Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

stackless fixes #37

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

stackless fixes #37

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
67c7672
stackless fixes
mattjj Dec 10, 2024
f0f99d8
remove dead code
mattjj Dec 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
163 changes: 72 additions & 91 deletions quax/_core.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,76 +141,51 @@ def _wrap_if_array(x: Union[ArrayLike, "Value"]) -> "Value":


class _QuaxTrace(core.Trace[_QuaxTracer]):
def pure(self, val: ArrayLike) -> _QuaxTracer:
if _is_value(val):
raise TypeError(
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If it's important to keep this behavior, can you give me a test that exercises it?

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure thing! The following should raise an error:

import quax
import jax

x = jax.numpy.arange(4.).reshape(2, 2)
key = jax.random.key(0)
y = quax.lora.LoraArray(x, rank=1, key=key)

def f(x):
    return jax.lax.add_p.bind(x, y)

quax.quaxify(f)(y)

Basically this is just a check that all of our Quax values are properly wrapped into tracers.

f"Encountered Quax value of type {type(val)}. These must be "
"transformed by passing them across a `quax.quaxify` boundary before "
"being used.\n"
"For example, the following is incorrect, as `SomeValue()` is not "
"explicitly passed across the API boundary:\n"
"```\n"
"def f(x):\n"
" return x + SomeValue()\n"
"\n"
"quax.quaxify(f)(AnotherValue())\n"
"```\n"
"This should instead be written as the following:\n"
"explicitly passed across the API boundary:\n"
"```\n"
"def f(x, y):\n"
" return x + y\n"
"\n"
"quax.quaxify(f)(AnotherValue(), SomeValue())\n"
"```\n"
"To better understand this, remember that the purpose of Quax is "
"take a JAX program (given as a function) that acts on arrays, and to "
"instead run it with array-ish types. But in the first example above, "
"the original program already has an array-ish type, even before the "
"`quaxify` is introduced."
)
if not eqx.is_array_like(val):
raise TypeError(f"{type(val)} is not a JAX type.")
return _QuaxTracer(self, _DenseArrayValue(val)) # pyright: ignore

def lift(self, tracer: core.Tracer) -> _QuaxTracer:
return _QuaxTracer(self, _DenseArrayValue(tracer))
def __init__(self, parent_trace, tag):
self.tag = tag
self.parent_trace = parent_trace

def sublift(self, tracer: _QuaxTracer) -> _QuaxTracer:
return tracer
def to_value(self, val):
if isinstance(val, _QuaxTracer) and val._trace.tag is self.tag:
return val.value
else:
return _DenseArrayValue(val)
Comment on lines +148 to +152
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For my own understanding of how new-JAX works, is it expected that we ever hit the else branch? Old-JAX had it such that the tracers of process_primitive(..., tracers, ...) was guaranteed to be tracers from the current trace, which would imply always hitting the if statement here.


def process_primitive(self, primitive, tracers, params):
values = [t.value for t in tracers]
# params = dict(params); params.pop('sharding', None)
values = [self.to_value(t) for t in tracers]
values = tuple(
x.array if isinstance(x, _DenseArrayValue) else x for x in values
)
try:
rule = _rules[primitive]
except KeyError:
out = _default_process(primitive, values, params)
with core.set_current_trace(self.parent_trace):
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Aha! It's super interesting to see that new-JAX now has each trace record the parent in a stack like this. I've done a couple of hobby projects reimplementing various kinds of JAXlike designs (sadly none of them released yet), and I've always ended up going for something similar.

(Albeit I've tended to go the opposite way and do full-data-dependence rather than full-dynamic-context dependence, but I'm guessing you're constrained there by the desire to do omnistaging.)

out = _default_process(primitive, values, params)
else:
try:
method, _ = rule.resolve_method(values)
except plum.NotFoundLookupError:
out = _default_process(primitive, values, params)
with core.set_current_trace(self.parent_trace):
out = _default_process(primitive, values, params)
else:
out = method(*values, **params)
with core.set_current_trace(self.parent_trace):
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: I think you could hoist each core.set_current_trace out of all these try/except blocks and have just one wrapping the whole thing.

