search

HazyResearch / safari

Convolutions for Sequence Modeling
Apache License 2.0
869 stars 71 forks source link

lose lm.yaml #48

Open Peilin-FF opened 3 days ago

Peilin-FF commented 3 days ago

I'm struggling with the configuration that hydra doesn't recognize in configs：

I'm reproducing H3's experiment on wikitext, using the setting of -/model: lm in the h3.yaml configuration of safari/configs/experiment/wt103wt103 to manipulate the original text. However, the safari/configs/model folder no longer contains lm.yaml. How do I find or build this profile? image image

DanFu09 commented 3 days ago

lm is a mapping to long_conv_lm: https://github.com/HazyResearch/safari/blob/main/src/utils/registry.py#L24

DanFu09 commented 3 days ago

See this on how to train with this repo: https://github.com/HazyResearch/safari/blob/main/experiments.md