SCOPD / models.json
enmingzhangzz's picture
Publish evaluation-and-model-card
a47010e verified
Raw History Blame Contribute Delete
10.8 kB
{
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"models": [
{
"id": "scopd",
"label": "SCOPD",
"objective": "forward KL: KL(teacher || student)",
"selection": "all response tokens",
"adapter_sha256": "0f882974e3e8565c7d9a99494e11289736c384e0d60fb9ba9e20ed5417a7b6a9",
"source_revision": "c7182c3867f9f8819b23af4bd57ba785c4daf5c8",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus",
"label": "SCOPD+",
"objective": "forward KL: KL(teacher || student)",
"selection": "top 10% budget-JSD response tokens",
"selection_fraction": 0.1,
"budget_delta": 0.01,
"adapter_sha256": "04c4590b14355f3767ebaa63a435f42def65fa83d386624b6ec1a75a23315e8d",
"source_revision": "863f3f0097db7eb4b9bca1a14d862edf138d407f",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus-top20",
"label": "SCOPD+ Top20",
"objective": "forward KL: KL(teacher || student)",
"selection": "top 20% budget-JSD response tokens",
"selection_fraction": 0.2,
"budget_delta": 0.01,
"adapter_sha256": "4f70e51e1c9cbfe953f7277786c82a32c6b754919bb6c8307fa2012f1ab8ec57",
"source_revision": "52b62edc3c51c310a75278ed53ca0187ed333a82",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus-top20",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus-top40",
"label": "SCOPD+ Top40",
"objective": "forward KL: KL(teacher || student)",
"selection": "top 40% budget-JSD response tokens",
"selection_fraction": 0.4,
"budget_delta": 0.01,
"adapter_sha256": "3d8791563dd4a732b3f9fe072a6f2588faa0bc192de131794d4914c60a658497",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus-top40",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus-top60",
"label": "SCOPD+ Top60",
"objective": "forward KL: KL(teacher || student)",
"selection": "top 60% budget-JSD response tokens",
"selection_fraction": 0.6,
"budget_delta": 0.01,
"adapter_sha256": "c02297cfe2fcd3ab01d769d0aca53cb4a3bb43dba6892895ecc282f23b053b3a",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus-top60",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "sft",
"label": "SFT",
"objective": "supervised fine-tuning",
"selection": "supervised response tokens",
"adapter_sha256": "9463250e17a196055c2f54e205587c7e727c3b5b6b7a30eb3158c5967ccb9f8e",
"source_revision": "0c0d37ac6d9cbc0243dbf431d426e16931e5c79f",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/sft",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "epic",
"label": "EPIC",
"objective": "official EPIC configuration",
"selection": "EPIC response supervision",
"adapter_sha256": "86acec7d328a0a895c695dbca7582f38ef0ddda40817f000e82830c50c38555b",
"source_revision": "0d0fe8bd4f4a1b280937d97a9c2839c814cf6a9d",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/epic",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "grpo",
"label": "GRPO",
"objective": "group-relative policy optimization",
"selection": "reward-based policy objective",
"adapter_sha256": "c38704d69eb2e1ac0164d9ec1be644d42449ff80eb130175855550c94745bcf4",
"optimizer_updates": 2560,
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/grpo",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-offpolicy",
"label": "SCOPD Off-Policy",
"objective": "forward KL on teacher-generated responses",
"selection": "all response tokens",
"adapter_sha256": "dbf02470f09a41bfe8bad808a17f96ba892c582e7510d12daf2df0e1047b9281",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-offpolicy",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus-reverse-kl",
"label": "SCOPD+ Reverse KL",
"objective": "reverse KL: KL(student || teacher)",
"selection": "top 10% budget-JSD response tokens",
"selection_fraction": 0.1,
"budget_delta": 0.01,
"adapter_sha256": "1a7c4004928c80f85a66d216c6d17b17005fe89740da946d7597225775b00baa",
"source_revision": "dc2490eaf9b8b35b7e5553080dd985d0ab6a86a3",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus-reverse-kl",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "scopd-plus-jsd",
"label": "SCOPD+ JSD",
"objective": "symmetric JSD(teacher, student)",
"selection": "top 10% budget-JSD response tokens",
"selection_fraction": 0.1,
"budget_delta": 0.01,
"adapter_sha256": "ea78a9ae00822a76a55824846b385662d9baae5998dcffa928a45be9f177c954",
"base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
"base_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/scopd-plus-jsd",
"tensor_count": 392,
"parameter_count": 40370176,
"weight_bytes": 161480704,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "qwen3-vl-4b-scopd",
"label": "Qwen3-VL-4B SCOPD",
"objective": "forward KL: KL(teacher || student)",
"selection": "all response tokens",
"adapter_sha256": "6aec6fba5176359d4d1a3d906d3722a51b9f7f21c79d5d5989120b075915102f",
"optimizer_updates": 320,
"base_model": "Qwen/Qwen3-VL-4B-Instruct",
"base_revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/qwen3-vl-4b-scopd",
"tensor_count": 504,
"parameter_count": 33030144,
"weight_bytes": 132120576,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
},
{
"id": "qwen3-vl-4b-scopd-plus",
"label": "Qwen3-VL-4B SCOPD+",
"objective": "forward KL: KL(teacher || student)",
"selection": "top 10% budget-JSD response tokens",
"selection_fraction": 0.1,
"budget_delta": 0.01,
"adapter_sha256": "febccb411829f5c314d421038c1cd89dcc543df074978778771504ab1dc91a0e",
"optimizer_updates": 320,
"base_model": "Qwen/Qwen3-VL-4B-Instruct",
"base_revision": "ebb281ec70b05090aa6165b016eac8ec08e71b17",
"merged": false,
"format": "peft_lora",
"dtype": "float32",
"inference_dtype": "bfloat16",
"questions_processed": 10240,
"path": "models/qwen3-vl-4b-scopd-plus",
"tensor_count": 504,
"parameter_count": 33030144,
"weight_bytes": 132120576,
"lora_rank": 16,
"lora_alpha": 32,
"lora_dropout": 0.0
}
]
}