Skip to content
VersatIL
qat_policy
Initializing search
Lorenzo-Mazza/VersatIL
VersatIL
Lorenzo-Mazza/VersatIL
Home
Getting Started
Getting Started
Installation
First Training Run
Configuration
Architecture
Architecture
Overview
Encoding Pipeline
Algorithms
Action Decoders
Data Pipeline
Inference
Explainability
Quantization
Post-Training Compression
Papers Reproducibility Instructions
Papers Reproducibility Instructions
Supervised MoE Surgical Grasping
API Reference
API Reference
versatil
versatil
checkpoint_loading
checkpoint_loading
base
compressed_policy
float_policy
qat_policy
qat_policy
Table of contents
qat_policy
common
common
dict_of_tensor_mixin
logging
module_attr_mixin
module_state
omegaconf_ops
set_cache_dir
tensor_ops
configs
configs
adaptation
data
data
augmentations
dataloader
metadata
raw
raw
schema
zarr_meta
task
tokenizer
decoding
decoding
action_head
algorithm
decoder
latent
deployment
encoding
encoding
encoder
fusion
pipeline
experiment
explainability
inference_client
loss
main
paths
policy
post_training_compression
quantization
store
store
policy
quantization
root
task
training
training
data
data
constants
dataloader
episodic_dataset
metadata
normalization
normalization
image_normalizer
normalizer
preprocessing
preprocessing
codecs
create_zarr_arrays
create_zarr_from_csv
create_zarr_from_hdf5
create_zarr_from_lerobot
create_zarr_from_synthetic
replay_buffer
resizers
sampler
processing
processing
action_processor
image_processor
transform
transform_builder
raw
raw
schemas
schemas
base
csv
custom
custom
libero
synthetic
tso
hdf5
lerobot
zarr_meta
sample_builder
synthetic
synthetic
constants
generators
renderer
task_layout
visualization
task
tokenization
tokenization
action_discretizer
action_token_id_mapping
action_tokenizer
binned_value_discretizer
fast
huggingface
observation_tokenizer
tokenizer
explainability
explainability
attribution
attribution
ablation
capture
gradients
maps
objectives
policy
constants
explanation_heatmaps
online_runtime
runner
sources
sources
dataset
online
schema_paths
typedefs
typedefs
vision_modules
visualization
writer
inference
inference
action_postprocessor
inference_client
observation_buffer
observation_preprocessor
policy_runtime
policy_runtime
base
compressed_runtime
executorch_adapter
float_runtime
protocol
socket_transport
synthetic_rollout
temporal_aggregation
metrics
metrics
accumulators
base
constants
kernels
losses
losses
classification
composite
divergence
gripper
latent_geometry
maximum_mean_discrepancy
metadata_passthrough
mixture
mixture_of_experts
optimal_transport
prior_denoising
regression
trajectory
vector_quantization
synthetic_metrics
models
models
adaptation
adaptation
constants
lora
decoding
decoding
action_heads
action_heads
base
blocks
conditional
gaussian
moe
single_output
action_masking
algorithm
algorithm
base
behavior_cloning
diffusion
flow_matching
variational
constants
decoders
decoders
autoregressive_mixin
base
discrete
factory
factory
act
action_transformer
autoregressive_vla
conditional_action_unet
diffusion_action_transformer
dit_block_action_transformer
gpt_action_transformer
lact
mode_act
openvla_oft
phase_act
pi0
smolvla
interleaved_vlm
llm_prefix_suffix_attention
moe
parallel_transformer
timestep_conditioning
vlm
generative_language_models
generative_language_models
base
constants
vision_language
vision_language
base
huggingface
paligemma
prismatic
smolvlm
latent
latent
posterior
posterior
base_posterior
transformer_encoder
vq_encoder
prior
prior
base_prior
codebook_prior
dit_prior
gaussian_prior
latent_standardizer
state_condition_pool
transformer_encoder
uniform_codebook_prior
vamp_prior
protocols
reparametrize
vq
vq
euclidean_codebook
residual_vq
vector_quantize
mixture_of_experts
transformer_input_builder
unet_input_builder
encoding
encoding
encoders
encoders
base
conditional
constants
cross_modal
cross_modal
rgbd
rgbd
dformerv2
geometric_rgbd
vision_language
vision_language
vlm_encoder
depth
depth
spatial
image_mixin
language
language
language
language_mixin
proprioceptive
proprioceptive
base
rgb
rgb
conditional_cnn
dinov2_siglip
flat
spatial
spatial_backbone
unconditional
explainability
fusion
fusion
attention
base
concat
mlp
pipeline
exportable_policy
feature_meta
input_specification
layers
layers
activation
conditional_unet
constants
convert_layers
convolution
convolution
conv1d
depthwise_conv2d
denoising
denoising
conditional_flow_matching
diffusion_process
ode_solvers
timestep_sampling
detr_transformer
detr_transformer
attention
transformer
transformer_decoder
transformer_encoder
diffusion_transformer
diffusion_transformer
cross_attention_dit
dit_block_transformer
final_prediction_layer
mmdit_transformer
drop_path
dynamic_feature_embedding
feature_projection
free_transformer
free_transformer
binary_mapper
free_transformer
frozen_batchnorm
gated_linear_unit
geometric_attention
geometric_attention
depth_decay
geometric_attention
geometric_attention_encoder
geometric_bias
spatial_decay
mlp
modulation
modulation
conditional_modulation
conditional_residual_block
film_residual_block
normalization
normalization
ada_norm
constants
factory
rms_norm
typedefs
unconditioned_norm
patch_embedding
pooling
pooling
attention_pooling
pooling_head
spatial_softmax
positional_encoding
positional_encoding
base
learned
rotary
sinusoidal
transformer
transformer
attention
attention
cached_attention
joint_attention
joint_attention_base
precomputed_joint_attention
query_key_norm
autoregressive_decoder
bidirectional_decoder
block
block
base
cross_attention
dual_stream_attention
dual_stream_base
feedforward
precomputed_cross_attention
precomputed_dual_stream_attention
self_attention
cache
cache
conditioning
generation
conditional_bidirectional_decoder
dual_stream_decoder
encoder
layer
layer
decoder_layer
dual_stream_layer
encoder_layer
precomputed_dual_stream_layer
precomputed_kv_layer
masking
positional_encoding
transformer_mixin
policy
post_training_compression
post_training_compression
compression_target
compressor
constants
deployment_backends
deployment_backends
base
executorch_xnnpack
torch_inductor
export
policy_context
policy_loading
preparation
preparation
batchnorm
fusion
pruning
pruning
base
structured
unstructured
report
serialization
quantization
quantization
calibration
constants
module_target
pt2e
pt2e
backends
backends
base
x86_inductor
xnnpack
torch_patches
workflows
workflows
base
eager
none
pt2e
training
training
callback_factory
callbacks
callbacks
confusion_matrix
early_stopping
ema
expert_usage
gradient_norm
latent_visualization
prior_target_standardization
provider
reduce_lr_on_plateau
synthetic_rollout
training_stage
wandb_figure
constants
lightning_policy
stage
validation
workspace
Known Issues
Changelog
Table of contents
qat_policy
qat_policy
qat_policy
¶
Quantization-Aware-Trained policy checkpoint restoration.
Back to top