salakash

/

SamKash-Tolstoy

like 145

If you plan a public demo:
provide quantized inference configs (int8/int4 where appropriate)
add streaming generation + max length guards
add “safe defaults” decoding presets for creativity vs coherence