Tianqi Chen
Tianqi Chen
@huevosabio this is likely due to you are using conda (x86) . Try to get a miniforge that works natively on arm64 instead
Thanks for the feedbacks, we didn't get our hand on other phones, but would love to continue improve. There is a recent update to reduce the RAM usage of APK...
We indeed have a limit on RAM we can handle. Latest app ships with 3B model that might help in such cases
registry pattern sounds right for this case
cc @junrushao @jinhongyii
@merrymercy We updated it to mostly follow the new design https://github.com/mlc-ai/mlc-llm/pull/251, there are a few places that might benefit from standardizing further, such as seps vs sep, sep1. Do you...
Thank you @manuongithub , do you mind send a PR to okenizers-cpp/web
cc @CharlieFRuan
let us build a validation function in gen_config to inspect the generated json, and check for required fields
pruned models may have degenerated perf, closing for now, feel free to open new issues