ssmits

/

Falcon2-5.5B-multilingual

like 2

Hey, I tried to reproduce this using https://huggingface.co/spaces/arcee-ai/mergekit-gui and got an error. Did you have to modify mergekit to get it to merge?

Config YAML:

slices:
  - sources:
      - model: tiiuae/falcon-11B
        layer_range: [0, 24]
  - sources:
      - model: tiiuae/falcon-11B
        layer_range: [55, 59]
merge_method: passthrough
dtype: bfloat16

Error output:

...
[2024-09-07 17:15:27] [INFO] File "/home/user/.pyenv/versions/3.10.14/lib/python3.10/site-packages/mergekit/io/tasks.py", line 86, in execute
[2024-09-07 17:15:27] [INFO] raise RuntimeError(
[2024-09-07 17:15:27] [INFO] RuntimeError: Tensor transformer.h.58.ln_mlp.weight required but not present in model tiiuae/falcon-11B
[2024-09-07 17:15:28] [ERROR] Command exited with code 1
[2024-09-07 17:15:28] [ERROR] Merge failed. Deleting repo as no model is uploaded.