change "repeat" to "repeat_interleave" and "tile" uniformly

examples/animatediff/ad/modules/attention.py

@@ -284,7 +284,7 @@ def construct(self, q, k, v, mask):
             else:
                 finfo_type = np.float32
             max_neg_value = -np.finfo(finfo_type).max
-            mask = mask.repeat(self.heads, axis=0)
+            mask = mask.repeat_interleave(self.heads, dim=0)
             mask = ops.expand_dims(mask, axis=1)
             sim.masked_fill(mask, max_neg_value)
 

examples/animatediff/ad/modules/diffusionmodules/util.py

@@ -189,7 +189,7 @@ def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
         if dim % 2:
             embedding = ops.concat((embedding, ops.ZerosLike()(embedding[:, :1])), axis=-1)
     else:
-        embedding = ops.reshape(timesteps.repeat(dim), (-1, dim))
+        embedding = ops.reshape(timesteps.repeat_interleave(dim), (-1, dim))
     return embedding
 
 

examples/animatediff/ad/utils/cond_data.py

@@ -24,7 +24,7 @@ def transform_conditional_images(image_paths, H, W, random_crop=True, normalize=
     if normalize:
 
         def image_norm(image):
-            image = image.mean(dim=0, keepdim=True).repeat(3, 1, 1)
+            image = image.mean(axis=0, keepdims=True).repeat(3, axis=0)
             image -= image.min()
             image /= image.max()
             return image

examples/dynamicrafter/lvdm/modules/attention.py

@@ -324,7 +324,7 @@ def construct(self, q, k, v, k_ip, v_ip, out_ip, mask):
             else:
                 finfo_type = np.float32
             max_neg_value = -np.finfo(finfo_type).max
-            mask = mask.repeat(self.head_num, axis=0)
+            mask = mask.repeat_interleave(self.head_num, dim=0)
             mask = ops.expand_dims(mask, axis=1)
             sim.masked_fill(mask, max_neg_value)
 

examples/dynamicrafter/lvdm/modules/networks/util.py

@@ -184,7 +184,7 @@ def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
         if dim % 2:
             embedding = ops.concat((embedding, ops.ZerosLike()(embedding[:, :1])), axis=-1)
     else:
-        embedding = ops.reshape(timesteps.repeat(dim), (-1, dim))
+        embedding = ops.reshape(timesteps.repeat_interleave(dim), (-1, dim))
     return embedding
 
 

examples/instantmesh/models/renderer/synthesizer.py

@@ -195,7 +195,7 @@ def forward_grid(self, planes, grid_size: int, aabb: ms.Tensor = None):
                     dtype=planes.dtype,
                 )
                 .unsqueeze(0)
-                .repeat(planes.shape[0], 1, 1)
+                .tile((planes.shape[0], 1, 1))
             )
         assert planes.shape[0] == aabb.shape[0], "Batch size mismatch for planes and aabb"
         N = planes.shape[0]

examples/magvit/tools/inflate_vae2d_to_3d.py

@@ -60,7 +60,7 @@ def inflate(vae2d_ckpt, save_fp):
                 shape_3d_new = tuple([shape_3d[0] // 2]) + shape_3d[1:]
                 new_w = ms.ops.zeros(shape_3d_new, dtype=weights.dtype)
                 new_w[:, :, -1, :, :] = weights
-                new_w = new_w.repeat(2, 0)
+                new_w = new_w.tile((2, 1, 1, 1, 1))
 
                 new_w = ms.Parameter(new_w, name=key_3d)
                 new_state_dict[key_3d] = new_w

examples/opensora_hpcai/opensora/utils/cond_data.py

@@ -49,7 +49,7 @@ def transform_conditional_images(image_paths, H, W, random_crop=True, normalize=
     if normalize:
 
         def image_norm(image):
-            image = image.mean(dim=0, keepdim=True).repeat(3, 1, 1)
+            image = image.mean(axis=0, keepdims=True).repeat(3, axis=0)
             image -= image.min()
             image /= image.max()
             return image

examples/opensora_pku/opensora/eval/cal_fvd.py

@@ -6,7 +6,7 @@
 def trans(x):
     # if greyscale images add channel
     if x.shape[-3] == 1:
-        x = x.repeat(1, 1, 3, 1, 1)
+        x = x.tile((1, 1, 3, 1, 1))
 
