CoolFace
Modelpublic

cstr/h2ovl-mississippi-2b-crispembed-GGUF

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
0likes157downloads
16 commits on main
1c620f22mo ago

Full-stack reference (54 stages, through the logits)

cstr
b7488ea2mo ago

Replace the 4-vision-layer partial reference with the full 24-layer dump

cstr
61a33022mo ago

Card: vision tower at F16, per-stage parity table

cstr
1472a692mo ago

q8_0 with vision tower at F16: vis_proj_output 0.912992 -> 0.999974 (CPU)

cstr
fd38fb02mo ago

Withdraw q4_k: decoder anti-correlated by layer 2, emits confident wrong text

cstr
20ff1e42mo ago

q8_0 verified; withdraw the measured-broken q4_k

cstr
e9b4c592mo ago

Model card

cstr
9bc28ba2mo ago

CrispEmbed-format GGUF (MSAC two-scale tiling required)

cstr
54ced0a2mo ago

CrispEmbed-format GGUF (MSAC two-scale tiling required)

cstr
4080d252mo ago

CrispEmbed-format GGUF (MSAC two-scale tiling required)

cstr
ba499342mo ago

Model card

cstr
e684ef12mo ago

f16 (UNVALIDATED — diagnostic)

cstr
cbfc0312mo ago

q8_0 (UNVALIDATED — diagnostic)

cstr
30ea38d2mo ago

q4_k (UNVALIDATED — diagnostic)

cstr
a4467e12mo ago

Card: UNVALIDATED, diagnostic only

cstr
97b0e792mo ago

initial commit

cstr