| name | add-pallas-kernel |
| description | Add, modify, or autotune a TPU/GPU Pallas kernel. |
Skill: Add or Update a Pallas Kernel
Use this skill to build or change Pallas kernels with explicit standards for
reference numerics, gradient safety, backend/fallback API design, performance
measurement, and block-size autotuning.
How to apply this skill
- For long-running kernel research, load
.agents/skills/run-research/SKILL.md
first.
- Apply the core workflow in this file.
- Load only the detail files needed for the task:
- Kernel sources: read when choosing an in-repo or
external kernel to imitate.
- Performance workflow: read before
benchmarking, profiling, roofline analysis, or autotuning.
- API patterns: read before adding or changing a
public kernel wrapper, fallback order, or block-size config.
- TPU tips: read for TPU Pallas/Mosaic kernels,
TPU-specific lowering failures, scoped VMEM, or TPU compiler dumps.
- GPU tips: read for GPU Pallas/Mosaic work.
- Deep references live under
docs/reference/; read them only when the
routed detail files point there.
- For atomic kernel changes, use only the task skills needed for the work.
Kernel Deliverables
For a kernel K, produce:
- A readable vanilla JAX reference with the target public API.
- A correctness harness validating value parity vs reference, gradient parity on
small shapes, and CPU + accelerator numerics where applicable.
- A Pallas kernel implementation plus wrapper with the same API.
- A roofline estimate for the relevant hardware type(s).
- A performance harness with steady-state timing on representative shape/dtype
grids and progress against the roofline.
- Autotuned block/tile sizes for requested hardware/shape regimes.
- A checked-in tuned table module for runtime selection, with explicit fallback
behavior.
- An autotune-on-miss fallback path that sweeps a bounded candidate set and
caches winning configs.
The general flow is: make it right, make it fast, make it usable, make it easy
to use.
Correctness Workflow
1. Start from a reference
Use an existing in-repo implementation, pseudocode, a PyTorch reference, or a
JAX baseline. The baseline must be obvious and stable, not clever. If the naive
baseline would materialize huge intermediates, use a streaming/blockwise
baseline with identical math.
2. Write a value and gradient harness
Minimum checks:
- Value parity over a shape/dtype grid.
- Gradient parity on small shapes.
- Backend numerics on CPU and accelerator backends as applicable.
- Pointwise deviation metrics such as max/mean absolute diff, not only
allclose.
Use explicit shape/dtype annotations for public APIs and references, such as
jaxtyping, where available.
3. Promote long-lived checks to pytest
For in-tree kernels, add or extend tests under lib/levanter/tests/kernels/.
Compare the default implementation against the reference on small CPU shapes and
accelerator-aligned shapes for fast paths. Read TESTING.md and the nearest
module AGENTS.md before writing or changing tests.
Pallas Kernel Workflow
Once the reference is correct, design the Pallas implementation. Use the
reference as both a correctness oracle and a performance baseline.
Use existing kernels for structure and API inspiration. Read
Kernel sources unless the user already named the
specific kernel to follow. Unless there is a stronger local pattern, start by
reimplementing the reference in Pallas.
Wrap accelerator kernel boundaries in an explicit jax.shard_map by default.
This applies to pl.pallas_call, Mosaic GPU kernels, and custom FFI calls.
Reshard inputs to the intended local PartitionSpec before the shard_map,
keep the sequence or other nonlocal dimensions unsharded unless the kernel is
explicitly written for them, and add a regression check that the lowered JAXPR
or HLO contains the expected shard_map. Do not rely on XLA to infer a good
sharding for an opaque kernel call boundary. Exceptions are limited to wrappers
whose inputs are explicitly documented and tested as fully local or replicated.
Check correctness against the harness and reference implementation before
tuning. Once the kernel is correct, run a performance harness on representative
shapes/dtypes and compare against the roofline. If performance is not near the
expected roofline, read Performance workflow and
investigate compiler dumps, pressure signals, and tile choices before broad
rewrites.
API Conventions
Read API patterns before adding or changing the public
wrapper, backend selection, block-size config, or input normalization contract.
Keep the reference/XLA path usable even when accelerator-specific constraints
are not met. Keep backend-specific validation in backend-specific modules.
Cost Estimate Requirement
Add cost_estimate= to each pl.pallas_call:
- Use
pl.estimate_cost on a body-equivalent JAX function, not a kernel body
with pl.program_id.
- Include IO bytes from call inputs/outputs.
from levanter.kernels.pallas.cost_estimate_utils import with_io_bytes_accessed
def _cost_estimate(
q: jax.Array,
k: jax.Array,
v: jax.Array,
*,
kernel_inputs_specs,
kernel_outputs_specs,
) -> pl.CostEstimate | None:
body_cost = pl.estimate_cost(reference_impl, q, k, v)
return with_io_bytes_accessed(
body_cost,
kernel_inputs_specs=kernel_inputs_specs,
kernel_outputs_specs=kernel_outputs_specs,
)
Definition of Done
- Values match reference within tolerance on the tested grid.
- Gradients match reference on small shapes.
- CPU/reference and accelerator fast paths are covered by tests where
applicable.
- Public API, fallback semantics, block-size config, and tuned table behavior
match API patterns.
- Every Pallas, Mosaic, or FFI kernel call is inside an explicit
shard_map, or
its wrapper documents and tests why the inputs are fully local or replicated.
Tests or profile evidence show it did not lower through unintended
all-gathers.
- Each
pl.pallas_call has a reviewed cost_estimate=.
- Benchmark/tuning artifacts include the required schema from
Performance workflow.
- Roofline performance is within expected bounds, or limitations are explicitly
documented.
- Performance improves on at least one realistic target shape, or limitations
are explicitly documented.
- Tuned table is checked in for requested hardware/shape regimes.
- Research artifacts, issue summaries, and snapshot links follow the
run-research workflow when the task is long-running.
PR Checklist
- Reference implementation and public wrapper are in place.
- Correctness tests cover values, gradients, and relevant backend paths.
- Pallas implementation has explicit backend/shape validation.
- Pallas, Mosaic, and FFI calls use an explicit
shard_map boundary when
operating on sharded inputs.
- Fallback behavior is tested for explicit and ordered implementation choices.
- Cost estimates are attached to Pallas calls.
- Benchmark or tuning script emits machine-readable rows.
- Tuned table and autotune-on-miss behavior are checked in when tuning is part
of the task.
- Long-running work has the required logbook/artifact/snapshot updates.