Forge SZL-MiniEmbed v1: REAL trained embeddings (co-occurrence+PPMI+TruncatedSVD, no gensim) + vocab/config + receipt + eval + bundled corpus + honest card & provenance

7842caa verified 4 months ago

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}})"
  }
}

	{
	"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}})"
	}
	}