out = method(*values, **params)
if primitive.multiple_results:
return [_QuaxTracer(self, _wrap_if_array(x)) for x in out] # pyright: ignore
else:
return _QuaxTracer(self, _wrap_if_array(out)) # pyright: ignore

def process_custom_jvp_call(self, primitive, fun, jvp, tracers, *, symbolic_zeros):
in_values = [t.value for t in tracers]
in_values = [self.to_value(t) for t in tracers]
# Each `t.value` will be some `Value`, and thus a PyTree. Here we flatten the
# `Value`-ness away.
in_leaves, in_treedef = jtu.tree_flatten(in_values)
fun, out_treedef1 = _custom_jvp_fun_wrap(fun, self.main, in_treedef) # pyright: ignore
jvp, out_treedef2 = _custom_jvp_jvp_wrap(jvp, self.main, in_treedef) # pyright: ignore
with jax.ensure_compile_time_eval():
out_leaves = primitive.bind(
fun, jvp, *in_leaves, symbolic_zeros=symbolic_zeros
)
fun, out_treedef1 = _custom_jvp_fun_wrap(fun, self.tag, in_treedef) # pyright: ignore
jvp, out_treedef2 = _custom_jvp_jvp_wrap(jvp, self.tag, in_treedef) # pyright: ignore
out_leaves = primitive.bind_with_trace(
self.parent_trace, (fun, jvp, *in_leaves), dict(symbolic_zeros=symbolic_zeros),
)
_, out_treedef = lu.merge_linear_aux(out_treedef1, out_treedef2)
out_values = jtu.tree_unflatten(out_treedef, out_leaves)
return [_QuaxTracer(self, x) for x in out_values]
Expand All @@ -219,51 +194,59 @@ def process_custom_jvp_call(self, primitive, fun, jvp, tracers, *, symbolic_zero


@lu.transformation_with_aux # pyright: ignore
def _custom_jvp_fun_wrap(main, in_treedef, *in_leaves):
trace = main.with_cur_sublevel()
def _custom_jvp_fun_wrap(tag, in_treedef, *in_leaves):
in_values = jtu.tree_unflatten(in_treedef, in_leaves)
in_tracers = [x if type(x) is SZ else _QuaxTracer(trace, x) for x in in_values]
out_tracers = yield in_tracers, {}
# The symbolic zero branch here will actually create a `quax.zero.Zero`!
out_tracers = [
jnp.zeros(t.aval.shape, t.aval.dtype) if type(t) is SZ else t # pyright: ignore
for t in out_tracers
]
out_values = [trace.full_raise(t).value for t in out_tracers]
with core.take_current_trace() as parent_trace:
trace = _QuaxTrace(parent_trace, tag)
in_tracers = [x if type(x) is SZ else _QuaxTracer(trace, x) for x in in_values]
with core.set_current_trace(trace):
out_tracers = yield in_tracers, {}
# The symbolic zero branch here will actually create a `quax.zero.Zero`!
out_tracers = [
jnp.zeros(t.aval.shape, t.aval.dtype) if type(t) is SZ else t # pyright: ignore
for t in out_tracers
]
out_values = [trace.to_value(t) for t in out_tracers]
del out_tracers
del trace, in_tracers
out_leaves, out_treedef = jtu.tree_flatten(out_values)
yield out_leaves, out_treedef


@lu.transformation_with_aux # pyright: ignore
def _custom_jvp_jvp_wrap(main, in_treedef, *in_primals_and_tangents):
trace = main.with_cur_sublevel()
def _custom_jvp_jvp_wrap(tag, in_treedef, *in_primals_and_tangents):
in_primals = in_primals_and_tangents[: len(in_primals_and_tangents) // 2]
in_tangents = in_primals_and_tangents[len(in_primals_and_tangents) // 2 :]
in_primal_values = jtu.tree_unflatten(in_treedef, in_primals)
in_tangent_values = jtu.tree_unflatten(in_treedef, in_tangents)
# Calling `_QuaxTracer` directly here, not using `trace.{pure,lift}` as each `x` is
# a `Value`, not an array (=> pure) or tracer (=> lift).
in_tracers = [
_QuaxTracer(trace, x) for x in it.chain(in_primal_values, in_tangent_values)
]
out_tracers = yield in_tracers, {}
# The symbolic zero branch here will actually create a `quax.zero.Zero`!
out_tracers = [
jnp.zeros(t.aval.shape, t.aval.dtype) if type(t) is SZ else t # pyright: ignore
for t in out_tracers
]
out_values = [trace.full_raise(t).value for t in out_tracers]
out_primal_values = out_values[: len(out_values) // 2]
out_tangent_values = out_values[len(out_values) // 2 :]
out_primal_values2 = []
out_tangent_values2 = []
assert len(out_primal_values) == len(out_tangent_values)
for primal, tangent in zip(out_primal_values, out_tangent_values):
if primal.__class__ != tangent.__class__:
primal = primal.materialise()
tangent = tangent.materialise()
out_primal_values2.append(primal)
out_tangent_values2.append(tangent)
with core.take_current_trace() as parent_trace:
trace = _QuaxTrace(parent_trace, tag)
in_tracers = [
_QuaxTracer(trace, x) for x in it.chain(in_primal_values, in_tangent_values)
]
with core.set_current_trace(trace):
out_tracers = yield in_tracers, {}
# The symbolic zero branch here will actually create a `quax.zero.Zero`!
out_tracers = [
jnp.zeros(t.aval.shape, t.aval.dtype) if type(t) is SZ else t # pyright: ignore
for t in out_tracers
]
out_values = [trace.to_value(t) for t in out_tracers]
out_primal_values = out_values[: len(out_values) // 2]
out_tangent_values = out_values[len(out_values) // 2 :]
out_primal_values2 = []
out_tangent_values2 = []
assert len(out_primal_values) == len(out_tangent_values)
for primal, tangent in zip(out_primal_values, out_tangent_values):
if primal.__class__ != tangent.__class__:
primal = primal.materialise()
tangent = tangent.materialise()
out_primal_values2.append(primal)
out_tangent_values2.append(tangent)
del out_tracers
del trace, in_tracers
out_primals, out_primal_treedef = jtu.tree_flatten(out_primal_values2)
out_tangents, out_tangent_treedef = jtu.tree_flatten(out_tangent_values2)
if out_primal_treedef != out_tangent_treedef:
Expand Down Expand Up @@ -307,21 +290,20 @@ def __wrapped__(self) -> CT:
return self.fn

