Upload folder using huggingface_hub

demo_options_qwen3.8-27b.json

@@ -0,0 +1 @@
+{"key": "Qwen/Qwen3.8-27B", "family": "Qwen", "options": [{"quant": "IQ2_M", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 10004594368, "bpw": 2.7, "fit_tier": "gpu", "repo_downloads": 11704}, {"quant": "UD-IQ2_S", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 8371970048, "bpw": 2.5, "fit_tier": "gpu", "repo_downloads": 6650349}, {"quant": "IQ2_XS", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 9088526048, "bpw": 2.31, "fit_tier": "gpu", "repo_downloads": 468425}, {"quant": "IQ2_XS", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 9090592448, "bpw": 2.31, "fit_tier": "gpu", "repo_downloads": 11704}, {"quant": "UD-IQ2_XXS", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 7266070528, "bpw": 2.06, "fit_tier": "gpu", "repo_downloads": 6650349}, {"quant": "IQ2_XXS", "repo": "byteshape/Qwen3.8-27B-GGUF", "size": 8843953536, "bpw": 2.06, "fit_tier": "gpu", "repo_downloads": 1421349}, {"quant": "IQ2_XXS", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 8881268448, "bpw": 2.06, "fit_tier": "gpu", "repo_downloads": 468425}, {"quant": "IQ2_XXS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 8977176032, "bpw": 2.06, "fit_tier": "gpu", "repo_downloads": 45504}, {"quant": "UD-IQ1_M", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 6729166848, "bpw": 1.75, "fit_tier": "gpu", "repo_downloads": 6650349}, {"quant": "IQ1_M", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 8498107872, "bpw": 1.75, "fit_tier": "gpu", "repo_downloads": 45504}, {"quant": "IQ1_M", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 6457847488, "bpw": 1.75, "fit_tier": "gpu", "repo_downloads": 11704}, {"quant": "UD-IQ1_S", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 6192222208, "bpw": 1.56, "fit_tier": "gpu", "repo_downloads": 6650349}, {"quant": "IQ3_XS", "repo": "byteshape/Qwen3.8-27B-GGUF", "size": 10353001856, "bpw": 3.3, "fit_tier": "partial", "repo_downloads": 1421349}, {"quant": "UD-Q2_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 9828981664, "bpw": 3.2, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "IQ3_XXS", "repo": "byteshape/Qwen3.8-27B-GGUF", "size": 9919481216, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 1421349}, {"quant": "Q2_K", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 10864592608, "bpw": 2.96, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q2_K", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 10864592192, "bpw": 2.96, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "IQ2_M", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 10522248928, "bpw": 2.7, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ2_S", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 9684043488, "bpw": 2.5, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ2_XS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 9889846752, "bpw": 2.31, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "IQ2_XS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 10221458912, "bpw": 2.31, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "IQ2_XS", "repo": "6block/Qwen3.8-27B-GGUF", "size": 10591877792, "bpw": 2.31, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ2_XS", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 10591877792, "bpw": 2.31, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "IQ2_XXS", "repo": "6block/Qwen3.8-27B-GGUF", "size": 9985014432, "bpw": 2.06, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ2_XXS", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 9985014432, "bpw": 2.06, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q2_K", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 10821543648, "bpw": 2.96, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ3_S", "repo": "byteshape/Qwen3.8-27B-GGUF", "size": 11035354496, "bpw": 3.44, "fit_tier": "partial", "repo_downloads": 1421349}, {"quant": "UD-IQ3_XXS", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 10934860704, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "IQ2_M", "repo": "6block/Qwen3.8-27B-GGUF", "size": 11271158432, "bpw": 2.7, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ2_M", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 11271158432, "bpw": 2.7, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q3_K_S", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 12256536288, "bpw": 3.5, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "IQ2_S", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 11141912032, "bpw": 2.5, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q2_K", "repo": "6block/Qwen3.8-27B-GGUF", "size": 11929795232, "bpw": 2.96, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q2_K", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 11929795232, "bpw": 2.96, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "UD-IQ3_S", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 12040883104, "bpw": 3.44, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "IQ3_XXS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 12075800032, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "IQ3_XXS", "repo": "6block/Qwen3.8-27B-GGUF", "size": 12451461792, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ3_XXS", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 12451461792, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q3_K_M", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 13500737248, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q3_K_M", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 13500736672, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 11704}, {"quant": "Q3_K_M", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 13500736832, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "IQ3_XXS", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 12320167648, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q3_K_S", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 12739884768, "bpw": 3.5, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q3_K_S", "repo": "6block/Qwen3.8-27B-GGUF", "size": 13060946592, "bpw": 3.5, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q3_K_S", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 13060946592, "bpw": 3.5, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "IQ3_XS", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 12799604448, "bpw": 3.3, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ4_XS", "repo": "byteshape/Qwen3.8-27B-GGUF", "size": 13083052416, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 1421349}, {"quant": "Q2_K_L", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 13081040480, "bpw": 3.1, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ3_XXS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 12984128992, "bpw": 3.06, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "UD-Q3_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 13146393504, "bpw": 4.4, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q3_K_L", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 14559799008, "bpw": 4.27, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q3_K_L", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 14559798592, "bpw": 4.27, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "IQ3_M", "repo": "6block/Qwen3.8-27B-GGUF", "size": 13489879712, "bpw": 3.66, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ3_M", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 13489879712, "bpw": 3.66, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q3_K_M", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 13404051168, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q3_K_M", "repo": "Chungulus/Qwen3.8-27B-Q3_K_M-GGUF", "size": 13500736896, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 5090}, {"quant": "Q3_K_M", "repo": "6block/Qwen3.8-27B-GGUF", "size": 14210448032, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q3_K_M", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 14210448032, "bpw": 3.91, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "IQ3_S", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 13838267872, "bpw": 3.44, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q3_K_L", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 14122817248, "bpw": 4.27, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ4_XS", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 15420450528, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "IQ4_XS", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 15309038752, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 11704}, {"quant": "Q4_0", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 15715853632, "bpw": 4.55, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "UD-IQ4_XS", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 14252845984, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "IQ3_S", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 14437471712, "bpw": 3.44, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q4_K_S", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 15825299168, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q4_K_S", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 15825298752, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "IQ3_M", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 14860629728, "bpw": 3.66, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "UD-Q4_K_S", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 15358213024, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q4_K_M", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 16810714848, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q4_K_M", "repo": "KikoCis/Qwen3.8-27B-GGUF", "size": 16810714272, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 11704}, {"quant": "Q4_K_M", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 16810714432, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "IQ4_XS", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 15475951328, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Qwen3.8-27B-MTP-IQ4_KS", "repo": "ubergarm/Qwen3.8-27B-GGUF", "size": 16921247392, "bpw": null, "fit_tier": "partial", "repo_downloads": 3945}, {"quant": "Q4_0", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 16056478688, "bpw": 4.55, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "IQ4_XS", "repo": "6block/Qwen3.8-27B-GGUF", "size": 16410040992, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ4_XS", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 16410040992, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "UD-Q4_K_M", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 16464440224, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q4_K_S", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 16363513568, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q4_K_S", "repo": "6block/Qwen3.8-27B-GGUF", "size": 16839957152, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q4_K_S", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 16839957152, "bpw": 4.58, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q4_0", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 16348767968, "bpw": 4.55, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q4_0", "repo": "6block/Qwen3.8-27B-GGUF", "size": 16769178272, "bpw": 4.55, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q4_0", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 16769178272, "bpw": 4.55, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q3_K_XL", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 16391919200, "bpw": 4.4, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q4_K_M", "repo": "lmstudio-community/Qwen3.8-27B-GGUF", "size": 16810714336, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 1172050}, {"quant": "Q4_K_M", "repo": "Abiray/Qwen3.8-27B-Q4_K_M-GGUF", "size": 16810714400, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 13078}, {"quant": "IQ4_NL", "repo": "6block/Qwen3.8-27B-GGUF", "size": 17039186592, "bpw": 4.5, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "IQ4_NL", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 17039186592, "bpw": 4.5, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "IQ4_XS", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 16512935392, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q4_K_M", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 17120781792, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q4_K_M", "repo": "6block/Qwen3.8-27B-GGUF", "size": 17743780512, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q4_K_M", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 17743780512, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "IQ4_XS", "repo": "Jackrong/Qwen3.8-27B-MTP-GGUF", "size": 15420450112, "bpw": 4.25, "fit_tier": "partial", "repo_downloads": 6348}, {"quant": "Q5_K_S", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 18971682528, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q5_0", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 18971682112, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "Q5_K_S", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 18971682112, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "UD-Q4_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 17559178144, "bpw": 5.05, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q4_1", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 17540705248, "bpw": 5.0, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q4_1", "repo": "6block/Qwen3.8-27B-GGUF", "size": 18140191392, "bpw": 5.0, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q4_1", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 18140191392, "bpw": 5.0, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q4_K_M", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 17442399968, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "IQ4_NL", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 17442399968, "bpw": 4.5, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q5_K_M", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 19535701728, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q5_K_M", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 19535701312, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "Q4_1", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 17825457888, "bpw": 5.0, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "UD-Q5_K_S", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 18665753504, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q4_K_M", "repo": "ggml-org/Qwen3.8-27B-GGUF", "size": 18973870528, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 2212866}, {"quant": "Q4_K_M", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 18546845152, "bpw": 4.85, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q5_K_S", "repo": "6block/Qwen3.8-27B-GGUF", "size": 19555768992, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q5_K_S", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 19555768992, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q4_K_L", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 18820703968, "bpw": 5.0, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q5_K_M", "repo": "6block/Qwen3.8-27B-GGUF", "size": 20080794272, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q5_K_M", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 20080794272, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q5_K_S", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 19566913248, "bpw": 5.54, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "UD-Q5_K_M", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 19771509664, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q5_K_M", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 20232512992, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q6_K", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 22431000288, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q6_K", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 22430999872, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "UD-Q5_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 20876938144, "bpw": 5.9, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q5_K_M", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 20923877088, "bpw": 5.69, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q5_K_L", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 21537478240, "bpw": 5.8, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "UD-Q6_K", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 21983677344, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q6_K", "repo": "6block/Qwen3.8-27B-GGUF", "size": 22563871392, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q6_K", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 22563871392, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "Q6_K", "repo": "lmstudio-community/Qwen3.8-27B-GGUF", "size": 22430999776, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 1172050}, {"quant": "UD-Q6_K_M", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 23088409504, "bpw": null, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q6_K_S", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 22857455328, "bpw": null, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q5_K", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 23081711072, "bpw": null, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q6_K", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 23860565728, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "UD-Q6_K_L", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 24193919904, "bpw": 6.7, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "UD-Q6_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 25299061664, "bpw": 6.9, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q6_K_L", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 24958908128, "bpw": 6.7, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q6_K", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 25005438432, "bpw": 6.56, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "Q8_0", "repo": "mradermacher/Qwen3.8-27B-GGUF", "size": 29047084768, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 14248}, {"quant": "Q8_0", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 29047084352, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "Q8_0", "repo": "ggml-org/Qwen3.8-27B-GGUF", "size": 28595763648, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 2212866}, {"quant": "Q8_0", "repo": "6block/Qwen3.8-27B-GGUF", "size": 28049234592, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 6052}, {"quant": "Q8_0", "repo": "qtum/Qwen3.8-27B-GGUF", "size": 28049234592, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 5308}, {"quant": "UD-Q8_K_L", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 28045695904, "bpw": null, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q8_0", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 29047086048, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "Q8_0", "repo": "lmstudio-community/Qwen3.8-27B-GGUF", "size": 29047084256, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 1172050}, {"quant": "Q8_0", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 29116388960, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "Q8_0", "repo": "AtomicChat/Qwen3.8-27B-GGUF", "size": 28887832032, "bpw": 8.5, "fit_tier": "partial", "repo_downloads": 45504}, {"quant": "UD-Q8_K_XL", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 31457991680, "bpw": 8.8, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "BF16", "repo": "ggml-org/Qwen3.8-27B-GGUF", "size": 53808282048, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 2212866}, {"quant": "BF16", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 54657733952, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "F16", "repo": "prithivMLmods/Qwen3.8-27B-GGUF", "size": 54657733952, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 8939}, {"quant": "BF16", "repo": "unsloth/Qwen3.8-27B-GGUF", "size": 54657735616, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 6650349}, {"quant": "BF16", "repo": "bartowski/Qwen3.8-27B-GGUF", "size": 54657734144, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 468425}, {"quant": "BF16", "repo": "Jackrong/Qwen3.8-27B-MTP-GGUF", "size": 54657733952, "bpw": 16.0, "fit_tier": "partial", "repo_downloads": 6348}]}