     # permute BTCHW -> BCTHW
     x = x.permute(0, 2, 1, 3, 4)

examples/opensora_pku/opensora/eval/cal_lpips.py

@@ -22,7 +22,7 @@
 def trans(x):
     # if greyscale images add channel
     if x.shape[-3] == 1:
-        x = x.repeat(3, axis=2)
+        x = x.tile((1, 1, 3, 1, 1))
 
     # value range [0, 1] -> [-1, 1]
     x = x * 2 - 1

examples/opensora_pku/opensora/models/causalvideovae/model/modules/quant.py

@@ -21,7 +21,7 @@ def _tile(self, x):
         if d < self.n_codes:
             n_repeats = (self.n_codes + d - 1) // d
             std = 0.01 / np.sqrt(ew)
-            x = x.repeat(n_repeats, 1)
+            x = x.tile((n_repeats, 1))
             x = x + ops.randn_like(x) * std
         return x
 

examples/stable_diffusion_v2/depth_to_image.py

@@ -157,7 +157,7 @@ def prepare_conditions(depth, txt, num_samples=1, height=512, width=512, vae_sca
 
     # repeat to [bs, 1, h_z, w_z]
     depth = np.expand_dims(depth, axis=[0, 1])
-    depth = depth.repeat(num_samples, axis=0)
+    depth = depth.tile((num_samples, 1, 1, 1))
     assert len(depth.shape) == 4 and depth.shape[1] == 1, f"expect shape [n, 1, h, w], but got {depth.shape}"
 
     depth = Tensor(depth, dtype=mstype.float32)

examples/stable_diffusion_v2/inference/sd_lite_infer.py

@@ -128,7 +128,7 @@ def main(args):
         )
         init_image = Image.open(args.inputs.image_path).convert("RGB")
         img = img_processor.preprocess(init_image, height=args.inputs.H, width=args.inputs.W)
-        inputs["img"] = img.repeat(batch_size, axis=0).asnumpy()
+        inputs["img"] = img.tile((batch_size, 1, 1, 1)).asnumpy()
         init_timestep = min(int(args.sampling_steps * args.inputs.strength), args.sampling_steps)
         t_start = max(args.sampling_steps - init_timestep, 0)
         inputs["timesteps"] = inputs["timesteps"][t_start * scheduler.order :]

examples/stable_diffusion_v2/inpaint.py

@@ -45,10 +45,10 @@ def make_batch_sd(image, mask, txt, num_samples=1):
     masked_image = image * (mask < 0.5)
 
     batch = {
-        "image": image.repeat(num_samples, axis=0),
+        "image": image.repeat_interleave(num_samples, dim=0),
         "txt": num_samples * [txt],
-        "mask": mask.repeat(num_samples, axis=0),
-        "masked_image": masked_image.repeat(num_samples, axis=0),
+        "mask": mask.repeat_interleave(num_samples, dim=0),
+        "masked_image": masked_image.repeat_interleave(num_samples, dim=0),
     }
     return batch
 

examples/stable_diffusion_v2/ldm/modules/diffusionmodules/util.py

@@ -134,7 +134,7 @@ def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
         if dim % 2:
             embedding = ops.concat((embedding, ops.ZerosLike()(embedding[:, :1])), axis=-1)
     else:
-        embedding = ops.reshape(timesteps.repeat(dim), (-1, dim))
+        embedding = ops.reshape(timesteps.repeat_interleave(dim), (-1, dim))
     return embedding
 
