showcase / data.js
stevorob's picture
Upload data.js with huggingface_hub
f45239d verified
Raw History Blame Contribute Delete
5.98 kB
// GENERATED by scripts/generate_showcase_data.py from the ModelForge provenance
// DB. Do not hand-edit: the next run overwrites it. Add a model by adding a
// FAMILIES entry in that script and re-running. Published artifacts only.
// 7 families / 76 rungs.
const MODELS = [{"repo":"Qwen3.8-27B-abliterated-GGUF","short":"Qwen3.8-27B","base":"Qwen3.8-27B-abliterated","tag":"dense + vision","upstream":"Qwen/Qwen3.8-27B","blurb":"Our first pass at a dense model. Everything else here is a mixture of experts running a fraction of its weights per token; this one runs all 27B every time, so it is a different tradeoff rather than a smaller box. Also our first with vision: the tower survives abliteration and the f16 mmproj ships beside the ladder.","rungs":[{"q":"Q8_0","gb":29.05,"bpw":8.51,"i":0},{"q":"Q6_K","gb":22.43,"bpw":6.57,"i":1},{"q":"Q5_K_M","gb":19.54,"bpw":5.72,"i":1},{"q":"Q4_K_M","gb":16.84,"bpw":4.93,"i":1},{"q":"IQ4_XS","gb":15.37,"bpw":4.5,"i":1},{"q":"Q3_K_M","gb":13.59,"bpw":3.98,"i":1},{"q":"IQ3_XS","gb":12.26,"bpw":3.59,"i":1},{"q":"IQ2_M","gb":10.3,"bpw":3.02,"i":1}]},{"repo":"Qwen3.5-REAP-262B-A17B-abliterated-GGUF","short":"REAP-262B-A17B","base":"Qwen3.5-REAP-262B-A17B-abliterated","tag":"flagship","upstream":"OpenMOSE/Qwen3.5-REAP-262B-A17B","blurb":"A 34% expert-pruned cut of Qwen3.5-397B, 333 experts, abliterated and massive-activation-guarded. Every rung cut from one Q8_0 master, so the ladder is a single lineage rather than a requant chain.","rungs":[{"q":"Q5_K_M","gb":185.71,"bpw":5.68,"i":0},{"q":"Q4_K_M","gb":158.61,"bpw":4.85,"i":0},{"q":"Q4_K_S","gb":148.71,"bpw":4.55,"i":0},{"q":"IQ4_XS","gb":139.7,"bpw":4.28,"i":1},{"q":"Q3_K_M","gb":125.0,"bpw":3.83,"i":0},{"q":"IQ3_M","gb":114.6,"bpw":3.51,"i":1},{"q":"IQ3_XS","gb":107.28,"bpw":3.29,"i":1},{"q":"IQ2_M","gb":85.68,"bpw":2.62,"i":1},{"q":"IQ2_XS","gb":77.11,"bpw":2.36,"i":1},{"q":"IQ2_XXS","gb":69.26,"bpw":2.12,"i":1}]},{"repo":"Qwen3.5-REAP-212B-A17B-abliterated-GGUF","short":"REAP-212B-A17B","base":"Qwen3.5-REAP-212B-A17B-abliterated","tag":"cluster-fit","upstream":"OpenMOSE/Qwen3.5-REAP-212B-A17B","blurb":"48% REAP of Qwen3.5-397B. Ships alongside an attention-priority mix cut to a specific VRAM and context budget rather than to a generic rung.","rungs":[{"q":"Q5_K_M","gb":150.29,"bpw":5.68,"i":1},{"q":"Q4_K_M","gb":128.37,"bpw":4.85,"i":1},{"q":"IQ4_XS","gb":113.16,"bpw":4.27,"i":1},{"q":"IQ3_M","gb":92.87,"bpw":3.51,"i":1},{"q":"IQ3_XS","gb":86.97,"bpw":3.29,"i":1},{"q":"IQ2_M","gb":69.54,"bpw":2.63,"i":1},{"q":"fit1m","gb":67.41,"bpw":2.55,"i":1},{"q":"IQ2_XS","gb":62.6,"bpw":2.36,"i":1},{"q":"IQ2_XXS","gb":56.27,"bpw":2.13,"i":1}]},{"repo":"Qwen3.6-35B-A3B-abliterated-GGUF","short":"Qwen3.6-35B-A3B","base":"Qwen3.6-35B-A3B-abliterated","tag":"moe","upstream":"Qwen/Qwen3.6-35B-A3B","blurb":"Abliterated in-house from the base rather than re-cut from someone else's abliteration. Thirteen