"""
Pipeline for training and using a Flux1 model for simulation-based inference.
Examples:
.. code-block:: python
import grain
import numpy as np
import jax
from jax import numpy as jnp
from gensbi.recipes import Flux1Pipeline
# Define your training and validation datasets.
train_data = jax.random.rand((1024, 4)) # your training dataset
val_data = jax.random.rand((128, 4)) # your validation dataset
batch_size = 32
train_dataset_grain = (
grain.MapDataset.source(np.array(train_data)[...,None])
.shuffle(42)
.repeat()
.to_iter_dataset()
.batch(batch_size)
# .mp_prefetch() # Uncomment if you want to use multiprocessing prefetching
)
val_dataset_grain = (
grain.MapDataset.source(np.array(val_data)[...,None])
.shuffle(42)
.repeat()
.to_iter_dataset()
.batch(batch_size)
# .mp_prefetch() # Uncomment if you want to use multiprocessing prefetching
)
# Define the model
dim_theta = 2 # Dimension of the parameter space
dim_x = 2 # Dimension of the observation space
pipeline = Flux1Pipeline(train_dataset_grain, val_dataset_grain, dim_theta, dim_x)
# Train the model
rngs = jax.random.PRNGKey(0)
pipeline.train(rngs)
# Sample from the posterior
x_o = jnp.array([0.5, -0.2]) # Example
samples = pipeline.sample(rngs, x_o, nsamples=10000, step_size=0.01)
.. note::
If you plan on using multiprocessing prefetching, ensure that your script is wrapped in a `if __name__ == "__main__":` guard. See https://docs.python.org/3/library/multiprocessing.html
"""
import jax
import jax.numpy as jnp
from flax import nnx
import optax
from optax.contrib import reduce_on_plateau
from numpyro import distributions as dist
from tqdm.auto import tqdm
from functools import partial
import orbax.checkpoint as ocp
from gensbi.flow_matching.path import AffineProbPath
from gensbi.flow_matching.path.scheduler import CondOTScheduler
from gensbi.flow_matching.solver import ODESolver
from gensbi.diffusion.path import EDMPath
from gensbi.diffusion.path.scheduler import EDMScheduler, VEScheduler
from gensbi.diffusion.solver import SDESolver
from gensbi.models import Flux1, Flux1Params, ConditionalCFMLoss, ConditionalWrapper, ConditionalDiffLoss
from einops import repeat
from gensbi.utils.model_wrapping import _expand_dims
import os
import yaml
from gensbi.recipes.conditional_pipeline import ConditionalFlowPipeline, ConditionalDiffusionPipeline
[docs]
def parse_flux1_params(config_path: str):
"""
Parse a Flux1 configuration file.
Parameters
----------
config_path : str
Path to the configuration file.
Returns
-------
config : dict
Parsed configuration dictionary.
"""
with open(config_path, "r") as f:
config = yaml.safe_load(f)
model_params = config.get("model", {})
params_dict = dict(
in_channels=model_params.get("in_channels", 1),
vec_in_dim=model_params.get("vec_in_dim", None),
context_in_dim=model_params.get("context_in_dim", 1),
mlp_ratio=model_params.get("mlp_ratio", 4),
num_heads=model_params.get("num_heads", 6),
depth=model_params.get("depth", 8),
depth_single_blocks=model_params.get("depth_single_blocks", 16),
axes_dim=model_params.get("axes_dim", [6]),
qkv_bias=model_params.get("qkv_bias", True),
theta=model_params.get("theta", -1),
param_dtype=getattr(jnp, model_params.get("param_dtype", "float32")),
)
return params_dict
[docs]
def parse_training_config(config_path: str):
"""
Parse a training configuration file.
Parameters
----------
config_path : str
Path to the configuration file.
Returns
-------
config : dict
Parsed configuration dictionary.