fork.js

@@ -1,12 +1,12 @@
 /* Quant-picking frontend. Catalog + fit data come live from the
- * LocalLLaMA explorer API (CORS-open); best-quant choice comes from
+ * LocalLLaMA explorer API; the best-quant choice per model comes from
  * rankQuants.js + quant_scores.json (measured KL benchmarks). */
 "use strict";
 
 const API = "https://codelion-local-model-explorer.hf.space";
 const $ = (s, r = document) => r.querySelector(s);
 const $$ = (s, r = document) => [...r.querySelectorAll(s)];
-let META = null, SCORES = null, runId = 0;
+let META = null, SCORES = null, runId = 0, offset = 0, total = 0;
 
 function el(tag, cls, text) {
   const n = document.createElement(tag);
@@ -14,31 +14,29 @@ function el(tag, cls, text) {
   if (text !== undefined) n.textContent = text;
   return n;
 }
-function opt(sel, value, label, selected) {
+function opt(value, label, selected) {
   const o = document.createElement("option");
   o.value = value; o.textContent = label;
   if (selected) o.selected = true;
   return o;
 }
-function badge(quant) {
-  const b = labelFor(quant, SCORES);
-  const s = el("span", "badge", b);
-  const cls = { "bench default band": "b-default", "bench step-up": "b-step",
-    "near-lossless": "b-archival", "tight budget": "b-tight", "min viable": "b-knee",
-    "emergency only": "b-emergency", "dominated": "b-avoid",
-    "full precision (ref)": "" }[b.replace(" (est.)", "")];
-  if (cls) s.classList.add(cls);
-  return s;
-}
+const gb = (v, d = 1) => v === null || v === undefined ? "?" : `${Number(v).toFixed(d)} GB`;
+const bytesGB = (b) => gb(b / 1e9);
+const params = (p) => p >= 1e9 ? `${+(p / 1e9).toFixed(p >= 1e10 ? 0 : 1)}B` : `${Math.round(p / 1e6)}M`;
+const compact = (n) => n >= 1e6 ? `${(n / 1e6).toFixed(1)}M` : n >= 1e3 ? `${Math.round(n / 1e3)}k` : String(n || 0);
+const ctxLabel = (c) => c >= 1024 ? `${Math.round(c / 1024)}k` : String(c);
 
