szl-kernels / config.json
betterwithage's picture
Forge SZL-MiniEmbed v1: REAL trained embeddings (co-occurrence+PPMI+TruncatedSVD, no gensim) + vocab/config + receipt + eval + bundled corpus + honest card & provenance
7842caa verified
Raw
History Blame Contribute Delete
399 Bytes
{
"model": "SZL-MiniEmbed",
"method": "term-term co-occurrence (distance-weighted, window=5) -> PPMI -> TruncatedSVD",
"dim": 128,
"vocab_size": 3290,
"min_count": 5,
"window": 5,
"seed": 20260721,
"normalization": "L2 row-normalized",
"files": {
"vectors": "vectors.npz (key 'vectors', float32 [V,dim])",
"vocab": "vocab.json ({'vocab':[term...], 'index':{term:i}})"
}
}