenting with vision-enabling long context quant.

merged

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

Merge Details

Merge Method

This model was merged using the passthrough merge method.

Models Merged

The following models were included in the merge:

/Projects/llama.cpp/models/Unhinged-Llama3-8B-524K

Configuration

The following YAML configuration was used to produce this model:

dtype: float16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 8]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [4, 12]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [8, 16]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [12, 20]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [16, 24]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [20, 28]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K
- sources:
  - layer_range: [24, 32]
    model: /Projects/llama.cpp/models/Unhinged-Llama3-8B-524K