 function rig() {
   const kind = $('input[name="kind"]:checked').value;
   const out = { kind, gpus: [], ram_gb: 32 };
   if (kind === "gpu") {
-    const name = $("#gpu").value;
-    const each = name === "custom" ? +$("#custom-vram").value : META.hardware.gpus[name];
-    const n = +$("#gpu-count").value || 1;
-    out.gpus = Array.from({ length: n }, () => each);
+    const one = (sel, vsel, csel, fb) => {
+      const name = $(sel).value;
+      const each = name === "custom" ? Math.max(1, Math.min(256, +$(vsel).value || fb)) : META.hardware.gpus[name];
+      return Array.from({ length: +$(csel).value || 1 }, () => each);
+    };
+    out.gpus = one("#gpu", "#custom-vram", "#gpu-count", 24);
+    if (!$("#gpu2-wrap").hidden) out.gpus = out.gpus.concat(one("#gpu2", "#custom-vram2", "#gpu2-count", 8)).slice(0, 16);
     out.ram_gb = +$("#ram").value;
   } else if (kind === "unified") {
     out.unified_system = $("#usys").value;
@@ -49,61 +47,62 @@ function rig() {
   return out;
 }
 
-function params(p) {
-  return p >= 1e9 ? `${+(p / 1e9).toFixed(p >= 1e10 ? 0 : 1)}B` : `${Math.round(p / 1e6)}M`;
+function query(append) {
+  const r = rig();
+  return {
+    hardware: { kind: r.kind, gpus: r.gpus, ram_gb: r.ram_gb,
+      ...(r.unified_system ? { unified_system: r.unified_system } : {}) },
+    context: +$("#ctx").value, kv_type: $("#kv").value,
+    priority: $('input[name="p"]:checked').value, min_bits: +$("#min-bits").value,
+    sort: $("#sort").value, family: $("#family").value || null, size: $("#size").value || null,
+    search: $("#search").value.trim() || null,
+    moe_only: $("#moe-only").checked, vision_only: $("#vision-only").checked,
+    fits_only: $("#fits-only").checked,
+    vram_only: r.kind === "gpu" && $("#vram-only").checked, mmproj: $("#mmproj").value,
+    offset: append ? offset : 0, limit: 24,
+  };
 }
-const gb = (b) => `${(b / 1e9).toFixed(2)} GB`;
 
-function pickLine(pick) {
-  const d = el("div", "choice");
-  const b = el("b", null, pick.quant);
-  d.append(b, ` from ${pick.repo.split("/")[0]} — ${gb(pick.size)} · `);
-  d.append(badge(pick.quant));
-  if (pick.fit) {
-    d.append(el("div", "muted",
-      `${pick.fit.label}: ${gb(pick.fit.gpu_gb * 1e9)} on GPU, ${gb(pick.fit.ram_gb * 1e9)} in RAM`));
-  }
-  return d;
+const SCORE_PARTS = [["fit", "Fit"], ["capacity", "Size"], ["popularity", "Downloads"], ["recency", "New"], ["community", "Reports"]];
+function scoreTitle(m) {
+  const w = META.score_weights || {};
+  return SCORE_PARTS.map(([k, l]) => `${l}: ${Math.round((m.components[k] || 0) * w[k] * 100)} of ${Math.round(w[k] * 100)}`).join("\n");
 }
 
-function quantTable(detail, pick) {
-  const t = el("table", "quants");
-  const head = el("tr");
-  for (const h of ["Uploader", "Quant", "Size", "Quality", "Fit"]) head.append(el("th", null, h));
-  const tb = el("tbody");
-  const ranked = [...detail.options].sort((a, b) => a.size - b.size);
-  for (const o of ranked) {
-    const tr = el("tr");
-    if (pick && o.quant === pick.quant && o.repo === pick.repo) tr.className = "is-pick";
-    const qtd = el("td");
-    const qb = el("b", null, o.quant);
-    qtd.append(qb);
-    if (pick && o.quant === pick.quant && o.repo === pick.repo) qtd.append(" ◀ pick");
-    const btd = el("td"); btd.append(badge(o.quant));
-    const ftd = el("td", null, o.fit ? o.fit.label : "?");
-    tr.append(el("td", null, o.repo.split("/")[0]), qtd, el("td", "num", gb(o.size)), btd, ftd);
-    tb.append(tr);
+function rail(fit, r) {
+  if (!fit) return null;
+  const box = el("div", "rail");
+  const pct = (v, cap) => `${Math.max(0, Math.min(100, (v / cap) * 100))}%`;
+  const vram = r.kind === "gpu" ? r.gpus.reduce((a, v) => a + v, 0) : 0;
+  const ram = r.ram_gb;
+  if (r.kind === "unified") {
+    const share = (META.hardware.unified_shares || {})[r.unified_system];
+    const cap = r.ram_gb * (share || (r.ram_gb <= 36 ? 0.67 : 0.75));
+    const pool = el("div", "pool");
+    pool.style.flex = "1";
+    const i = el("i", "gpu"); i.style.width = pct(fit.gpu_gb, cap); pool.append(i);
+    if (fit.gpu_gb > cap) { const o = el("i", "over"); o.style.width = "30%"; pool.append(o); }
+    box.append(pool);
+    return box;
   }
-  t.append(head, tb);
-  return t;
+  if (vram) {
+    const pool = el("div", "pool"); pool.style.flex = `${vram / (vram + ram)}`;
+    const i = el("i", "gpu"); i.style.width = pct(fit.gpu_gb, vram); pool.append(i);
+    box.append(pool);
+  }
+  const ramPool = el("div", "pool"); ramPool.style.flex = `${ram / ((vram || 1) + ram)}`;
+  const j = el("i", fit.tier === "gpu_moe" ? "exp" : "ram"); j.style.width = pct(fit.ram_gb, ram);
+  ramPool.append(j);
+  if (fit.tier === "no" || fit.tier === "no_vram") { const o = el("i", "over"); o.style.width = "30%"; ramPool.append(o); }
+  box.append(ramPool);
+  return box;
 }
 