def __call__(self, *args, **kwargs):
with core.new_main(_QuaxTrace, dynamic=self.dynamic) as main:
trace = _QuaxTrace(main, core.cur_sublevel())
# Note that we do *not* wrap arraylikes here. We let that happen in
# `_QuaxTrace.{pure,lift}` as necessary. This means that we can do e.g.
# quaxify(jnp.moveaxis)(array, source=0, destination=-1).
dynamic, static = eqx.partition(
(self.fn, args, kwargs), self.filter_spec, is_leaf=_is_value
)
dynamic, static = eqx.partition(
(self.fn, args, kwargs), self.filter_spec, is_leaf=_is_value
)
tag = core.TraceTag()
with core.take_current_trace() as parent_trace:
trace = _QuaxTrace(parent_trace, tag)
dynamic = jtu.tree_map(
ft.partial(_wrap_tracer, trace),
dynamic,
is_leaf=_is_value,
)
fn, args, kwargs = eqx.combine(dynamic, static)
out = fn(*args, **kwargs)
with core.set_current_trace(trace):
out = fn(*args, **kwargs)
out = jtu.tree_map(ft.partial(_unwrap_tracer, trace), out)
return out

Expand Down Expand Up @@ -543,8 +525,7 @@ def _(*args: Union[ArrayLike, ArrayValue], jaxpr, inline, **kwargs):
else:
leaves, treedef = jtu.tree_flatten(args) # remove all Values
flat_fun = lambda x: fun(*jtu.tree_unflatten(treedef, x))
with jax.ensure_compile_time_eval(): # replace the dynamic QuaxTrace
return jax.jit(flat_fun)(leaves) # now we can call without Quax.
return jax.jit(flat_fun)(leaves) # now we can call without Quax.


@register(jax.lax.while_p)
Expand Down
2 changes: 1 addition & 1 deletion quax/examples/sparse/_core.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ def _op_sparse_to_dense(x, y, op):


@quax.register(lax.broadcast_in_dim_p)
def _(value: BCOO, *, broadcast_dimensions, shape) -> BCOO:
def _(value: BCOO, *, broadcast_dimensions, shape, sharding=None) -> BCOO:
n_extra_batch_dims = len(shape) - value.ndim
if broadcast_dimensions != tuple(range(n_extra_batch_dims, len(shape))):
raise NotImplementedError(
Expand Down
22 changes: 9 additions & 13 deletions quax/examples/zero/_core.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,23 +42,19 @@ def materialise(self):

@quax.register(lax.broadcast_in_dim_p)
def _(
value: ArrayLike, *, broadcast_dimensions, shape
value: ArrayLike, *, broadcast_dimensions, shape, sharding=None
) -> Union[ArrayLike, quax.ArrayValue]:
aval = jax.core.get_aval(value)
if isinstance(aval, jax.core.ConcreteArray) and aval.shape == () and aval.val == 0:
return Zero(shape, np.result_type(value))
else:
# Avoid an infinite loop, by pushing a new interpreter to the dynamic
# interpreter stack.
with jax.ensure_compile_time_eval():
out = lax.broadcast_in_dim_p.bind(
value, broadcast_dimensions=broadcast_dimensions, shape=shape
)
return out # pyright: ignore
# Avoid an infinite loop using ensure_compile_time_eval.
with jax.ensure_compile_time_eval():
out = lax.broadcast_in_dim_p.bind(
value, broadcast_dimensions=broadcast_dimensions, shape=shape,
sharding=sharding
)
return out # pyright: ignore


@quax.register(lax.broadcast_in_dim_p)
def _(value: Zero, *, broadcast_dimensions, shape) -> Zero:
def _(value: Zero, *, broadcast_dimensions, shape, sharding=None) -> Zero:
del broadcast_dimensions
return Zero(shape, value.dtype)

Expand Down
Loading