"""
with open(config_path, "r") as f:
config = yaml.safe_load(f)
# Training parameters
train_params = config.get("training", {})
multistep = train_params.get("multistep", 1)
experiment_id = train_params.get("experiment_id", 1)
early_stopping = train_params.get("early_stopping", True)
nsteps = train_params.get("nsteps", 30000) * multistep
val_every = train_params.get("val_every", 100) * multistep
# Optimizer parameters
opt_params = config.get("optimizer", {})
PATIENCE = opt_params.get("patience", 10)
COOLDOWN = opt_params.get("cooldown", 2)
FACTOR = opt_params.get("factor", 0.5)
ACCUMULATION_SIZE = opt_params.get("accumulation_size", 100) * multistep
RTOL = opt_params.get("rtol", 1e-4)
MAX_LR = opt_params.get("max_lr", 1e-3)
MIN_LR = opt_params.get("min_lr", 0.0)
MIN_SCALE = MIN_LR / MAX_LR if MAX_LR > 0 else 0.0
ema_decay = opt_params.get("ema_decay", 0.999)
training_config = {}
# overwrite the defaults with the config file values
training_config["num_steps"] = nsteps
training_config["ema_decay"] = ema_decay
training_config["patience"] = PATIENCE
training_config["cooldown"] = COOLDOWN
training_config["factor"] = FACTOR
training_config["accumulation_size"] = ACCUMULATION_SIZE
training_config["rtol"] = RTOL
training_config["max_lr"] = MAX_LR
training_config["min_lr"] = MIN_LR
training_config["min_scale"] = MIN_SCALE
training_config["val_every"] = val_every
training_config["early_stopping"] = early_stopping
training_config["experiment_id"] = experiment_id
training_config["multistep"] = multistep
return training_config
[docs]
class Flux1FlowPipeline(ConditionalFlowPipeline):
def __init__(
self,
train_dataset,
val_dataset,
dim_theta: int,
dim_x: int,
params=None,
training_config=None,
):
"""
Flow pipeline for training and using a Flux1 model for simulation-based inference.
Parameters
----------
train_dataset : grain dataset or iterator over batches
Training dataset.
val_dataset : grain dataset or iterator over batches
Validation dataset.
dim_theta : int
Dimension of the parameter space.
dim_x : int
Dimension of the observation space.
params : Flux1Params, optional
Parameters for the Flux1 model. If None, default parameters are used.
training_config : dict, optional
Configuration for training. If None, default configuration is used.
"""
super().__init__(
None, train_dataset, val_dataset, dim_theta, dim_x, params, training_config
)
if params is None:
self.params = self._get_default_params()
[docs]
self.model = self._make_model(self.params)
[docs]
self.ema_model = nnx.clone(self.model)
@classmethod
[docs]
def init_pipeline_from_config(
cls,
train_dataset,
val_dataset,
dim_theta: int,
dim_x: int,
config_path: str,
checkpoint_dir: str,
):
"""
Initialize the pipeline from a configuration file.
Parameters
----------
config_path : str
Path to the configuration file.
"""
with open(config_path, "r") as f:
config = yaml.safe_load(f)
# methodology
strategy = config.get("strategy", {})
method = strategy.get("method")
model_type = strategy.get("model")
assert method == "flow", f"Method {method} not supported in Flux1FlowPipeline."
assert (
model_type == "flux"
), f"Model type {model_type} not supported in Flux1FlowPipeline."
# Model parameters from config
dim_joint = dim_theta + dim_x
params_dict = parse_flux1_params(config_path)
if params_dict["theta"] == -1:
params_dict["theta"] = 4 * dim_joint
params = Flux1Params(
rngs=nnx.Rngs(0),
obs_dim=dim_theta,
cond_dim=dim_x,
**params_dict,
)
# Training parameters
training_config = cls._get_default_training_config()
training_config["checkpoint_dir"] = checkpoint_dir
training_config_ = parse_training_config(config_path)
for key, value in training_config_.items():
training_config[key] = value # update with config file values
pipeline = cls(
train_dataset,
val_dataset,
dim_theta,
dim_x,
params,
training_config,
)
return pipeline
[docs]
def _make_model(self, params):
"""
Create and return the Flux1 model to be trained.