 

examples/story_diffusion/utils/pipeline.py

@@ -406,9 +406,9 @@ def __call__(
                     prompt_embeds = self.id_encoder(id_pixel_values, prompt_embeds, class_tokens_mask)
 
                     bs_embed, seq_len, _ = prompt_embeds.shape
-                    prompt_embeds = prompt_embeds.repeat(num_images_per_prompt, axis=1)
+                    prompt_embeds = prompt_embeds.tile((1, num_images_per_prompt, 1))
                     prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
-                    pooled_prompt_embeds = pooled_prompt_embeds.repeat(num_images_per_prompt, axis=1).view(
+                    pooled_prompt_embeds = pooled_prompt_embeds.tile((1, num_images_per_prompt, 1)).view(
                         bs_embed * num_images_per_prompt, -1
                     )
                     pooled_prompt_embeds_arr.append(pooled_prompt_embeds)
@@ -472,7 +472,7 @@ def __call__(
                 text_encoder_projection_dim=text_encoder_projection_dim,
             )
             add_time_ids = ops.cat([add_time_ids, add_time_ids], axis=0)
-            add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, axis=0)
+            add_time_ids = add_time_ids.tile((batch_size * num_images_per_prompt, 1))
 
             # print(latents.shape)
             # print(add_time_ids.shape)

examples/sv3d/modules/temporal_ae.py

@@ -152,7 +152,7 @@ def construct(self, x: Tensor, timesteps: Tensor, skip_video: bool = False):
         x = x.transpose(0, 2, 3, 1).reshape(b, -1, c)  # b c h w -> b (h w) c
 
         x_mix = x
-        num_frames = ops.arange(timesteps).repeat(b // timesteps)
+        num_frames = ops.arange(timesteps).repeat_interleave(b // timesteps)
         t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False)
         emb = self.video_time_embed(t_emb)  # b, n_channels
         emb = emb[:, None, :]

examples/sv3d/sgm/modules/diffusionmodules/guiders.py

@@ -40,7 +40,7 @@ def construct(self, x: Tensor, sigma: Tensor) -> Tensor:
         x_c = x_c.reshape(-1, self.num_frames, *x_c.shape[1:])
 
         scale = ops.linspace(self.min_scale, self.max_scale, self.num_frames)[None, :]
-        scale = scale.repeat(x_u.shape[0], axis=0)  # 1 t -> b t
+        scale = scale.repeat_interleave(x_u.shape[0], dim=0)  # 1 t -> b t
         scale = append_dims(scale, x_u.ndim)
 
         out = x_u + scale * (x_c - x_u)

examples/sv3d/simple_video_sample.py

@@ -94,7 +94,7 @@ def get_batch(
             "cond_frames_without_noise": cond_frames_without_noise,
         }
         batch_uc = {}
-        batch["cond_aug"] = Tensor(self.cond_aug).repeat(math.prod([1, self.num_frames]))
+        batch["cond_aug"] = Tensor(self.cond_aug).repeat_interleave(math.prod([1, self.num_frames]))
 
         for key in batch.keys():
             if key not in batch_uc and isinstance(batch[key], Tensor):
@@ -128,10 +128,10 @@ def __call__(self, image: Tensor) -> Tensor:
 
         for k in ["crossattn", "concat"]:
             uc[k] = self.expand_dims_ops(uc[k], 1)
-            uc[k] = uc[k].repeat(self.num_frames, axis=1)
+            uc[k] = uc[k].repeat_interleave(self.num_frames, dim=1)
             uc[k] = uc[k].flatten(order="C", start_dim=0, end_dim=1)
             c[k] = self.expand_dims_ops(c[k], 1)
-            c[k] = c[k].repeat(self.num_frames, axis=1)
+            c[k] = c[k].repeat_interleave(self.num_frames, dim=1)
             c[k] = c[k].flatten(order="C", start_dim=0, end_dim=1)
 
         randn = Tensor(self.randn_n)