-function cmdBlock(detail, pick) {
-  const o = detail.options.find((x) => pick && x.repo === pick.repo && x.quant === pick.quant)
-    || detail.options[0];
-  if (!o || !o.commands) return el("p", "muted", "No run command available.");
-  const pre = el("pre", "cmd", o.commands["llama.cpp"]);
-  const btn = el("button", "btn ghost", "Copy");
-  btn.addEventListener("click", async () => {
-    try { await navigator.clipboard.writeText(pre.textContent); btn.textContent = "Copied"; }
-    catch (e) { btn.textContent = "Select and copy"; }
-    setTimeout(() => (btn.textContent = "Copy"), 1500);
-  });
-  const wrap = el("div");
-  const row = el("div", "row");
-  row.append(el("span", "muted", "Run it:"), btn);
-  wrap.append(row, pre);
-  return wrap;
+function fitText(fit) {
+  if (!fit) return "";
+  const parts = [fit.label];
+  if (fit.tier !== "no" && fit.tier !== "no_vram") parts.push(`${gb(fit.gpu_gb)} GPU, ${gb(fit.ram_gb)} RAM`);
+  return parts.join(" · ");
 }
 
 function card(m) {
@@ -112,42 +111,153 @@ function card(m) {
   const btn = el("button", null, m.name);
   h.append(btn);
   c.append(h);
-  c.append(el("div", "meta",
-    `${m.params ? params(m.params) + " · " : ""}${m.family} · ${m.repo_count} GGUF repos · ${(m.downloads / 1e6).toFixed(1)}M downloads`));
+  const tags = [
+    m.params ? params(m.params) + (m.active_params ? ` (${params(m.active_params)} active)` : "") : null,
+    m.moe ? "MoE" : null, m.vision ? "vision" : null, m.family,
+    `${m.repo_count} GGUF repos`, `${compact(m.downloads)} downloads`,
+  ].filter(Boolean).join(" · ");
+  c.append(el("div", "meta", tags));
+  const side = el("div", "r-side");
+  const sc = el("span", "score", String(Math.round(m.score)));
+  sc.title = scoreTitle(m) + "\nHardware ranking, not model quality.";
+  side.append(sc, el("span", "muted", "score"));
+  c.append(side);
   const status = el("p", "muted", "Ranking quants…");
   c.append(status);
   const body = el("div");
   c.append(body);
   btn.addEventListener("click", () => {
-    body.replaceChildren();
-    if (!m._detail) return;
-    body.append(cmdBlock(m._detail, m._pick), quantTable(m._detail, m._pick));
+    if (!m._detail || body.children.length) { body.replaceChildren(); return; }
+    renderDetail(m, body);
   });
   return { c, status, body };
 }
 
+function renderDetail(m, body) {
+  const d = m._detail, pick = m._pick;
+  const g = d.geometry;
+  const facts = [
+    d.params ? `${params(d.params)} parameters${d.active_params ? `, ${params(d.active_params)} active` : ""}` : null,
+    g ? `${g.n_layers} layers` : null,
+    g && g.moe ? `${g.expert_count} experts, ${g.expert_used} per token` : null,
+    d.context_length ? `${ctxLabel(d.context_length)} native context` : null,
+    g && g.notes && g.notes.length ? g.notes.join(", ") : null,
+  ].filter(Boolean).join("  •  ");
+  if (facts) body.append(el("p", "muted", facts));
+
+  const opts = d.options || [];
+  const sel = opts.find((o) => pick && o.repo === pick.repo && o.quant === pick.quant) || opts[0];
+  if (sel && sel.commands) {
+    const row = el("div", "row");
+    const copy = el("button", "btn ghost", "Copy command");
+    const pre = el("pre", "cmd", sel.commands["llama.cpp"]);
+    copy.addEventListener("click", async () => {
+      try { await navigator.clipboard.writeText(pre.textContent); copy.textContent = "Copied"; }
+      catch { copy.textContent = "Select and copy"; }
+      setTimeout(() => (copy.textContent = "Copy command"), 1500);
+    });
+    row.append(el("span", "muted", `Run it (${sel.quant} from ${sel.repo.split("/")[0]}):`), copy);
+    body.append(row, pre);
+    if (sel.fit) body.append(el("div", "muted", fitText(sel.fit)));
+  }
+
+  body.append(el("h3", null, `Every quant (${opts.length})`));
+  if (d.hidden_below_min_bits) {
+    body.append(el("p", "muted",
+      `${d.hidden_below_min_bits} quant(s) hidden by the bit floor. Lower "Lowest quant to consider" to see them.`));
+  }
+  const t = el("table", "quants");
+  const head = el("tr");
+  for (const [h, num] of [["Uploader", 0], ["Quant", 0], ["Size", 1], ["On your hardware", 0]]) {
+    head.append(el("th", num ? "num" : null, h));
+  }
+  const tb = el("tbody");
+  for (const o of [...opts].sort((a, b) => a.size - b.size)) {
+    const tr = el("tr");
+    if (pick && o.quant === pick.quant && o.repo === pick.repo) tr.className = "is-pick";
+    const qtd = el("td");
+    qtd.append(el("b", null, o.quant));
+    if (pick && o.quant === pick.quant && o.repo === pick.repo) qtd.append(" ◀ pick");
+    const link = el("a", null, o.repo.split("/")[0]);
+    link.href = o.url; link.target = "_blank"; link.rel = "noopener";
+    const utd = el("td"); utd.append(link);
+    tr.append(utd, qtd, el("td", "num", bytesGB(o.size)), el("td", null, o.fit ? o.fit.label : "?"));
+    tb.append(tr);
+  }
+  t.append(head, tb);
+  body.append(t);
+
+  if (d.kv_cache_gb) {
+    body.append(el("h3", null, "KV cache"));
+    const strip = el("div", "kv-strip");
+    for (const [ctx, v] of Object.entries(d.kv_cache_gb)) {
+      const s = el("span", null, `${ctxLabel(+ctx)} `);
+      s.append(el("b", null, gb(v, v < 10 ? 2 : 1)));
+      strip.append(s);
+    }
+    body.append(strip);
+  }
+  const fm = d.other_formats || {};
+  const names = { mlx: "MLX", onnx: "ONNX", awq: "AWQ", gptq: "GPTQ", fp8: "FP8", exl3: "EXL3", exl2: "EXL2" };
+  const any = Object.keys(names).filter((k) => fm[k] && fm[k].length);
+  if (any.length) {
+    body.append(el("h3", null, "Other formats"));
+    for (const k of any) {
+      const p = el("p", "muted", `${names[k]} (${fm[k].length}): `);
+      for (const id of fm[k].slice(0, 4)) {
+        const a = el("a", null, id.split("/")[1] || id);
+        a.href = `https://huggingface.co/${id}`; a.target = "_blank"; a.rel = "noopener";
+        p.append(a, "  ");
+      }
+      body.append(p);
+    }
+  }
+  const cm = d.community;
+  if (cm && cm.bench_rows && cm.bench_rows.length) {
+    body.append(el("h3", null, "Community benchmarks"));
+    const bt = el("table", "quants");
+    const bh = el("tr");
+    for (const [h, num] of [["Quant", 0], ["Hardware", 0], ["Backend", 0], ["Prompt t/s", 1], ["Gen t/s", 1]]) {
+      bh.append(el("th", num ? "num" : null, h));
+    }
+    const bb = el("tbody");
+    for (const b of cm.bench_rows) {
+      const tr = el("tr");
+      tr.append(el("td", null, b.quant || "?"),
+        el("td", null, b.gpu_name || b.gpu_model || (b.vram_class ? `${b.vram_class} GB GPU` : "CPU")),
+        el("td", null, b.backend || "?"),
+        el("td", "num", b.prompt_tps ? String(Math.round(b.prompt_tps)) : "—"),
+        el("td", "num", b.generation_tps ? b.generation_tps.toFixed(1) : "—"));
+      bb.append(tr);
+    }
+    bt.append(bh, bb);
+    body.append(bt);
+  }
+}
+
 async function detailFor(key, q, signal) {
   const r = await fetch(`${API}/api/model`, { method: "POST",
     headers: { "content-type": "application/json" }, signal,
     body: JSON.stringify({ key, hardware: q.hardware, context: q.context,
-      kv_type: "f16", priority: q.priority, min_bits: 0, vram_only: false, mmproj: "gpu" }) });
+      kv_type: q.kv_type, priority: q.priority, min_bits: q.min_bits,
+      vram_only: q.vram_only, mmproj: q.mmproj }) });
   if (!r.ok) throw new Error(`model ${r.status}`);
   return r.json();
 }
 
