| { | |
| "model": "B-16M-sparse (github.com/re133/sparse-memory-lm)", | |
| "rows": 16777216, | |
| "dim": 384, | |
| "table_keys": [ | |
| "layers.2.ffn.values.weight", | |
| "layers.6.ffn.values.weight", | |
| "layers.10.ffn.values.weight" | |
| ], | |
| "q4": "kernels.quantize_q4: per row d = fp16(max-abs value with sign / -8), code = clamp(round(w/d),-8,7)+8, low nibble = dims 0..dim/2-1, high nibble = dims dim/2..dim-1", | |
| "cpu_vs_gpu_q4_check": { | |
| "rows": 1048576, | |
| "packed_equal": true, | |
| "scales_equal": true | |
| }, | |
| "sha256": { | |
| "hot_rows.npy": "7380ad9608603d05289e3698259a69472056723c7a35bb0fe925833b3e109fb5", | |
| "rest.pt": "c7ff78fa2b2afe3eaaab29e993d2b988628a858436e8d629364161cca242dbc8", | |
| "scales_q4.bin": "3a260537dc23ef4c7d73cfd358f9bd4dd9bd95ce89aede4d48ae73a683271294", | |
| "values_q4.bin": "67f8900ba65e34a2b6d93116f61636fd9ce43d13c96d5ed870f955bd6c54916c" | |
| } | |
| } |