examples/svd/models/diffusion.py

@@ -97,8 +97,8 @@ def construct(
         tokens = (cond_frames_without_noise, fps_id, motion_bucket_id, cond_frames, cond_aug)
 
         vector, crossattn, concat = self.conditioner(*tokens)
-        crossattn = crossattn.repeat(num_frames, axis=0)
-        concat = concat.repeat(num_frames, axis=0)
+        crossattn = crossattn.repeat_interleave(num_frames, dim=0)
+        concat = concat.repeat_interleave(num_frames, dim=0)
 
         c_skip, c_out, c_in, c_noise = self.denoiser(sigmas, noised_input.ndim)
         model_output = self.model(

examples/svd/modules/diffusionmodules/guiders.py

@@ -29,7 +29,7 @@ def construct(self, x: Tensor, sigma: Tensor, num_frames: int) -> Tensor:
         x_c = x_c.reshape(-1, num_frames, *x_c.shape[1:])
 
         scale = ops.linspace(self.min_scale, self.max_scale, num_frames)[None, :]
-        scale = scale.repeat(x_u.shape[0], axis=0)  # 1 t -> b t
+        scale = scale.repeat_interleave(x_u.shape[0], dim=0)  # 1 t -> b t
         scale = append_dims(scale, x_u.ndim)
 
         out = x_u + scale * (x_c - x_u)

examples/svd/modules/encoders/modules.py

@@ -42,8 +42,8 @@ def construct(
             sigmas = self.sigma_sampler(b)
             if self.sigma_cond is not None:
                 sigma_cond = self.sigma_cond(sigmas)
-                sigma_cond = sigma_cond.repeat(self.n_copies, axis=0)  # b d -> (b t) d
-            sigmas = sigmas.repeat(self.n_copies, axis=0)  # b -> (b t)
+                sigma_cond = sigma_cond.repeat_interleave(self.n_copies, dim=0)  # b d -> (b t) d
+            sigmas = sigmas.repeat_interleave(self.n_copies, dim=0)  # b -> (b t)
             noise = ops.randn_like(vid)
             vid = vid + noise * append_dims(sigmas, vid.ndim)
 
@@ -61,7 +61,7 @@ def construct(
 
         vid = vid.reshape(-1, self.n_cond_frames, *vid.shape[1:])  # (b t) c h w -> b t c h w
         vid = vid.reshape(vid.shape[0], -1, *vid.shape[3:])  # b t c h w -> b (t c) h w
-        vid = vid.repeat(self.n_copies, axis=0)  # b (t c) h w -> (b s) (t c) h w
+        vid = vid.repeat_interleave(self.n_copies, dim=0)  # b (t c) h w -> (b s) (t c) h w
 
         if self.sigma_cond is not None:
             return vid, sigma_cond
@@ -84,6 +84,6 @@ def __init__(
     def construct(self, vid: Tensor) -> Tensor:
         vid = self.open_clip(vid)
         vid = vid.reshape(-1, self.n_cond_frames, vid.shape[1])  # (b t) d -> b t d
-        vid = vid.repeat(self.n_copies, axis=0)  # b t d -> (b s) t d
+        vid = vid.repeat_interleave(self.n_copies, dim=0)  # b t d -> (b s) t d
 
         return vid

examples/svd/modules/temporal_ae.py

@@ -148,7 +148,7 @@ def construct(self, x: Tensor, timesteps: Tensor, skip_video: bool = False):
         x = x.transpose(0, 2, 3, 1).reshape(b, -1, c)  # b c h w -> b (h w) c
 
         x_mix = x
-        num_frames = ops.arange(timesteps).repeat(b // timesteps)
+        num_frames = ops.arange(timesteps).repeat_interleave(b // timesteps)
         t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False)
         emb = self.video_time_embed(t_emb)  # b, n_channels
         emb = emb[:, None, :]