-async function search() {
+function famGuess(key) {
+  if (/qwen/i.test(key)) return "Qwen";
+  if (/gemma/i.test(key)) return "Gemma";
+  return null;
+}
+
+async function search(append) {
   const id = ++runId;
-  const r = rig();
-  const q = {
-    hardware: { kind: r.kind, gpus: r.gpus, ram_gb: r.ram_gb,
-      ...(r.unified_system ? { unified_system: r.unified_system } : {}) },
-    context: +$("#ctx").value,
-    priority: $('input[name="p"]:checked').value,
-    search: $("#search").value.trim() || null,
-  };
+  const q = query(append);
+  if (!append) offset = 0;
   const box = $("#results");
-  box.replaceChildren();
-  $("#count").textContent = "Finding models…";
+  if (!append) box.replaceChildren();
+  $("#count").textContent = append ? $("#count").textContent : "Finding models…";
   const ctrl = new AbortController();
   const alive = () => id === runId;
 
@@ -155,10 +265,7 @@ async function search() {
   try {
     const rr = await fetch(`${API}/api/recommend`, { method: "POST",
       headers: { "content-type": "application/json" }, signal: ctrl.signal,
-      body: JSON.stringify({ hardware: q.hardware, context: q.context, kv_type: "f16",
-        priority: q.priority, min_bits: 0, sort: "best", family: null, size: null,
-        search: q.search, moe_only: false, vision_only: false, fits_only: false,
-        vram_only: false, mmproj: "gpu", offset: 0, limit: 12 }) });
+      body: JSON.stringify(q) });
     if (!rr.ok) throw new Error(`recommend ${rr.status}`);
     res = await rr.json();
   } catch {
@@ -166,56 +273,68 @@ async function search() {
     return;
   }
   if (!alive()) return;
-  $("#count").textContent = `${res.total.toLocaleString()} models — top ${res.results.length} with quant picks`;
-  const fams = res.results.map((m) => {
-    const fam = /qwen/i.test(m.key) ? "Qwen" : /gemma/i.test(m.key) ? "Gemma"
-      : /lfm|minimax|kimi/i.test(m.key) ? null : null;
-    return fam;
-  });
+  total = res.total;
+  offset += res.results.length;
+  $("#count").textContent = `${total.toLocaleString()} models`;
+  $("#more").hidden = offset >= total;
+  const r = rig();
+
+  for (const m of res.results) {
+    const ui = card(m);
+    m._ui = ui;
+    box.append(ui.c);
+  }
 
-  // bounded parallel detail fetch; each card upgrades when its data lands
-  const queue = res.results.map((m, i) => ({ m, fam: fams[i] }));
+  const queue = [...res.results];
   async function worker() {
     while (alive() && queue.length) {
-      const { m, fam } = queue.shift();
+      const m = queue.shift();
       try {
         const d = await detailFor(m.key, q, ctrl.signal);
         if (!alive()) return;
         const opts = d.options.map((o) => ({ ...o,
           fit_ok: o.fit && o.fit.tier !== "no" && o.fit.tier !== "no_vram" }));
-        const famGuess = fam || (d.family && /qwen|gemma/i.test(d.family) ? d.family : null);
-        const { pick, warnings } = rankOptions(opts, q.priority, famGuess, SCORES);
+        const { pick, warnings } = rankOptions(opts, q.priority, famGuess(m.key), SCORES);
         m._detail = d; m._pick = pick;
         const ui = m._ui;
         if (!ui) continue;
         ui.status.replaceChildren();
         if (pick) {
-          ui.status.append(pickLine(pick));
+          const line = el("div", "choice");
+          line.append(el("b", null, pick.quant),
+            ` from ${pick.repo.split("/")[0]} — ${bytesGB(pick.size)}${pick.full === false ? " (partial offload)" : ""}`);
+          ui.status.append(line);
+          ui.status.append(el("div", "muted", fitText(pick.fit)));
           for (const w of warnings) ui.status.append(el("div", "warn", w));
-          ui.body.replaceChildren(cmdBlock(d, pick), quantTable(d, pick));
+          const railEl = rail(pick.fit, r);
+          if (railEl) ui.status.append(railEl);
         } else {
-          ui.status.textContent = (warnings[0] || "Doesn't fit this hardware.");
+          ui.status.textContent = warnings[0] || "Doesn't fit this hardware.";
         }
       } catch (err) {
         if (alive() && m._ui) m._ui.status.textContent =
-          `Quant ranking failed for this model (${String(err).slice(0, 100)}).`;
+          `Quant ranking failed (${String(err).slice(0, 100)}).`;
       }
     }
   }
-
-  for (const m of res.results) {
-    const ui = card(m);
-    m._ui = ui;
-    box.append(ui.c);
-  }
   await Promise.all(Array.from({ length: 4 }, worker));
 }
 
-// wire kind visibility
+let deb;
+function schedule(append) {
+  clearTimeout(deb);
+  deb = setTimeout(() => search(append || false), 250);
+}
+
 function syncKind() {
   const kind = $('input[name="kind"]:checked').value;
-  $$("[data-show]").forEach((n) => { n.hidden = !n.dataset.show.split(" ").includes(kind); });
+  $$("[data-show]").forEach((n) => {
+    if (n.id === "gpu2-wrap") return; // toggle-owned, not kind-owned
+    n.hidden = !n.dataset.show.split(" ").includes(kind);
+  });
   $("#custom-vram-field").hidden = kind !== "gpu" || $("#gpu").value !== "custom";
+  $("#custom-vram2-field").hidden = $("#gpu2").value !== "custom";
+  $("#gpu2-toggle").textContent = $("#gpu2-wrap").hidden ? "Add a different card" : "Remove the other card";
 }
 
