Instructions to use qgfvadfuvads/Q-Prefer-D2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use qgfvadfuvads/Q-Prefer-D2 with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-VL-4B-Instruct") model = PeftModel.from_pretrained(base_model, "qgfvadfuvads/Q-Prefer-D2") - Notebooks
- Google Colab
- Kaggle
| { | |
| "date": "2026-08-12", | |
| "scope": "fresh release CUDA forward versus original D2 preprocessing and saved legacy GPU predictions", | |
| "environment": { | |
| "release_forward_device": "NVIDIA RTX 6000 Ada Generation", | |
| "release_forward_dtype": "bfloat16", | |
| "torch": "2.8.0+cu128", | |
| "transformers": "5.3.0", | |
| "peft": "0.18.1", | |
| "qwen_vl_utils": "0.0.14" | |
| }, | |
| "executed_fresh_forward": true, | |
| "model_load_seconds": 24.421089079231024, | |
| "artifact": { | |
| "adapter_sha256": "3f3788b942cb8e259582ee6cb2ac7c7bb2592a63e7183fe89c8434aef86476cf", | |
| "adapter_tensor_count": 505, | |
| "special_token_embeddings_sha256": "52c5aa200d63d093e9d46b16cfbcca14f1b00e0baf4fb5708de414d01511007f", | |
| "reward_head_shape": [3, 2560] | |
| }, | |
| "real_pair_checks": [ | |
| { | |
| "task": "t2v", | |
| "pair_id": "480p_helios_longcat_000009", | |
| "original_and_release_preprocessing_elementwise_equal": true, | |
| "preprocessing_max_abs_diff": 0.0, | |
| "fresh_cuda_video_scores": { | |
| "video_a": { | |
| "visual_quality": -3.7902991771698, | |
| "text_alignment": -5.020113468170166 | |
| }, | |
| "video_b": { | |
| "visual_quality": 0.4633995294570923, | |
| "text_alignment": -4.430180072784424 | |
| } | |
| }, | |
| "fresh_cuda_margins": { | |
| "visual_quality": -4.253698706626892, | |
| "text_alignment": -0.5899333953857422 | |
| }, | |
| "saved_legacy_gpu_margins": { | |
| "visual_quality": -4.253698706626892, | |
| "text_alignment": -0.5899333953857422 | |
| }, | |
| "absolute_margin_difference": { | |
| "visual_quality": 0.0, | |
| "text_alignment": 0.0 | |
| }, | |
| "inference_seconds": 0.8286775168962777 | |
| }, | |
| { | |
| "task": "i2v", | |
| "pair_id": "720p_skyreels_wan_A bald man in a black polo shirt and jeans stands beside a red convertible car w_5b1d6b8b", | |
| "includes_reference_image": true, | |
| "original_and_release_preprocessing_elementwise_equal": true, | |
| "preprocessing_max_abs_diff": 0.0, | |
| "fresh_cuda_video_scores": { | |
| "video_a": { | |
| "visual_quality": -0.981022834777832, | |
| "text_alignment": -2.165818214416504 | |
| }, | |
| "video_b": { | |
| "visual_quality": -0.30839744210243225, | |
| "text_alignment": -1.5762799978256226 | |
| } | |
| }, | |
| "fresh_cuda_margins": { | |
| "visual_quality": -0.6726253926753998, | |
| "text_alignment": -0.5895382165908813 | |
| }, | |
| "saved_legacy_gpu_margins": { | |
| "visual_quality": -0.6726253926753998, | |
| "text_alignment": -0.5895382165908813 | |
| }, | |
| "absolute_margin_difference": { | |
| "visual_quality": 0.0, | |
| "text_alignment": 0.0 | |
| }, | |
| "inference_seconds": 0.39105592016130686 | |
| } | |
| ], | |
| "recomputed_saved_prediction_metrics": { | |
| "t2v_quality_with_ties": 0.5933, | |
| "t2v_quality_without_ties": 0.7796, | |
| "t2v_alignment_with_ties": 0.4647, | |
| "t2v_alignment_without_ties": 0.625, | |
| "i2v_quality_with_ties": 0.5758, | |
| "i2v_quality_without_ties": 0.7791, | |
| "i2v_alignment_with_ties": 0.5247, | |
| "i2v_alignment_without_ties": 0.7401, | |
| "main_macro_with_ties": 0.5396, | |
| "main_macro_without_ties": 0.7309 | |
| }, | |
| "limitations": [ | |
| "This file records the initial two-pair tensor and forward check; the later full fresh-CUDA runs are summarized in release_validation_summary.json.", | |
| "Raw full-run predictions and media-derived scores remain local and are intentionally excluded from Git.", | |
| "Re-run frozen-pair parity after changing CUDA, PyTorch, Transformers, PEFT, or the attention implementation." | |
| ] | |
| } | |