Replace Pyro with NumPyro for fully Bayesian NUTS inference (96% reduction in fit time) (#5087)#5087
Open
sdaulton wants to merge 1 commit intofacebook:mainfrom
Open
Replace Pyro with NumPyro for fully Bayesian NUTS inference (96% reduction in fit time) (#5087)#5087sdaulton wants to merge 1 commit intofacebook:mainfrom
sdaulton wants to merge 1 commit intofacebook:mainfrom
Conversation
meta-cla
bot
added
the
CLA Signed
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 21, 2026
…ction in fit time) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
meta-codesync
bot
changed the title
sdaulton
force-pushed
the
export-D97159025
branch
from
a9b8319 to
719844c
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 22, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 22, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
from
719844c to
7e16e2a
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 23, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 23, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
from
7e16e2a to
df0ff94
Compare
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
2 times, most recently
from
b76911c to
675444d
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 Pull Request resolved: facebook#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
from
675444d to
7ad52f3
Compare
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (meta-pytorch#3247) Summary: Pull Request resolved: meta-pytorch#3247 X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
from
7ad52f3 to
78f1619
Compare
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 24, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
…ction in fit time) (facebook#5087) Summary: X-link: meta-pytorch/botorch#3247 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
force-pushed
the
export-D97159025
branch
from
78f1619 to
e1bde48
Compare
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 25, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
sdaulton
added a commit
to sdaulton/botorch
that referenced
this pull request
Mar 25, 2026
…ction in fit time) (meta-pytorch#3247) Summary: X-link: facebook/Ax#5087 This diff replaces all Pyro usage in BoTorch's fully Bayesian models with NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is significantly faster than Pyro's PyTorch-based one, delivering 25x speedups on CPU with equivalent model quality. Key changes: - Replace `pyro.sample`/`pyro.distributions` with `numpyro.sample`/`numpyro.distributions` - Replace `pyro.deterministic` with `numpyro.deterministic` - Rewrite kernel functions (`matern52_kernel`, `linear_kernel`, `compute_dists`) in JAX - Replace Pyro's `NUTS`/`MCMC` with NumPyro equivalents (`dense_mass` instead of `full_mass`, `num_warmup` instead of `warmup_steps`, explicit `jax.random.PRNGKey`) - Remove `register_exception_handler` (Pyro-specific); NumPyro handles divergences via built-in NaN detection - Add `botorch/utils/jax_utils.py` with `torch_to_jax`/`jax_to_torch` conversion helpers - Convert training data to JAX arrays in `PyroModel.set_inputs()`; all `sample()` methods now operate in JAX-land - Post-process MCMC samples by converting JAX arrays back to torch tensors at the boundary - Pin numpyro to 0.18.0 in PACKAGE for vectorized chain support - Update BUCK to replace `pyro-ppl` dep with `jax` + `numpyro` Differential Revision: D97159025
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
X-link: meta-pytorch/botorch#3247
This diff replaces all Pyro usage in BoTorch's fully Bayesian models with
NumPyro (backed by JAX). NumPyro's JAX-based NUTS implementation is
significantly faster than Pyro's PyTorch-based one, delivering 25x
speedups on CPU with equivalent model quality.
Key changes:
pyro.sample/pyro.distributionswithnumpyro.sample/numpyro.distributionspyro.deterministicwithnumpyro.deterministicmatern52_kernel,linear_kernel,compute_dists) in JAXNUTS/MCMCwith NumPyro equivalents (dense_massinstead offull_mass,num_warmupinstead ofwarmup_steps, explicitjax.random.PRNGKey)register_exception_handler(Pyro-specific); NumPyro handles divergences via built-in NaN detectionbotorch/utils/jax_utils.pywithtorch_to_jax/jax_to_torchconversion helpersPyroModel.set_inputs(); allsample()methods now operate in JAX-landpyro-ppldep withjax+numpyroDifferential Revision: D97159025