update

hao-ai-lab · Dec 31, 2024 · b2a75a5 · b2a75a5
1 parent 68a38cc
commit b2a75a5
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 12 deletions.
diff --git a/fastvideo/distill.py b/fastvideo/distill.py
@@ -54,8 +54,9 @@ def reshard_fsdp(model):
 
 
 def get_norm(model_pred, norms, gradient_accumulation_steps):
-    fro_norm = (torch.linalg.matrix_norm(model_pred, ord="fro") / # codespell:ignore
-                gradient_accumulation_steps)
+    fro_norm = (
+        torch.linalg.matrix_norm(model_pred, ord="fro") /  # codespell:ignore
+        gradient_accumulation_steps)
     largest_singular_value = (torch.linalg.matrix_norm(model_pred, ord=2) /
                               gradient_accumulation_steps)
     absolute_mean = torch.mean(
@@ -65,7 +66,7 @@ def get_norm(model_pred, norms, gradient_accumulation_steps):
     dist.all_reduce(fro_norm, op=dist.ReduceOp.AVG)
     dist.all_reduce(largest_singular_value, op=dist.ReduceOp.AVG)
     dist.all_reduce(absolute_mean, op=dist.ReduceOp.AVG)
-    norms["fro"] += torch.mean(fro_norm).item() # codespell:ignore
+    norms["fro"] += torch.mean(fro_norm).item()  # codespell:ignore
     norms["largest singular value"] += torch.mean(
         largest_singular_value).item()
     norms["absolute mean"] += absolute_mean.item()
@@ -100,7 +101,7 @@ def distill_one_step(
     total_loss = 0.0
     optimizer.zero_grad()
     model_pred_norm = {
-        "fro": 0.0, # codespell:ignore
+        "fro": 0.0,  # codespell:ignore
         "largest singular value": 0.0,
         "absolute mean": 0.0,
         "absolute max": 0.0,
@@ -555,7 +556,7 @@ def get_num_phases(multi_phased_distill_schedule, step):
                     "grad_norm":
                     grad_norm,
                     "pred_fro_norm":
-                    pred_norm["fro"], # codespell:ignore
+                    pred_norm["fro"],  # codespell:ignore
                     "pred_largest_singular_value":
                     pred_norm["largest singular value"],
                     "pred_absolute_mean":

diff --git a/fastvideo/models/hunyuan/modules/models.py b/fastvideo/models/hunyuan/modules/models.py
@@ -608,11 +608,11 @@ def forward(
         txt = encoder_hidden_states[:, 1:]
         text_states_2 = encoder_hidden_states[:, 0, :self.config.
                                               text_states_dim_2]
-        _, _, ot, oh, ow = x.shape # codespell:ignore
+        _, _, ot, oh, ow = x.shape  # codespell:ignore
         tt, th, tw = (
-            ot // self.patch_size[0], # codespell:ignore
-            oh // self.patch_size[1], # codespell:ignore
-            ow // self.patch_size[2], # codespell:ignore
+            ot // self.patch_size[0],  # codespell:ignore
+            oh // self.patch_size[1],  # codespell:ignore
+            ow // self.patch_size[2],  # codespell:ignore
         )
         original_tt = nccl_info.sp_size * tt
         freqs_cos, freqs_sin = self.get_rotary_pos_embed((original_tt, th, tw))

diff --git a/fastvideo/models/mochi_hf/modeling_mochi.py b/fastvideo/models/mochi_hf/modeling_mochi.py
@@ -488,8 +488,10 @@ def _create_rope(self, freqs: torch.Tensor,
                      pos: torch.Tensor) -> torch.Tensor:
         with torch.autocast(freqs.device.type, enabled=False):
             # Always run ROPE freqs computation in FP32
-            freqs = torch.einsum("nd,dhf->nhf", pos.to(torch.float32), # codespell:ignore
-                                 freqs.to(torch.float32))
+            freqs = torch.einsum(
+                "nd,dhf->nhf",
+                pos.to(torch.float32),  # codespell:ignore
+                freqs.to(torch.float32))
         freqs_cos = torch.cos(freqs)
         freqs_sin = torch.sin(freqs)
         return freqs_cos, freqs_sin

diff --git a/fastvideo/utils/env_utils.py b/fastvideo/utils/env_utils.py
@@ -26,7 +26,7 @@
     if is_torch_npu_available():
         info["PyTorch version"] += " (NPU)"
         info["NPU type"] = torch.npu.get_device_name()
-        info["CANN version"] = torch.version.cann # codespell:ignore
+        info["CANN version"] = torch.version.cann  # codespell:ignore
 
     try:
         import bitsandbytes