tangger/lerobot
1
0
Fork 1
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Update pre-commits (#733)

Browse Source
This commit is contained in:
Simon Alibert
2025-02-15 15:51:17 +01:00
committed by GitHub
parent 2cb0bf5d41
commit c4c2ce04e7
16 changed files with 69 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
lerobot/common/policies/act/modeling_act.py
View File

@@ -409,9 +409,9 @@ class ACT(nn.Module):
latent dimension.
"""
if self.config.use_vae and self.training:
assert (
"action" in batch
), "actions must be provided when using the variational objective in training mode."
assert "action" in batch, (
"actions must be provided when using the variational objective in training mode."
)
batch_size = (
batch["observation.images"]

2
lerobot/common/policies/diffusion/configuration_diffusion.py
View File

@@ -221,7 +221,7 @@ class DiffusionConfig(PreTrainedConfig):
for key, image_ft in self.image_features.items():
if image_ft.shape != first_image_ft.shape:
raise ValueError(
f"`{key}` does not match `{first_image_key}`, but we " "expect all image shapes to match."
f"`{key}` does not match `{first_image_key}`, but we expect all image shapes to match."
)
@property

6
lerobot/common/policies/tdmpc/modeling_tdmpc.py
View File

@@ -594,9 +594,9 @@ class TDMPCTOLD(nn.Module):
self.apply(_apply_fn)
for m in [self._reward, *self._Qs]:
assert isinstance(
m[-1], nn.Linear
), "Sanity check. The last linear layer needs 0 initialization on weights."
assert isinstance(m[-1], nn.Linear), (
"Sanity check. The last linear layer needs 0 initialization on weights."
)
nn.init.zeros_(m[-1].weight)
nn.init.zeros_(m[-1].bias) # this has already been done, but keep this line here for good measure

2
lerobot/common/policies/vqbet/configuration_vqbet.py
View File

@@ -184,7 +184,7 @@ class VQBeTConfig(PreTrainedConfig):
for key, image_ft in self.image_features.items():
if image_ft.shape != first_image_ft.shape:
raise ValueError(
f"`{key}` does not match `{first_image_key}`, but we " "expect all image shapes to match."
f"`{key}` does not match `{first_image_key}`, but we expect all image shapes to match."
)
@property

44
lerobot/common/policies/vqbet/vqbet_utils.py
View File

@@ -203,9 +203,9 @@ class GPT(nn.Module):
def forward(self, input, targets=None):
device = input.device
b, t, d = input.size()
assert (
t <= self.config.gpt_block_size
), f"Cannot forward sequence of length {t}, block size is only {self.config.gpt_block_size}"
assert t <= self.config.gpt_block_size, (
f"Cannot forward sequence of length {t}, block size is only {self.config.gpt_block_size}"
)
# positional encodings that are added to the input embeddings
pos = torch.arange(0, t, dtype=torch.long, device=device).unsqueeze(0) # shape (1, t)
@@ -273,10 +273,10 @@ class GPT(nn.Module):
assert len(inter_params) == 0, "parameters {} made it into both decay/no_decay sets!".format(
str(inter_params)
)
assert (
len(param_dict.keys() - union_params) == 0
), "parameters {} were not separated into either decay/no_decay set!".format(
str(param_dict.keys() - union_params),
assert len(param_dict.keys() - union_params) == 0, (
"parameters {} were not separated into either decay/no_decay set!".format(
str(param_dict.keys() - union_params),
)
)
decay = [param_dict[pn] for pn in sorted(decay)]
@@ -419,9 +419,9 @@ class ResidualVQ(nn.Module):
# and the network should be able to reconstruct
if quantize_dim < self.num_quantizers:
assert (
self.quantize_dropout > 0.0
), "quantize dropout must be greater than 0 if you wish to reconstruct from a signal with less fine quantizations"
assert self.quantize_dropout > 0.0, (
"quantize dropout must be greater than 0 if you wish to reconstruct from a signal with less fine quantizations"
)
indices = F.pad(indices, (0, self.num_quantizers - quantize_dim), value=-1)
# get ready for gathering
@@ -472,9 +472,9 @@ class ResidualVQ(nn.Module):
all_indices = []
if return_loss:
assert not torch.any(
indices == -1
), "some of the residual vq indices were dropped out. please use indices derived when the module is in eval mode to derive cross entropy loss"
assert not torch.any(indices == -1), (
"some of the residual vq indices were dropped out. please use indices derived when the module is in eval mode to derive cross entropy loss"
)
ce_losses = []
should_quantize_dropout = self.training and self.quantize_dropout and not return_loss
@@ -887,9 +887,9 @@ class VectorQuantize(nn.Module):
# only calculate orthogonal loss for the activated codes for this batch
if self.orthogonal_reg_active_codes_only:
assert not (
is_multiheaded and self.separate_codebook_per_head
), "orthogonal regularization for only active codes not compatible with multi-headed with separate codebooks yet"
assert not (is_multiheaded and self.separate_codebook_per_head), (
"orthogonal regularization for only active codes not compatible with multi-headed with separate codebooks yet"
)
unique_code_ids = torch.unique(embed_ind)
codebook = codebook[:, unique_code_ids]
@@ -999,9 +999,9 @@ def gumbel_sample(
ind = sampling_logits.argmax(dim=dim)
one_hot = F.one_hot(ind, size).type(dtype)
assert not (
reinmax and not straight_through
), "reinmax can only be turned on if using straight through gumbel softmax"
assert not (reinmax and not straight_through), (
"reinmax can only be turned on if using straight through gumbel softmax"
)
if not straight_through or temperature <= 0.0 or not training:
return ind, one_hot
@@ -1209,9 +1209,9 @@ class EuclideanCodebook(nn.Module):
self.gumbel_sample = gumbel_sample
self.sample_codebook_temp = sample_codebook_temp
assert not (
use_ddp and num_codebooks > 1 and kmeans_init
), "kmeans init is not compatible with multiple codebooks in distributed environment for now"
assert not (use_ddp and num_codebooks > 1 and kmeans_init), (
"kmeans init is not compatible with multiple codebooks in distributed environment for now"
)
self.sample_fn = sample_vectors_distributed if use_ddp and sync_kmeans else batched_sample_vectors
self.kmeans_all_reduce_fn = distributed.all_reduce if use_ddp and sync_kmeans else noop