mergekit_config.yml

merge_method: linear # use linear so we can include multiple models, albeit at a zero weight
parameters:
  weight: 1.0 # weight everything as 1 unless specified otherwise - linear with one model weighted at 1 is a no-op like passthrough
slices:
  - sources:
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001 # embed_tokens comes along with the ride with whatever is the first layer
        layer_range: [0, 1]
      - model: lizpreciatior/lzlv_70b_fp16_hf # add dummy second model with 0 weight so tokenizer-based merge routine is invoked for embed_tokens
        layer_range: [0, 1]
        parameters:
          weight: 0
  - sources:
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001
        layer_range: [1, 20]
  - sources:
      - model: lizpreciatior/lzlv_70b_fp16_hf
        layer_range: [10, 30]
  - sources:
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001
        layer_range: [20, 40]
  - sources:
      - model: lizpreciatior/lzlv_70b_fp16_hf
        layer_range: [30, 50]
  - sources:
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001
        layer_range: [40, 60]
  - sources:
      - model: lizpreciatior/lzlv_70b_fp16_hf
        layer_range: [50, 70]
  - sources:
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001
        layer_range: [60, 79]
  - sources: # same as above, but for lm_head with the last layer
      - model: alchemonaut/QuartetAnemoi-70B-t0.0001
        layer_range: [79, 80]
      - model: lizpreciatior/lzlv_70b_fp16_hf
        layer_range: [79, 80]
        parameters:
          weight: 0
dtype: float16
tokenizer_source: model:alchemonaut/QuartetAnemoi-70B-t0.0001 # keep exact tokenizer used by Quartet - or you could use `union` if you add all of the input models to the first/last slice, but they would need to be non-zero weight or you'll get NaNs in your embeddings