| { | |
| "upstream_draft": "z-lab/Qwen3.8-27B-DFlash2", | |
| "upstream_revision": "50307d4c4cde6860d4eee73e2547cd786fe8e8a4", | |
| "license_at_pinned_revision": "apache-2.0", | |
| "compatible_target": "matching Qwen3.8-27B planner with target tokenizer, image processor, and chat template; target weights not included", | |
| "training_complete": true, | |
| "epoch": 2, | |
| "optimizer_step": 126, | |
| "draft_parameters": 1924404480, | |
| "num_hidden_layers": 5, | |
| "trained_block_size": 8, | |
| "selected_inference_block_size": 12, | |
| "dtype_of_exported_tensors": "BF16", | |
| "full_finetuning": true, | |
| "target_frozen": true, | |
| "lora": false, | |
| "train_examples": 1999, | |
| "validation_examples": 251, | |
| "test_examples": 250, | |
| "connected_split_groups": 205, | |
| "learning_rate": 2e-05, | |
| "global_batch_size": 32, | |
| "anchors_per_example": 8, | |
| "ddp_world_size": 8, | |
| "target_layer_ids": [ | |
| 5, | |
| 19, | |
| 33, | |
| 47, | |
| 61 | |
| ], | |
| "validation_ce": [ | |
| 1.3144736349772383, | |
| 0.8333338581543247, | |
| 0.8122866464276635 | |
| ] | |
| } | |