Upload metrics/GAN_FINETUNE_METRICS.md with huggingface_hub

metrics/GAN_FINETUNE_METRICS.md

@@ -125,12 +125,12 @@ starved by DPG/training sharing the same nodes; awaiting a dedicated card).
 | baseline         | 25k | 0.7062 | 75.464 | 0.6320 | 29.162 |
 | gan_patchgan     | 25k | 0.7017 | 75.422 | 0.6270 | 29.288 |
 | gan_dinov2_text  | 25k | 0.7666 | 76.489 | 0.7100 | **25.526** |
-| gan_dinov2       | 25k | 0.7803 | 76.373 | 0.7149 | — |
+| gan_dinov2       | 25k | 0.7803 | 76.373 | 0.7149 | 27.845 |
 | gan_dinov2_50    | 25k | 0.7750 | **76.640** | 0.7217 | 27.319 |
 | **gan_dinov2_t035** | 25k | **0.7827** | 76.577 | **0.7263** | 27.661 |
 
 (All generated with the 25-step teacher `gen_coco_teacher.py`, scored with pyiqa (NR-IQA) + clean-fid
-(pFID) over the same 10k COCO prompts. `gan_dinov2` pFID still missing — its 10k images weren't regenerated.)
+(pFID) over the same 10k COCO prompts. All seven models now have full NR-IQA + pFID.)
 
 **Takeaways**
 - **DINOv2-disc GANs dominate perceptual quality**; **`gan_dinov2_t035` now leads NR-IQA** (best TOPIQ 0.7827