 async function boot() {
@@ -227,38 +346,57 @@ async function boot() {
     return;
   }
   const H = META.hardware;
-  for (const [label, names] of H.gpu_groups) {
-    const g = document.createElement("optgroup");
-    g.label = label;
-    for (const n of names) g.append(opt(null, n, `${n.replace(/^(NVIDIA|AMD|Intel) /, "")} (${H.gpus[n]} GB)`));
-    $("#gpu").append(g);
-  }
-  $("#gpu").append(opt(null, "custom", "Other: enter VRAM"));
+  const groups = (sel) => {
+    for (const [label, names] of H.gpu_groups) {
+      const g = document.createElement("optgroup");
+      g.label = label;
+      for (const n of names) g.append(opt(n, `${n.replace(/^(NVIDIA|AMD|Intel) /, "")} (${H.gpus[n]} GB)`));
+      $(sel).append(g);
+    }
+    $(sel).append(opt("custom", "Other: enter VRAM"));
+  };
+  groups("#gpu"); groups("#gpu2");
   $("#gpu").value = "NVIDIA RTX 5070";
+  $("#gpu2").value = "NVIDIA RTX 3060 12GB";
   const vramSet = [...new Set(Object.values(H.gpus))].sort((a, b) => a - b);
-  for (const v of vramSet) $("#custom-vram").append(opt(null, v, `${v} GB`, v === 12));
-  for (const n of [1, 2, 3, 4]) $("#gpu-count").append(opt(null, n, n, n === 1));
-  for (const v of H.ram) $("#ram").append(opt(null, v, `${v} GB`, v === 64));
-  for (const s of Object.keys(H.unified)) $("#usys").append(opt(null, s, s));
+  for (const sel of ["#custom-vram", "#custom-vram2"]) {
+    for (const v of vramSet) $(sel).append(opt(v, `${v} GB`, v === 12));
+  }
+  for (const sel of ["#gpu-count", "#gpu2-count"]) {
+    for (const n of [1, 2, 3, 4, 6, 8]) $(sel).append(opt(n, n, n === 1));
+  }
+  for (const v of H.ram) $("#ram").append(opt(v, `${v} GB`, v === 64));
+  for (const s of Object.keys(H.unified)) $("#usys").append(opt(s, s));
   const syncMem = () => {
     const mems = H.unified[$("#usys").value] || [];
-    $("#umem").replaceChildren(...mems.map((m) => opt(null, m, `${m} GB`)));
+    $("#umem").replaceChildren(...mems.map((m) => opt(m, `${m} GB`)));
   };
-  $("#usys").addEventListener("change", syncMem);
+  $("#usys").addEventListener("change", () => { syncMem(); schedule(); });
   syncMem();
-  for (const c of H.contexts) $("#ctx").append(opt(null, c, `${c >= 1024 ? Math.round(c / 1024) + "k" : c} tokens`, c === 8192));
+  for (const c of H.contexts) $("#ctx").append(opt(c, `${ctxLabel(c)} tokens`, c === 16384));
+  for (const f of META.families) $("#family").append(opt(f.family, `${f.family} (${f.count})`));
+  for (const s of META.sizes) $("#size").append(opt(s, s));
 
-  $("#kind").addEventListener("change", syncKind);
+  $("#kind").addEventListener("change", () => { syncKind(); schedule(); });
   $("#gpu").addEventListener("change", syncKind);
-  syncKind();
-  let deb;
-  $("#go").addEventListener("click", search);
-  $("#search").addEventListener("input", () => { clearTimeout(deb); deb = setTimeout(search, 500); });
-  $$('input[name="kind"],input[name="p"]').forEach((i) => i.addEventListener("change", search));
-  for (const id of ["gpu", "gpu-count", "custom-vram", "ram", "usys", "umem", "ctx"]) {
-    document.getElementById(id).addEventListener("change", search);
+  $("#gpu2").addEventListener("change", syncKind);
+  $("#gpu2-toggle").addEventListener("click", () => {
+    $("#gpu2-wrap").hidden = !$("#gpu2-wrap").hidden;
+    syncKind(); schedule();
+  });
+  $("#more").addEventListener("click", () => search(true));
+  $("#search").addEventListener("input", () => schedule());
+  for (const sel of ["#gpu", "#custom-vram", "#gpu-count", "#gpu2", "#custom-vram2", "#gpu2-count",
+      "#ram", "#usys", "#umem", "#ctx", "#kv", "#mmproj", "#min-bits",
+      "#family", "#size", "#sort"]) {
+    $(sel).addEventListener("change", () => schedule());
   }
-  search();
+  for (const sel of ["#vram-only", "#fits-only", "#moe-only", "#vision-only"]) {
+    $(sel).addEventListener("change", () => schedule());
+  }
+  $$('input[name="kind"],input[name="p"]').forEach((i) => i.addEventListener("change", () => schedule()));
+  syncKind();
+  search(false);
 }
 
 boot();

