File size: 5,695 Bytes
031a8dd 3439960 031a8dd e30cf0a 3439960 e30cf0a 3439960 e30cf0a 07594dc e30cf0a b6ac22b e30cf0a b6ac22b 3439960 b6ac22b 98851d9 3439960 e30cf0a 98851d9 3439960 e30cf0a 2f2c499 de8d4d4 98851d9 3439960 98851d9 3439960 031a8dd | 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 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | {
"schema_version": 1,
"model_name": "Winnow-12B",
"repository": "EldanRing/Winnow-12B",
"status": "GGUF release: BF16, Q8_0 and NVFP4 language models, optional F16 vision projector and optional matching12B BF16 GGUF MTP assistant",
"dataset_released": false,
"base_model": "google/gemma-4-12B-it",
"base_revision": "707f0a3b8a3c7ad586ed01e27eafbad8a27dd0f7",
"training": {
"method": "LoRA merged into the weights, exported as BF16/Q8_0 GGUF and quantized as NVFP4 GGUF; quantization adds no training",
"rank": 32,
"alpha": 64,
"dropout": 0.0,
"target_modules": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
],
"data_description": "Private mixture of synthetic decision scenarios, teacher-supervised examples and labeled semantic tasks; data and training pipeline are not released."
},
"license": "apache-2.0",
"inference_repository": "https://github.com/EldanRing/winnow-inference",
"conversion_repository": "https://github.com/ggml-org/llama.cpp",
"conversion_revision": "911f6cdc8ab8a530b2bee09ee61471a6f3178eeb",
"calibration": {
"decision_temperature": 1.0,
"fitted_posthoc_map": false
},
"artifacts": [
{
"path": "chat_template.jinja",
"sha256": "ae53464bf3be25802b3a5b37def7fd89667067d7577049b3b2d74c4d8de4c6d4",
"size_bytes": 18683,
"role": "model_asset"
},
{
"path": "config.json",
"sha256": "b50e760a9107b629e1a140ab7a4392e875c37baefee022f27c98b90b9d947cf0",
"size_bytes": 4953,
"role": "model_asset"
},
{
"path": "generation_config.json",
"sha256": "a8349d9bd64cc5841297fcb5002f0fdc4749c473c8f1b10ea337f9ce4ee7014e",
"size_bytes": 260,
"role": "model_asset"
},
{
"path": "processor_config.json",
"sha256": "6b938e76555b3e9946890770e1abcd442a4718f34041a58e8139dc8ad34545c9",
"size_bytes": 1382,
"role": "model_asset"
},
{
"path": "tokenizer.json",
"sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
"size_bytes": 32169626,
"role": "model_asset"
},
{
"path": "tokenizer_config.json",
"sha256": "a62f4e85a47c0c136edaaa3a4f591fd6783717299a9def47e5ad03a49f6a5eb9",
"size_bytes": 3089,
"role": "model_asset"
},
{
"path": "gguf/Winnow-12B-Q8_0.gguf",
"sha256": "b710efc4c0d048ee61eed92c5fef5ce323a4d17e7c51f9f0533cc72ae50818ea",
"size_bytes": 12669646592,
"role": "q8_0_gguf"
},
{
"path": "gguf/mmproj-Winnow-12B.gguf",
"sha256": "91f086971e56d7a7d8d39e271873fccdb49541bd259d6e02c401a4f1cb7a219e",
"size_bytes": 175115840,
"role": "vision_projector",
"precision": "F16",
"companion_for": [
"gguf/Winnow-12B-BF16.gguf",
"gguf/Winnow-12B-Q8_0.gguf",
"gguf/Winnow-12B-NVFP4.gguf"
]
},
{
"path": "gguf/Winnow-12B-BF16.gguf",
"sha256": "10e6b41e00a3fc6668b53a92fb5d9c85179f47145e0786e7a3bc7b5b09ba8968",
"size_bytes": 23832065792,
"role": "bf16_gguf"
},
{
"path": "gguf/Winnow-12B-NVFP4.gguf",
"sha256": "9aaf25d04b8ab8259be54b153c99e2ab799948256a10abf51b2cd79e665b1f2f",
"size_bytes": 8163448416,
"role": "nvfp4_gguf"
},
{
"path": "gguf/Gemma-4-12B-IT-Assistant-BF16.gguf",
"sha256": "dc630b2f3f3cb8c3170d865d200f414d2da91676ae37c1738e898f7ba92fbeae",
"size_bytes": 861519840,
"role": "mtp_assistant_bf16_gguf",
"source_model": "google/gemma-4-12B-it-assistant",
"source_revision": "46d4c6f13f0ac0ad827b915669b8df9b81c64c51",
"license": "apache-2.0",
"documentation": "docs/assistants/README.md"
}
],
"evaluation_reports": [
"docs/BENCHMARKS.md",
"https://github.com/EldanRing/winnow-inference/blob/main/docs/BENCHMARKS.md"
],
"bf16_benchmark_format": "BF16 GGUF; current download and recorded benchmark SHA256 are compared in bf16_export_verification",
"projector_description": "Optional F16 vision projector for BF16, Q8_0 or NVFP4; a companion file, not another language model.",
"local_test_profile": {
"gpu": "NVIDIA GeForce RTX 5070 Ti",
"gpu_memory_gb": 16,
"host_ram_gb": 96,
"context_positions": 65536,
"kv_cache": "q8_0",
"decision_parallel": 4,
"chat_parallel": 1,
"memory_policy": "exclusive"
},
"inference_revision": "6c2b3c04e248a319f2cb43832628eba03e55fe38",
"weight_format": "GGUF",
"available_model_precisions": [
"BF16",
"Q8_0",
"NVFP4"
],
"bf16_export_verification": {
"source_shards_match_previous_release": true,
"converter_revision": "911f6cdc8ab8a530b2bee09ee61471a6f3178eeb",
"previous_benchmark_sha256": "a7ec058c65b4e73da9f6066a0239b449c74d644c234a5f901aebb19e56de3a22",
"current_sha256": "10e6b41e00a3fc6668b53a92fb5d9c85179f47145e0786e7a3bc7b5b09ba8968",
"matches_previous_benchmark_file": false
},
"previous_model_artifact_revision": "b2b14213dfa252e6d6b543c8b334762e51772d29",
"training_pipeline_released": false,
"optional_inference": {
"repository": "https://github.com/EldanRing/winnow-inference",
"runtime_lock_sha256": "9560b74c9fd736c80880c2733569075e066e7e8f49a7b465c8a4589604c4f65e",
"context": 8192,
"native_parallel": 4,
"chat_parallel": 1,
"cache": "q8_0",
"scope": "Linux/CUDA singleGPU tested RTX5070Ti16GB; experimental adaptive calibration text only",
"supported_presets": [
"12b-nvfp4-vision8k-mtp",
"12b-q8-text8k-mtp"
],
"adaptive_policies": [
"nvfp4-entropy-v1",
"q8-fixed50-v1"
]
}
}
|