ChiTako/unofficial-photon-repro-ja-250m-v3-900m
0
mark clearly as an unofficial reimplementation
correct the perplexity methodology
PHOTON-JP small v3 (900M tokens, full mixture): trained checkpoint, eval and logs
initial commit
mark clearly as an unofficial reimplementation
correct the perplexity methodology
PHOTON-JP small v3 (900M tokens, full mixture): trained checkpoint, eval and logs
initial commit