index.html

@@ -4,82 +4,143 @@
 <meta charset="utf-8">
 <meta name="viewport" content="width=device-width, initial-scale=1">
 <title>Local Model Explorer + quant picks</title>
-<meta name="description" content="Which GGUF models fit your hardware, and exactly which quant to download. Fit data live; quant ranking from measured KL-divergence benchmarks.">
+<meta name="description" content="Which GGUF models fit your GPU, Mac or CPU, and exactly which quant to download for each.">
 <style>
   html,body{margin:0;background:#0d1217;color:#e8edf2;font:15px system-ui,sans-serif}
-  .shell{max-width:1080px;margin:0 auto;padding:20px}
-  header.top{display:flex;justify-content:space-between;align-items:center;flex-wrap:wrap;gap:8px;margin-bottom:12px}
+  .shell{max-width:1120px;margin:0 auto;padding:20px}
   header.top h1{font-size:20px;margin:0}
-  header.top p{margin:2px 0 0;color:#9aa7b4;font-size:13px}
-  .layout{display:grid;grid-template-columns:280px 1fr;gap:18px}
-  @media(max-width:860px){.layout{grid-template-columns:1fr}}
+  header.top h1 .tag{font-size:12px;border:1px solid #0e9aa5;color:#7fe0c3;border-radius:20px;padding:1px 9px;white-space:nowrap}
+  header.top p{margin:4px 0 14px;color:#9aa7b4;font-size:13.5px}
+  .layout{display:grid;grid-template-columns:290px 1fr;gap:18px}
+  @media(max-width:860px){.layout{grid-template-columns:1fr}.rig{position:static !important}}
   .rig{border:1px solid #1d2733;border-radius:10px;padding:14px;height:fit-content;position:sticky;top:12px}
-  .rig h2{font-size:15px;margin:0 0 10px}
+  .rig h2{font-size:15px;margin:0 0 6px}
   .field{margin:10px 0}
   .field label{display:block;font-size:12.5px;color:#9aa7b4;margin-bottom:4px}
   .field select,.field input[type=search]{width:100%;background:#131c25;color:#e8edf2;border:1px solid #2a3542;border-radius:7px;padding:7px 9px;font-size:14px;box-sizing:border-box}
   .seg{display:flex;gap:6px;flex-wrap:wrap}
   .seg label{border:1px solid #2a3542;border-radius:8px;padding:6px 10px;cursor:pointer;font-size:13px}
-  .seg input{accent-color:#0e9aa5}
+  .seg input,.check input{accent-color:#0e9aa5}
+  .check{display:flex;gap:7px;align-items:center;font-size:13.5px;margin:8px 0;cursor:pointer}
+  .row2{display:grid;grid-template-columns:1fr 1fr;gap:8px}
   .btn{background:#0e9aa5;border:0;color:#fff;border-radius:8px;padding:8px 14px;font-size:14px;cursor:pointer}
-  .btn.ghost{background:none;border:1px solid #2a3542;color:#e8edf2}
+  .btn.ghost{background:none;border:1px solid #2a3542;color:#e8edf2;font-size:13px;padding:5px 12px}
+  .btn.small{font-size:12.5px;padding:4px 10px}
+  .filters{display:flex;gap:8px;flex-wrap:wrap;margin-bottom:6px}
+  .filters .grow{flex:1;min-width:180px}
+  .filters select,.filters input[type=search]{background:#131c25;color:#e8edf2;border:1px solid #2a3542;border-radius:7px;padding:7px 9px;font-size:14px}
+  .toggles{display:flex;gap:12px;align-items:center}
   .card{border:1px solid #1d2733;border-radius:10px;padding:14px 16px;margin:0 0 12px}
-  .card h3{margin:0 0 2px;font-size:17px}
+  .card h3{margin:0;font-size:17px}
   .card h3 button{background:none;border:0;color:#e8edf2;font-size:17px;cursor:pointer;padding:0;text-align:left}
-  .meta{color:#9aa7b4;font-size:13px;margin:2px 0 8px}
+  .meta{color:#9aa7b4;font-size:13px;margin:3px 0 8px}
+  .r-side{display:flex;gap:14px;align-items:baseline;margin:6px 0}
+  .score{font-size:22px;font-weight:700}
   .choice{font-size:14.5px;margin:6px 0}
   .choice b{font-size:16px}
-  .badge{display:inline-block;padding:1px 9px;border-radius:20px;font-size:12px;border:1px solid #2a3542;color:#9aa7b4;white-space:nowrap}
-  .b-default{color:#7fe0c3;border-color:#0e9aa5}.b-archival,.b-step{color:#8fb6e6}
-  .b-tight,.b-knee{color:#e8d27f}.b-emergency,.b-avoid{color:#e88a7f}
   .warn{color:#e8b93e;font-size:13px;margin:4px 0}
   .muted{color:#9aa7b4;font-size:13px}
+  .rail{display:flex;gap:3px;height:10px;margin:8px 0 2px}
+  .rail .pool{position:relative;background:#1d2733;border-radius:5px;overflow:hidden;height:100%}
+  .rail i{position:absolute;left:0;top:0;bottom:0}
+  .rail i.gpu{background:#0e9aa5}.rail i.ram{background:#d69532}.rail i.exp{background:#7f6ce0}.rail i.over{background:#e85a5a}
+  .legend{display:flex;gap:14px;color:#9aa7b4;font-size:12.5px;margin:8px 0}
+  .legend i{display:inline-block;width:10px;height:10px;border-radius:3px;margin-right:5px}
   table.quants{width:100%;border-collapse:collapse;font-size:13px;margin-top:8px}
   table.quants th,table.quants td{text-align:left;padding:5px 8px;border-bottom:1px solid #1d2733}
-  table.quants td.num{text-align:right;font-variant-numeric:tabular-nums}
+  table.quants td.num,table.quants th.num{text-align:right;font-variant-numeric:tabular-nums}
   tr.is-pick td{background:#12242b}
   pre.cmd{background:#080c10;border:1px solid #1d2733;border-radius:8px;padding:10px;overflow-x:auto;font-size:12.5px}
   .row{display:flex;gap:8px;align-items:center;flex-wrap:wrap}
+  .kv-strip{display:flex;gap:12px;flex-wrap:wrap;font-size:13px;color:#9aa7b4}
+  .kv-strip b{color:#e8edf2}
   footer{color:#9aa7b4;font-size:12.5px;margin-top:24px}
   footer a{color:#8fb6e6}
   #count{color:#9aa7b4}
+  #more{display:block;margin:12px auto}
 </style>
 </head>
 <body>
 <div class="shell">
   <header class="top">
-    <div><h1>Local Model Explorer <span class="badge b-default">+ quant picks</span></h1>
-    <p>Same models-and-fit catalog, but every model gets the exact best quant for your machine.</p></div>
+    <h1>Local Model Explorer <span class="tag">+ quant picks</span></h1>
+    <p>Set your GPU, Mac or CPU. Each model names its exact best quant for your machine, with the command to run it.</p>
   </header>
   <div class="layout">
-    <aside class="rig">
+    <aside class="rig" aria-label="Your hardware">
       <h2>Your hardware</h2>
       <div class="seg" id="kind">
         <label><input type="radio" name="kind" value="gpu" checked> GPU</label>
-        <label><input type="radio" name="kind" value="unified"> Mac/unified</label>
-        <label><input type="radio" name="kind" value="cpu"> CPU</label>
+        <label><input type="radio" name="kind" value="unified"> Mac / unified</label>
+        <label><input type="radio" name="kind" value="cpu"> CPU only</label>
       </div>
       <div class="field" data-show="gpu"><label for="gpu">Graphics card</label><select id="gpu"></select></div>
-      <div class="field" data-show="gpu" id="custom-vram-field" hidden><label for="custom-vram">VRAM per card (GB)</label><select id="custom-vram"></select></div>
-      <div class="field" data-show="gpu"><label for="gpu-count">Number of cards</label><select id="gpu-count"></select></div>
+      <div class="row2" data-show="gpu">
+        <div class="field" id="custom-vram-field" hidden><label for="custom-vram">VRAM/card (GB)</label><select id="custom-vram"></select></div>
+        <div class="field"><label for="gpu-count">Cards</label><select id="gpu-count"></select></div>
+      </div>
+      <div data-show="gpu" id="gpu2-wrap" hidden>
+        <div class="field"><label for="gpu2">Other card</label><select id="gpu2"></select></div>
+        <div class="row2">
+          <div class="field" id="custom-vram2-field" hidden><label for="custom-vram2">VRAM/card (GB)</label><select id="custom-vram2"></select></div>
+          <div class="field"><label for="gpu2-count">Cards</label><select id="gpu2-count"></select></div>
+        </div>
+      </div>
+      <div data-show="gpu"><button type="button" class="btn ghost small" id="gpu2-toggle">Add a different card</button></div>
       <div class="field" data-show="unified"><label for="usys">System</label><select id="usys"></select></div>
       <div class="field" data-show="unified"><label for="umem">Memory</label><select id="umem"></select></div>
       <div class="field" data-show="gpu cpu"><label for="ram">System RAM</label><select id="ram"></select></div>
       <div class="field"><label for="ctx">Context length</label><select id="ctx"></select></div>
+      <div class="field"><label for="kv">KV cache precision</label><select id="kv">
+        <option value="f16">Full (f16), llama.cpp default</option>
+        <option value="q8_0">8-bit (q8_0)</option>
+        <option value="q4_0">4-bit (q4_0)</option>
+      </select></div>
+      <div class="field"><label for="mmproj">Vision projector (vision models)</label><select id="mmproj">
+        <option value="gpu">On the GPU, llama-server default</option>
+        <option value="ram">In system RAM (--no-mmproj-offload)</option>
+        <option value="off">Don't load it, text only (--no-mmproj)</option>
+      </select></div>
       <div class="field"><label>Prefer</label><div class="seg" id="priority">
         <label><input type="radio" name="p" value="balanced" checked> Balanced</label>
         <label><input type="radio" name="p" value="quality"> Quality</label>
         <label><input type="radio" name="p" value="small"> Smallest</label>
       </div></div>
-      <div class="field"><label for="search">Search models</label><input type="search" id="search" placeholder="e.g. qwen3, gemma, 70b"></div>
-      <div class="row"><button class="btn" id="go">Find models</button></div>
+      <div class="field"><label for="min-bits">Lowest quant to consider</label><select id="min-bits">
+        <option value="0">Any, down to 1-bit</option>
+        <option value="2">2-bit and up</option>
+        <option value="3">3-bit and up</option>
+        <option value="4">4-bit and up (Q4, IQ4, MXFP4)</option>
+        <option value="5">5-bit and up</option>
+        <option value="6">6-bit and up</option>
+        <option value="8">8-bit and up</option>
+      </select></div>
+      <label class="check" data-show="gpu"><input type="checkbox" id="vram-only"> Keep the whole model in VRAM</label>
+      <label class="check"><input type="checkbox" id="fits-only"> Only models that run on this hardware</label>
     </aside>
     <section>
+      <div class="filters">
+        <div class="grow"><input type="search" id="search" placeholder="Search models, e.g. qwen3, gemma, 70b" maxlength="80" style="width:100%"></div>
+        <select id="family"><option value="">All families</option></select>
+        <select id="size"><option value="">Any size</option></select>
+        <select id="sort">
+          <option value="best">Best for my hardware</option>
+          <option value="popular">Most downloaded</option>
+          <option value="recent">Newest</option>
+          <option value="community">Most community reports</option>
+        </select>
+        <div class="toggles">
+          <label class="check"><input type="checkbox" id="moe-only"> MoE</label>
+          <label class="check"><input type="checkbox" id="vision-only"> Vision</label>
+        </div>
+      </div>
+      <div class="legend"><span><i style="background:#0e9aa5"></i>On GPU</span><span><i style="background:#7f6ce0"></i>Experts in RAM</span><span><i style="background:#d69532"></i>In RAM</span><span><i style="background:#e85a5a"></i>Over capacity</span></div>
       <p id="count"></p>
       <div id="results"></div>
+      <button type="button" id="more" class="btn" hidden>Show more models</button>
     </section>
   </div>
-  <footer>Fit + catalog data live from <a href="https://huggingface.co/spaces/LocalLLaMA/local-model-explorer">codelion's Local Model Explorer</a>. Quant ranking from <a href="https://huggingface.co/spaces/spl16/local-model-explorer">this Space's</a> bench file <a href="quant_scores.json">quant_scores.json</a> (KL-divergence vs full precision, 150 GGUFs).</footer>
+  <footer>Model catalog and fit estimates live from <a href="https://huggingface.co/spaces/LocalLLaMA/local-model-explorer">codelion's Local Model Explorer</a>. Quant picks use measured KL-divergence benchmarks (<a href="quant_scores.json">quant_scores.json</a>): full offloads preferred, dominated formats skipped when a better twin is in the same repo.</footer>
 </div>
 <script src="rankQuants.js"></script>
 <script src="fork.js"></script>