examples/svd/modules/unet3d.py

@@ -292,9 +292,9 @@ def construct(
 
             time_context = context
             time_context_first_timestep = time_context[::timesteps]
-            time_context = time_context_first_timestep.repeat(h * w, axis=0)  # b ... -> (b n) ...
+            time_context = time_context_first_timestep.repeat_interleave(h * w, dim=0)  # b ... -> (b n) ...
         elif time_context is not None and not self.use_spatial_context:
-            time_context = time_context.repeat(h * w, axis=0)  # b ... -> (b n) ...
+            time_context = time_context.repeat_interleave(h * w, dim=0)  # b ... -> (b n) ...
             if time_context.ndim == 2:
                 time_context = time_context.expand_dims(1)  # b c -> b 1 c
 

examples/t2v_turbo/intern_vid2/models/backbones/bert/xbert.py

@@ -985,7 +985,7 @@ def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple
             if is_decoder:
                 batch_size, seq_length = input_shape
                 seq_ids = ops.arange(seq_length)
-                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                causal_mask = seq_ids[None, None, :].tile((batch_size, seq_length, 1)) <= seq_ids[None, :, None]
                 # in case past_key_values are used we need to add a prefix ones mask to the causal mask
                 # causal and attention masks must have same type with pytorch version < 1.3
                 causal_mask = causal_mask.to(attention_mask.dtype)

examples/t2v_turbo/lvdm/modules/encoders/ip_resampler.py

@@ -132,7 +132,7 @@ def __init__(
             )
 
     def construct(self, x):
-        latents = self.latents.repeat(x.shape[0], 1, 1)
+        latents = self.latents.tile((x.shape[0], 1, 1))
 
         x = self.proj_in(x)
 

examples/t2v_turbo/pipeline/t2v_turbo_ms_pipeline.py

@@ -67,7 +67,7 @@ def _encode_prompt(
 
         bs_embed, seq_len, _ = prompt_embeds.shape
         # duplicate text embeddings for each generation per prompt, using mps friendly method
-        prompt_embeds = prompt_embeds.repeat(num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.tile((1, num_videos_per_prompt, 1))
         prompt_embeds = prompt_embeds.view(bs_embed * num_videos_per_prompt, seq_len, -1)
 
         # Don't need to get uncond prompt embedding because of LCM Guided Distillation
@@ -177,7 +177,7 @@ def __call__(
         bs = batch_size * num_videos_per_prompt
 
         # 6. Get Guidance Scale Embedding
-        w = ms.Tensor(guidance_scale).repeat(bs)
+        w = ms.Tensor(guidance_scale).tile((bs,))
         w_embedding = self.get_w_embedding(w, embedding_dim=256)
 
         # 7. LCM MultiStep Sampling Loop:

examples/t2v_turbo/pipeline/t2v_turbo_vc2_pipeline.py

@@ -56,7 +56,7 @@ def _encode_prompt(
 
         bs_embed, seq_len, _ = prompt_embeds.shape
         # duplicate text embeddings for each generation per prompt, using mps friendly method
-        prompt_embeds = prompt_embeds.repeat(num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.tile((1, num_videos_per_prompt, 1))
         prompt_embeds = prompt_embeds.view(bs_embed * num_videos_per_prompt, seq_len, -1)
 
         # Don't need to get uncond prompt embedding because of LCM Guided Distillation
@@ -171,7 +171,7 @@ def __call__(
         bs = batch_size * num_videos_per_prompt
 
         # 6. Get Guidance Scale Embedding
-        w = ms.Tensor(guidance_scale).repeat(bs)
+        w = ms.Tensor(guidance_scale).tile((bs,))
         w_embedding = self.get_w_embedding(w, embedding_dim=256)
 
