Charles O. Goddard

Does mergekit really need an internal library for generic computational graph scheduling and execution? Maybe not. But it's sure getting one.  Huge, invasive change for...

Addresses #80. Also does some plumbing adjustments to more robustly handle GPT 2 based models. Going to keep this as a draft for a while until I can test it...

* Allow using parameter interpolation by setting `smooth: true` in genome definition * Enable different parameter values using filters * Add `--timeout` option

NuSLERP

1

Adds a new merge method `nuslerp`. This method allows for a superset of the functionality of `slerp`. If provided with a base model `nuslerp` will perform spherical interpolation of the...

Done in a dumb way for now. Going to revisit this and make it more general at some point in the future. To try: ```yaml merge_method: dare_ties base_model: model_base.safetensors models:...

Allows using bitsandbytes quantization in `mergekit-evolve` when a) not using vLLM and b) not using in-memory mode.