search

foundation-model-stack / fms-extras

Apache License 2.0
20 stars 9 forks source link

Add weight tying and input scaling to MLPSpeculator #36

Closed sahilsuneja1 closed 4 months ago

sahilsuneja1 commented 5 months ago

... and corresponding options to MLPSpeculatorConfig to create the HF version. Builds on top of @daviswer's PR #25

sahilsuneja1 commented 5 months ago

@JRosenkranz @daviswer Can we please get this merged-- needed for supporting speculators in watsonx

sahilsuneja1 commented 5 months ago

@JRosenkranz updated to use only 1 flag now-- tie_weights as per our discussion