         # 7. LCM MultiStep Sampling Loop:

examples/t2v_turbo/viclip/viclip_vision.py

@@ -219,11 +219,11 @@ def inflate_weight(weight_2d, time_dim, center=True):
     logger.info(f"Init center: {center}")
     if center:
         weight_3d = ops.zeros(*weight_2d.shape)
-        weight_3d = weight_3d.unsqueeze(2).repeat(1, 1, time_dim, 1, 1)
+        weight_3d = weight_3d.unsqueeze(2).tile((1, 1, time_dim, 1, 1))
         middle_idx = time_dim // 2
         weight_3d[:, :, middle_idx, :, :] = weight_2d
     else:
-        weight_3d = weight_2d.unsqueeze(2).repeat(1, 1, time_dim, 1, 1)
+        weight_3d = weight_2d.unsqueeze(2).tile((1, 1, time_dim, 1, 1))
         weight_3d = weight_3d / time_dim
     return weight_3d
 

examples/videocomposer/vc/models/attention.py

@@ -133,7 +133,7 @@ def rearange_in(x):
             else:
                 finfo_type = np.float32
             max_neg_value = -np.finfo(finfo_type).max
-            mask = mask.repeat(self.heads, axis=0)
+            mask = mask.repeat_interleave(self.heads, dim=0)
             mask = ops.expand_dims(mask, axis=1)
             sim.masked_fill(mask, max_neg_value)
 

mindone/models/dit.py

@@ -299,7 +299,9 @@ def construct(self, x, mask=None):
         if mask is not None:
             if mask.ndim == 2:
                 # mask shape is (batch_size, key_len)
-                mask = ops.expand_dims(mask, axis=1).repeat(q_n, axis=1)  # (b, k_n) -> (b, 1, k_n) -> (b, q_n, k_n)
+                mask = ops.expand_dims(mask, axis=1).repeat_interleave(
+                    q_n, dim=1
+                )  # (b, k_n) -> (b, 1, k_n) -> (b, q_n, k_n)
                 mask = ops.select(
                     ~mask,
                     ops.ones((q_b, q_n, k_n), self.dtype) * (-ms.numpy.inf),
@@ -308,7 +310,9 @@ def construct(self, x, mask=None):
             elif mask.ndim == 3:
                 # mask shape is (batch_size, query_len, key_len), the query_len maybe one
                 if mask.shape[-2] == 1:
-                    mask = mask.repeat(q_n, axis=-1)  # manually broadcast to key length to avoid FA shape error
+                    mask = mask.repeat_interleave(
+                        q_n, dim=-1
+                    )  # manually broadcast to key length to avoid FA shape error
                 assert mask.shape[-2] == q_n, "Expect mask shape to be (bs, query_len, key_len), "
                 f"but the mask query length {mask.shape[-2]} is different from the input query length {q_n}"
                 assert mask.shape[-1] == k_n, "Expect mask shape to be (bs, query_len, key_len), "
@@ -335,7 +339,7 @@ def construct(self, x, mask=None):
             k = self._rearange_in(k, h)
             v = self._rearange_in(v, h)
             if mask is not None and mask.shape[0] != q.shape[0]:
-                mask = mask.repeat(h, axis=0)  # (b, q_n, k_n) -> (b*h, q_n, k_n)
+                mask = mask.repeat_interleave(h, dim=0)  # (b, q_n, k_n) -> (b*h, q_n, k_n)
             out = self.attention(q, k, v, mask)
             # (b*h, n, d) -> (b, n, h*d)
             out = self._rearange_out(out, h)

mindone/models/mmdit.py

@@ -184,7 +184,7 @@ def construct(self, x, c, mask=None, spatial_freq=None):
             k = ops.reshape(k, (B * h, T + L, hd))
             v = ops.reshape(v, (B * h, T + L, hd))
             if mask is not None and mask.shape[0] != q.shape[0]:
-                mask = mask.repeat(h, axis=0)
+                mask = mask.repeat_interleave(h, dim=0)
             out = self.attention(q, k, v, mask)
             out = out.swapaxes(1, 2).view(B, T + L, C)  # b, nh, T+L, hd -> b, T+L, nh*hd=C