rankQuants.js

@@ -6,8 +6,13 @@
  * size alone. Pure functions, no DOM: wire into the existing
  * recommend/model handlers; fit estimation stays untouched.
  *
+ * Fit awareness: full offloads (tier gpu/gpu_moe/cpu/unified/ram) always
+ * outrank partial ones. A partial "quality" pick (e.g. 12 of 64 layers on
+ * GPU) runs 3-10x slower with no speed data on the page to warn you, so
+ * the ranker only goes partial when nothing fits fully, with a warning.
+ *
  * Option shape (maps 1:1 to the /api/model option objects):
- *   { quant, repo, size, fit: { tier }, repo_downloads }
+ *   { quant, repo, size, fit: { tier, ngl, layers }, repo_downloads }
  * fit.tier "no" / "no_vram" count as not fitting.
  *
  * Usage:
@@ -18,6 +23,8 @@
 "use strict";
 
 const FAIL_TIERS = new Set(["no", "no_vram"]);
+// tiers where the whole model is resident (GPU, shared memory, or RAM)
+const FULL_TIERS = new Set(["gpu", "gpu_moe", "cpu", "unified", "ram"]);
 const KNOWN_REFS = new Set(["BF16", "F16", "FP16"]);
 const BAND_PREFERENCE = {
   balanced: ["default", "tight", "knee", "step", "archival"],
@@ -47,7 +54,9 @@ function scoreMap(scoresJson) {
 function classifyUnknown(quant) {
   const q = quant.toUpperCase();
   if (/Q4_0|Q4_1|Q3_K_S|Q3_K_M|Q5_K_S|Q4_K_S|Q4_K_L|Q5_0|Q3_0/.test(q)) return ["avoid", 0.10];
-  if (/IQ1|1BIT|_I1/.test(q)) return ["emergency", 0.60];
+  const bit = q.match(/(\d(\.\d)?)\s*BIT/);
+  if (bit) return parseFloat(bit[1]) < 4 ? ["emergency", 0.50] : ["tight", 0.12];
+  if (/IQ1|1BIT|_I1|TERNARY|Q1(?!\d)/.test(q)) return ["emergency", 0.60];
   if (/Q2|IQ2/.test(q)) return ["emergency", 0.40];
   if (/IQ3|Q3/.test(q)) return ["tight", 0.12];
   if (/Q4|IQ4|MXFP4/.test(q)) return ["default", 0.05];
@@ -93,15 +102,23 @@ function rankOptions(options, priority, family, scoresJson) {
   const map = scoreMap(scoresJson);
   let pool = [];
   for (const o of options) {
-    if (o.fit && FAIL_TIERS.has(o.fit.tier)) continue;
+    if (!o.fit || FAIL_TIERS.has(o.fit.tier)) continue;
     const [kl, band, measured] = qualityOf(o.quant, map);
-    pool.push({ ...o, kl, band, measured });
+    pool.push({ ...o, kl, band, measured,
+      full: FULL_TIERS.has(o.fit.tier) });
   }
   if (!pool.length) return { pick: null, ranked: [], warnings: ["nothing-fits"] };
   pool = pool.filter((o) => o.band !== "ref");
   if (!pool.length) return { pick: null, ranked: [], warnings: ["nothing-fits"] };
   const warnings = [];
 
+  // full offloads first; partial only when nothing fits fully
+  if (pool.some((o) => o.full)) {
+    pool = pool.filter((o) => o.full);
+  } else {
+    warnings.push("partial-offload: nothing fits fully; these picks split layers between GPU and RAM and run slower");
+  }
+
   const nonEmergency = pool.filter((o) => o.band !== "emergency");
   if (nonEmergency.length) {
     pool = nonEmergency;