File size: 4,523 Bytes
b7c9d17 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 | {
"name": "bend-models/models",
"version": "g30-imatrix-1",
"date": "2026-10-14",
"lane": "gap-chase/G30 (from G17 SHIP-DECISION)",
"license": "apache-2.0 (from GGUF general.license; same as upstream bases)",
"note": "Additive tree. The live phone app still fetches kev-4b-q4km/ (unchanged); nothing here is wired into any app.",
"evidence": {
"decision": "/home/omen/lab/fabric/runs/gap-chase/G17/SHIP-DECISION.md",
"kld_kev": "/home/omen/lab/fabric/runs/gap-chase/G2/receipts/kev4b-rep/RECEIPT.md + /home/omen/lab/fabric/runs/gap-chase/G2/receipts/kev4b/RECEIPT.md",
"kld_27b": "/home/omen/lab/fabric/runs/gap-chase/G17/receipts/27b/raw",
"battery": "/home/omen/lab/fabric/runs/gap-chase/G17/receipts/battery/",
"reverify": "/home/omen/lab/fabric/runs/gap-chase/G30/receipts/"
},
"files": [
{
"id": "kev-4b-q4km-imat",
"repo_path": "models/kev-4b-q4km-imat/Kev-4B-Q4_K_M-imat.gguf",
"sha256": "0535181cf3164a7271a7d3d1752a9ceab673c32bb316a081133ec238c186b51c",
"bytes": 2709547712,
"quant": "Q4_K_M + imatrix (kev4b-calib)",
"base": "Kev-4B BF16 (sha c690ee43…) incl. shipped cls.output head",
"replaces": "kev-4b-q4km/Kev-4B-Q4_K_M.gguf (plain, sha 93c66545…) — same tensor types/shapes, +288 B metadata",
"kld_vs_bf16": 0.02894,
"kld_shipped": 0.04573,
"top1": 89.81,
"battery": "38/40 labels vs BF16 (flips only on BF16-margin ≤0.104 items); goldens 8/8 (rot3 = lucky flip on a borderline golden)",
"native_gate": "types identical to shipped Q4_K_M → engine compat unchanged; G17 native smoke queued on GPU locks, not yet receipted",
"status": "ship (llama.cpp path); app NOT repointed",
"url": "https://huggingface.co/jethrovic/bend-models/resolve/main/models/kev-4b-q4km-imat/Kev-4B-Q4_K_M-imat.gguf"
},
{
"id": "kev-4b-iq4xs-imat",
"repo_path": "models/kev-4b-iq4xs-imat/Kev-4B-IQ4_XS-imat.gguf",
"sha256": "8a45e54a03a689368413de3bb6d33c1129ea416dfa7a8aade64c2f8c2c3b47dc",
"bytes": 2514987712,
"quant": "IQ4_XS + imatrix (kev4b-calib)",
"base": "Kev-4B BF16 (sha c690ee43…) incl. shipped cls.output head",
"replaces": "— (new small tier, −185 MiB vs Q4_K_M)",
"kld_vs_bf16": 0.03593,
"kld_shipped": 0.04573,
"top1": 89.22,
"battery": "40/40 labels vs BF16 (best 4-bit row); goldens 7/8 = same as BF16",
"native_gate": "PENDING — llb native greedy gate (G17 sessionkev) queued on GPU locks; must pass before any native-engine/phone use",
"status": "published for llama.cpp; native path BLOCKED until gate receipt",
"url": "https://huggingface.co/jethrovic/bend-models/resolve/main/models/kev-4b-iq4xs-imat/Kev-4B-IQ4_XS-imat.gguf"
},
{
"id": "qwen38-27b-q4ks-imat-openjev-wide",
"repo_path": "models/qwen38-27b-q4ks-imat-openjev-wide/Qwen3.8-27B-Q4_K_S-imat-openjev-wide.gguf",
"sha256": "c1e505b2e4266d964cc75e8a801e1c54dfec1bd4205c6ea7072d15c9b907fc0f",
"bytes": 15586316480,
"quant": "Q4_K_S + imatrix (qwen38-27b-calib) + openjev-wide metadata",
"base": "Qwen3.8-27B BF16",
"replaces": "board Qwen3.8-27B-Q4_K_S-openjev-wide (plain, sha 47a0bf5b…) — same type mix",
"kld_vs_q6k": 0.02126,
"kld_shipped_vs_q6k": 0.03652,
"top1": 93.3,
"battery": "40/40 labels vs shipped 27B (capfix llama-server); goldens 6/8 = shipped (Kev goldens, consistency only)",
"native_gate": "n/a — board 27B is served by llama-server (capfix), not llb",
"status": "ship (board, llama-server); promotion into the board store is an owner action",
"url": "https://huggingface.co/jethrovic/bend-models/resolve/main/models/qwen38-27b-q4ks-imat-openjev-wide/Qwen3.8-27B-Q4_K_S-imat-openjev-wide.gguf"
},
{
"id": "imatrix-kev4b-calib",
"repo_path": "models/imatrix/kev4b-calib.imatrix.gguf",
"sha256": "fe439522f2acc99fd040960f045f9582ccb89fe055d570ca883c61600a2f4737",
"bytes": 3626496,
"quant": "imatrix",
"base": "Kev-4B, G2 calib.txt (lab prose, no TDB)",
"status": "provenance",
"url": "https://huggingface.co/jethrovic/bend-models/resolve/main/models/imatrix/kev4b-calib.imatrix.gguf"
},
{
"id": "imatrix-qwen38-27b-calib",
"repo_path": "models/imatrix/qwen38-27b-calib.imatrix.gguf",
"sha256": "bd707218d6275dbcb342b23846776b306e535f431cb76f316349fa451138488c",
"bytes": 13642656,
"quant": "imatrix",
"base": "Qwen3.8-27B Q4_K_S, G2 calib.txt 202×512",
"status": "provenance",
"url": "https://huggingface.co/jethrovic/bend-models/resolve/main/models/imatrix/qwen38-27b-calib.imatrix.gguf"
}
]
} |