Upload folder using huggingface_hub

predict.py

@@ -151,18 +151,23 @@ def predict_next_frame(ens, context_frames: np.ndarray) -> np.ndarray:
 
             direct_pred = _predict_8frames_direct(ens.pong_direct, context_tensor, last_tensor)
 
-            ar_preds = []
-            ctx = context_tensor.clone()
-            last_t = last_tensor.clone()
-            for step in range(PRED_FRAMES):
-                predicted = _predict_ar_frame(ens.models["pong"], ctx, last_t)
-                ar_preds.append(predicted)
-                ctx_frames = ctx.reshape(1, CONTEXT_FRAMES, 3, 64, 64)
-                ctx_frames = torch.cat([ctx_frames[:, 1:], predicted.unsqueeze(1)], dim=1)
-                ctx = ctx_frames.reshape(1, -1, 64, 64)
-                last_t = predicted
+            # Multi-run Pong AR with tiny noise diversity
+            all_pong_ar_runs = []
+            for noise_std in [0.0, 0.25/255.0, 0.5/255.0]:
+                ar_preds_run = []
+                ctx = context_tensor.clone()
+                last_t = last_tensor.clone()
+                for step in range(PRED_FRAMES):
+                    ctx_in = ctx if noise_std == 0 else torch.clamp(ctx + torch.randn_like(ctx) * noise_std, 0, 1)
+                    predicted = _predict_ar_frame(ens.models["pong"], ctx_in, last_t)
+                    ar_preds_run.append(predicted)
+                    ctx_frames = ctx.reshape(1, CONTEXT_FRAMES, 3, 64, 64)
+                    ctx_frames = torch.cat([ctx_frames[:, 1:], predicted.unsqueeze(1)], dim=1)
+                    ctx = ctx_frames.reshape(1, -1, 64, 64)
+                    last_t = predicted
+                all_pong_ar_runs.append(torch.stack(ar_preds_run, dim=1))
 
-            ar_pred = torch.stack(ar_preds, dim=1)
+            ar_pred = sum(all_pong_ar_runs) / len(all_pong_ar_runs)
 
             predicted = torch.zeros_like(direct_pred)
             for step in range(PRED_FRAMES):
@@ -195,19 +200,12 @@ def predict_next_frame(ens, context_frames: np.ndarray) -> np.ndarray:
             context_tensor = torch.from_numpy(context).to(DEVICE)
             last_tensor = torch.from_numpy(last_frame_t).to(DEVICE)
 
+            direct_orig = _predict_8frames_direct(ens.sonic_direct, context_tensor, last_tensor)
             context_flipped = torch.flip(context_tensor, dims=[3])
             last_flipped = torch.flip(last_tensor, dims=[3])
-
-            # Multi-run direct with noise diversity
-            all_direct_runs = []
-            for noise_std in [0.0, 0.5/255.0, 1.0/255.0]:
-                ctx_in = context_tensor if noise_std == 0 else torch.clamp(context_tensor + torch.randn_like(context_tensor) * noise_std, 0, 1)
-                ctx_flip_in = context_flipped if noise_std == 0 else torch.clamp(context_flipped + torch.randn_like(context_flipped) * noise_std, 0, 1)
-                direct_orig = _predict_8frames_direct(ens.sonic_direct, ctx_in, last_tensor)
-                direct_flipped = _predict_8frames_direct(ens.sonic_direct, ctx_flip_in, last_flipped)
-                direct_flipped = torch.flip(direct_flipped, dims=[4])
-                all_direct_runs.append((direct_orig + direct_flipped) / 2.0)
-            direct_pred = sum(all_direct_runs) / len(all_direct_runs)
+            direct_flipped = _predict_8frames_direct(ens.sonic_direct, context_flipped, last_flipped)
+            direct_flipped = torch.flip(direct_flipped, dims=[4])
+            direct_pred = (direct_orig + direct_flipped) / 2.0
 
             # Multi-run AR with noise diversity
             all_ar_runs = []