rungs, every one imatrix-guided, and at IQ2_XS it is the smallest file we publish: 11.2 GB, with 3B active per token. The one to reach for on a single card.","rungs":[{"q":"Q8_0","gb":37.8,"bpw":8.51,"i":0},{"q":"Q6_K","gb":29.21,"bpw":6.58,"i":1},{"q":"Q5_K_M","gb":25.42,"bpw":5.73,"i":1},{"q":"Q5_K_S","gb":24.67,"bpw":5.56,"i":1},{"q":"Q4_K_M","gb":21.86,"bpw":4.92,"i":1},{"q":"Q4_K_S","gb":20.58,"bpw":4.64,"i":1},{"q":"IQ4_XS","gb":19.42,"bpw":4.37,"i":1},{"q":"Q3_K_M","gb":17.46,"bpw":3.93,"i":1},{"q":"IQ3_M","gb":16.13,"bpw":3.63,"i":1},{"q":"Q3_K_S","gb":15.88,"bpw":3.57,"i":1},{"q":"IQ3_XS","gb":15.18,"bpw":3.42,"i":1},{"q":"IQ2_M","gb":12.35,"bpw":2.78,"i":1},{"q":"IQ2_XS","gb":11.2,"bpw":2.52,"i":1}]},{"repo":"Qwen3.5-122B-A10B-abliterated-GGUF","short":"Qwen3.5-122B-A10B","base":"Qwen3.5-122B-A10B-abliterated","tag":"widest ladder","upstream":"Qwen/Qwen3.5-122B-A10B","blurb":"The deepest ladder we publish: fourteen rungs from Q8_0 down to IQ2, MTP preserved throughout.","rungs":[{"q":"Q8_0","gb":132.55,"bpw":8.51,"i":0},{"q":"Q6_K","gb":102.38,"bpw":6.57,"i":1},{"q":"Q5_K_M","gb":88.94,"bpw":5.71,"i":1},{"q":"Q5_K_S","gb":86.27,"bpw":5.54,"i":1},{"q":"Q4_K_M","gb":76.28,"bpw":4.9,"i":1},{"q":"Q4_K_S","gb":71.72,"bpw":4.6,"i":1},{"q":"IQ4_XS","gb":67.53,"bpw":4.33,"i":1},{"q":"Q3_K_M","gb":60.68,"bpw":3.89,"i":1},{"q":"IQ3_M","gb":55.86,"bpw":3.58,"i":1},{"q":"Q3_K_S","gb":54.99,"bpw":3.53,"i":1},{"q":"IQ3_XS","gb":52.42,"bpw":3.36,"i":1},{"q":"IQ2_M","gb":42.34,"bpw":2.72,"i":1},{"q":"IQ2_XS","gb":38.33,"bpw":2.46,"i":1}]},{"repo":"Qwen3.5-122B-A10B-REAP-20-abliterated-GGUF","short":"122B REAP-20","base":"Qwen3.5-122B-A10B-REAP-20-abliterated","tag":"pruned","upstream":"0xSero/Qwen3.5-99B","blurb":"A 20% expert prune of the 122B, then abliterated. Same lineage, smaller box.","rungs":[{"q":"Q6_K","gb":81.33,"bpw":6.56,"i":1},{"q":"Q5_K_M","gb":70.44,"bpw":5.6,"i":1},{"q":"Q4_K_M","gb":60.19,"bpw":4.87,"i":1},{"q":"Q4_K_S","gb":56.51,"bpw":4.58,"i":1},{"q":"IQ4_XS","gb":53.15,"bpw":4.25,"i":1},{"q":"Q3_K_M","gb":47.57,"bpw":3.91,"i":1},{"q":"IQ3_M","gb":43.7,"bpw":3.66,"i":1},{"q":"IQ3_XS","gb":40.93,"bpw":3.3,"i":1},{"q":"IQ2_M","gb":32.78,"bpw":2.7,"i":1},{"q":"IQ2_XS","gb":29.54,"bpw":2.31,"i":1}]},{"repo":"Qwen3.5-122B-A10B-REAP-30-abliterated-GGUF","short":"122B REAP-30","base":"Qwen3.5-122B-A10B-REAP-30-abliterated","tag":"pruned","upstream":"0xSero/Qwen3.5-88B","blurb":"A 30% prune. The most aggressive cut of the 122B line we ship.","rungs":[{"q":"Q8_0","gb":93.25,"bpw":8.5,"i":0},{"q":"Q6_K","gb":72.02,"bpw":6.57,"i":1},{"q":"Q5_K_M","gb":62.39,"bpw":5.69,"i":1},{"q":"Q5_K_S","gb":60.48,"bpw":5.52,"i":1},{"q":"Q4_K_M","gb":53.32,"bpw":4.86,"i":1},{"q":"Q4_K_S","gb":50.06,"bpw":4.57,"i":1},{"q":"IQ4_XS","gb":47.12,"bpw":4.3,"i":1},{"q":"Q3_K_M","gb":42.16,"bpw":3.85,"i":1},{"q":"IQ3_M","gb":38.76,"bpw":3.54,"i":1},{"q":"Q3_K_S","gb":38.08,"bpw":3.47,"i":1},{"q":"IQ3_XS","gb":36.32,"bpw":3.31,"i":1},{"q":"IQ2_M","gb":29.12,"bpw":2.66,"i":1},{"q":"IQ2_XS","gb":26.24,"bpw":2.39,"i":1}]}];