"""
model = Flux1(params)
return model
[docs]
def _get_default_params(self):
"""
Return default parameters for the Flux1 model.
"""
params = Flux1Params(
in_channels=1,
vec_in_dim=None,
context_in_dim=1,
mlp_ratio=4,
num_heads=6,
depth=8,
depth_single_blocks=16,
axes_dim=[6],
qkv_bias=True,
obs_dim=self.dim_theta,
cond_dim=self.dim_x,
theta=10 * (self.dim_theta + self.dim_x),
rngs=nnx.Rngs(default=42),
param_dtype=jnp.float32,
)
return params
[docs]
class Flux1DiffusionPipeline(ConditionalDiffusionPipeline):
def __init__(
self,
train_dataset,
val_dataset,
dim_theta: int,
dim_x: int,
params=None,
training_config=None,
):
"""
Diffusion pipeline for training and using a Flux1 model for simulation-based inference.
Parameters
----------
train_dataset : grain dataset or iterator over batches
Training dataset.
val_dataset : grain dataset or iterator over batches
Validation dataset.
dim_theta : int
Dimension of the parameter space.
dim_x : int
Dimension of the observation space.
params : Flux1Params, optional
Parameters for the Flux1 model. If None, default parameters are used.
training_config : dict, optional
Configuration for training. If None, default configuration is used.
"""
super().__init__(
None, train_dataset, val_dataset, dim_theta, dim_x, params, training_config
)
if params is None:
self.params = self._get_default_params()
[docs]
self.model = self._make_model(self.params)
[docs]
self.ema_model = nnx.clone(self.model)
@classmethod
[docs]
def init_pipeline_from_config(
cls,
train_dataset,
val_dataset,
dim_theta: int,
dim_x: int,
config_path: str,
checkpoint_dir: str,
):
"""
Initialize the pipeline from a configuration file.
Parameters
----------
config_path : str
Path to the configuration file.
"""
with open(config_path, "r") as f:
config = yaml.safe_load(f)
# methodology
strategy = config.get("strategy", {})
method = strategy.get("method")
model_type = strategy.get("model")
assert (
method == "diffusion"
), f"Method {method} not supported in Flux1DiffusionPipeline."
assert (
model_type == "flux"
), f"Model type {model_type} not supported in Flux1DiffusionPipeline."
# Model parameters from config
dim_joint = dim_theta + dim_x
params_dict = parse_flux1_params(config_path)
if params_dict["theta"] == -1:
params_dict["theta"] = 4 * dim_joint
params = Flux1Params(
rngs=nnx.Rngs(0),
obs_dim=dim_theta,
cond_dim=dim_x,
**params_dict,
)
# Training parameters
training_config = cls._get_default_training_config()
training_config["checkpoint_dir"] = checkpoint_dir
training_config_ = parse_training_config(config_path)
for key, value in training_config_.items():
training_config[key] = value # update with config file values
pipeline = cls(
train_dataset,
val_dataset,
dim_theta,
dim_x,
params,
training_config,
)
return pipeline
[docs]
def _make_model(self, params):
"""
Create and return the Flux1 model to be trained.
"""
model = Flux1(params)
return model
[docs]
def _get_default_params(self):
"""
Return default parameters for the Flux1 model.
"""
params = Flux1Params(
in_channels=1,
vec_in_dim=None,
context_in_dim=1,
mlp_ratio=4,
num_heads=6,
depth=8,
depth_single_blocks=16,
axes_dim=[6],
qkv_bias=True,
obs_dim=self.dim_theta,
cond_dim=self.dim_x,
theta=10 * (self.dim_theta + self.dim_x),
rngs=nnx.Rngs(default=42),
param_dtype=jnp.float32,
)
return params
