RouteFM / manifest.json
AIGNLAI's picture
Add RouteFM paper, figures, and citation
0d533c0 verified
Raw History Blame Contribute Delete
2.92 kB
{
"schema_version": 2,
"name": "RouteFM",
"model_version": "1.0.0",
"package_version": "1.0.1",
"license": "Apache-2.0",
"paper": {
"title": "Pretrain Once, Route Anywhere: Towards a Foundation Model for LLM Routing",
"authors": [
"Guannan Lai",
"Han-Jia Ye"
],
"arxiv": "2609.37362",
"url": "https://arxiv.org/abs/2609.37362",
"doi": "10.48550/arXiv.2609.37362",
"published": "2026-09-29"
},
"source": {
"repository": "https://github.com/LAMDA-Model-Reuse/RouteFM",
"model_source_tag": "v1.0.0",
"model_source_commit": "4a9e4dbbae791d8d24e06d4b90fe46edc562ed80"
},
"hub": {
"repository": "AIGNLAI/RouteFM",
"artifact_revision": "b364300234794b0e6394453d224ddab7c94e41d0",
"model_tag": "model-v1.0.0"
},
"training": "one continuous 10,000-step random-initialization run per encoder",
"encoders": {
"qwen": {
"canonical_weight": "qwen/model.safetensors",
"canonical_weight_sha256": "881b648ead20565da38a7f6ee9516dd0a45d3747add41c44d300ac24433b9cb8",
"canonical_weight_size_bytes": 51145068,
"config": "qwen/config.json",
"config_sha256": "0a0f63b954fa3e4472fdd8c51abff33bbc6a12bf0ea267eab0a4604ebcaf6231",
"legacy_weight": "legacy/routefm_qwen.pt",
"legacy_weight_sha256": "e28d6203e99cf5f10f42293c561b80d17e866bc7bd69a696928ea836e6b7cccc",
"source_checkpoint_sha256": "5e7ad02b565718a1598f7392a32ee244b1eb66c3170d0d65a1f348ddd751d09b",
"training_config": "configs/pretrain_qwen.json",
"training_config_sha256": "2372f59ef6f8cf5ea8c121dd5dc1102a5276f1251ec532112d3fd2330834b9da",
"embedding_dimension": 4096,
"step": 10000,
"parameters": 12781059
},
"bge": {
"canonical_weight": "bge/model.safetensors",
"canonical_weight_sha256": "3245698f2f6bd4cb607cc3af8875d63ba04172f33abd25b286d1b2968c679662",
"canonical_weight_size_bytes": 44302700,
"config": "bge/config.json",
"config_sha256": "6cdc9e9d5ccad4ed9e2543e26ea5adf7def5830189a25e3349ae183c5e7535ea",
"legacy_weight": "legacy/routefm_bge.pt",
"legacy_weight_sha256": "0f7ce1b269499a6938a5722511a489eb1a701e17dfd2605f49d47a19edeffcde",
"source_checkpoint_sha256": "d177e8255e9f01441b9b1fd6f8d853dbc391074fa912c8484351bae8bc086577",
"training_config": "configs/pretrain_bge.json",
"training_config_sha256": "38a9703ae24eea5b4bd2c3a7f651c82d756d3ac3dbe20daf7c0b1122fd16b2a8",
"embedding_dimension": 768,
"step": 10000,
"parameters": 11070467
}
},
"mmrbench": {
"version": "MMR-Bench V1",
"split": "published seed-31010 within-dataset 40:60 IDs",
"status": "retrospective post-selected illustrative split"
},
"external_assets_not_included": [
"pretraining source datasets",
"MMR-Bench V1 records and images",
"Qwen3-VL embedding service/model",
"BGE-base-en-v1.5 model"
]
}