Upload configs/opd_12g.yaml with huggingface_hub

configs/opd_12g.yaml

@@ -0,0 +1,55 @@
+# OPD 蒸馏 - 适配 12G 显存 (4070 Ti)
+# 将 SFT box 专家 + SFT point 专家蒸馏为统一模型
+# 按任务路由 teacher: counting/spatial→box, maze/path→point
+# 使用 forward KL + 温度缩放，student 从 pretrain 初始化
+
+student_model: "outputs/sft_box/final"
+teacher_models:
+  - "outputs/sft_box/final"
+  - "outputs/sft_point/final"
+teacher_weights:
+  - 0.5
+  - 0.5
+
+freeze_vision_tower: true
+freeze_llm: false
+torch_dtype: "bfloat16"
+
+load_in_4bit: true
+load_in_8bit: false
+
+datasets:
+  - path: "data/sft/grounding/sft_grounding.jsonl"
+    image_root: "data/coco/train/coco/train"
+    task_type: "grounding"
+    weight: 2.0
+  - path: "data/sft/counting/counting_data.jsonl"
+    image_root: "data/coco/train/coco/train"
+    task_type: "counting"
+    weight: 2.0
+  - path: "data/sft/spatial/spatial_data.jsonl"
+    image_root: "data/sft/spatial"
+    task_type: "spatial"
+    weight: 1.0
+  - path: "data/sft/maze/maze_data.jsonl"
+    image_root: "data/sft/maze"
+    task_type: "maze"
+    weight: 1.0
+  - path: "data/sft/path/path_data.jsonl"
+    image_root: "data/sft/path"
+    task_type: "path"
+    weight: 1.0
+
+batch_size: 1
+gradient_accumulation_steps: 16
+epochs: 3
+learning_rate: 1.0e-6
+temperature: 1.0
+ce_coeff: 0.5
+weight_decay: 0.01
+warmup_ratio: 0.05
+max_length: 1024
+image_size: 336
+num_workers: 0
+save_every: 1
+device: "cuda"