def __call__(self, *args: Any, **kwargs: Any) -> ReturnValue: # pytype: disable=invalid-annotation
if not self.fwd or not self.bwd:
msg = "No VJP defined for custom_vjp function {} using defvjp."
raise AttributeError(msg.format(self.__name__))
args = _resolve_kwargs(self.fun, args, kwargs)
if config.jax_enable_custom_vjp_by_custom_transpose:
if self.nondiff_argnums:
raise NotImplementedError(
'nondiff_argnums not implemented for new custom_vjp')
return custom_vjp_by_custom_transpose(self.fun, self.fwd, self.bwd)(*args)
else:
if self.nondiff_argnums:
for i in self.nondiff_argnums: _check_for_tracers(args[i])
nondiff_argnums = set(self.nondiff_argnums)
dyn_argnums = [i for i in range(len(args)) if i not in nondiff_argnums]
f_, dyn_args = argnums_partial(lu.wrap_init(self.fun), dyn_argnums,
args, require_static_args_hashable=False)
static_args = [args[i] for i in self.nondiff_argnums]
fwd, _ = argnums_partial(lu.wrap_init(self.fwd), dyn_argnums, args,
require_static_args_hashable=False)
bwd = _add_args(lu.wrap_init(self.bwd), static_args)
else:
f_, dyn_args = lu.wrap_init(self.fun), args
fwd, bwd = lu.wrap_init(self.fwd), lu.wrap_init(self.bwd)
args_flat, in_tree = tree_flatten(dyn_args)
in_avals = [core.raise_to_shaped(core.get_aval(x)) for x in args_flat]
flat_fun, out_tree = flatten_fun_nokwargs(f_, in_tree)
flat_fwd, out_trees = _flatten_fwd(fwd, in_tree)
flat_bwd = _flatten_bwd(bwd, in_tree, in_avals, out_trees)
out_flat = custom_vjp_call_p.bind(flat_fun, flat_fwd, flat_bwd,
*args_flat, out_trees=out_trees)
fst, aux = lu.merge_linear_aux(out_tree, out_trees)
out_tree = aux if fst else aux[0]
return tree_unflatten(out_tree, out_flat)
def remat_transpose(params, call_jaxpr, primals_in, cotangents_in,
cotangent_in_avals, reduce_axes):
# backward_pass can only transpose linear computations, but the call_jaxpr embedded in
# remat contains primal (non-linear) equations too. Hence, we have to eliminate those
# (in this case via partial_eval) before we call into backward_pass again.
typed_call_jaxpr = core.ClosedJaxpr(call_jaxpr, [])
unknowns = map(is_undefined_primal, primals_in)
primal_jaxpr, tangent_jaxpr, out_unknowns = \
pe.partial_eval_jaxpr(typed_call_jaxpr, unknowns=unknowns, instantiate=True) # type: ignore
def do_transpose(primals_in, cotangents_in):
# NOTE: This is passing in undefined primals in place of tangent arguments, but it
# should all work out, because we're only computing the primal part here.
residuals = core.jaxpr_as_fun(primal_jaxpr)(
*primals_in)[len(cotangents_in):]
# Now that we have a purely linear jaxpr, we can transpose it
cotangents_out = backward_pass(tangent_jaxpr.jaxpr, reduce_axes, (),
primals_in + residuals, cotangents_in)
# backward_pass will return cotangents computed for all invars, but some of them
# are residuals appended by partial eval, so we need to skip those before we return.
return cotangents_out[:len(primals_in)]
flat_args, in_tree_def = tree_flatten((primals_in, cotangents_in))
flat_do_transpose, out_tree = flatten_fun_nokwargs(
lu.wrap_init(do_transpose), in_tree_def)
flat_cotangents_out = pe.remat_call_p.bind(flat_do_transpose, *flat_args,
**params)
return tree_unflatten(out_tree(), flat_cotangents_out)
def jvp_of_rule_rule(axis_size, in_batched, primals, tangents):
in_batched_ps, in_batched_ts = in_batched
mutually_batched = tree_map(operator.and_, in_batched_ps, in_batched_ts)
extra_batched_ps = tree_map(lambda pb, tb: 0 if pb and not tb else None,
in_batched_ps, in_batched_ts)
extra_batched_ts = tree_map(lambda pb, tb: 0 if tb and not pb else None,
in_batched_ps, in_batched_ts)
out_mutually_batched = lu.Store()
flat_ps_ts, tree_ps_ts = tree_flatten((primals, tangents))
flat_extra_batched_ps_ts, tree_ps_ts2 = tree_flatten(
(extra_batched_ps, extra_batched_ts),
is_leaf=lambda x: x is None)
# TODO(frostig): assert these also equal:
# treedef_tuple((in_tree, in_tree))
# once https://github.com/google/jax/issues/9066 is fixed
assert tree_ps_ts == tree_ps_ts2
del tree_ps_ts2
def to_jvp(*primals):
out, out_batched = call_rule(rule, axis_size, mutually_batched, primals)
check_vmap_rule_trees(
rule, out_tree, tree_structure(out), tree_structure(out_batched))
out_mutually_batched.store(out_batched)
return out
def to_vmap_over_extra_batched_dims(primals, tangents):
return jax.jvp(to_jvp, primals, tangents)
to_vmap_over_extra_batched_dims_flat, out_tree2 = flatten_fun_nokwargs(
lu.wrap_init(to_vmap_over_extra_batched_dims),
tree_ps_ts)
flat_out_ps_ts, flat_out_axes = vmap_unrestricted(
to_vmap_over_extra_batched_dims_flat, *flat_ps_ts,
in_axes=flat_extra_batched_ps_ts,
axis_name=core.no_axis_name, axis_size=axis_size)
n, ragged = divmod(len(flat_out_ps_ts), 2)
assert not ragged
flat_out_ps, flat_out_ts = flat_out_ps_ts[:n], flat_out_ps_ts[n:]
flat_out_axes_p, flat_out_axes_t = flat_out_axes[:n], flat_out_axes[n:]
flat_out_ps = map(maybe_bdim_at_front, flat_out_ps, flat_out_axes_p)
flat_out_extra_batched_ps = [d is not not_mapped for d in flat_out_axes_p]
flat_out_ts = map(maybe_bdim_at_front, flat_out_ts, flat_out_axes_t)
flat_out_extra_batched_ts = [d is not not_mapped for d in flat_out_axes_t]
out_ps, out_ts = tree_unflatten(
out_tree2(), [*flat_out_ps, *flat_out_ts])
out_extra_batched_ps, out_extra_batched_ts = tree_unflatten(
out_tree2(), [*flat_out_extra_batched_ps, *flat_out_extra_batched_ts])
out_batched_ps = tree_map(
operator.or_, out_mutually_batched.val, out_extra_batched_ps)
out_batched_ts = tree_map(
operator.or_, out_mutually_batched.val, out_extra_batched_ts)
return (out_ps, out_ts), (out_batched_ps, out_batched_ts)
def __call__(self, *args: Any, **kwargs: Any) -> ReturnValue: # pytype: disable=invalid-annotation
if not self.jvp:
msg = "No JVP defined for custom_jvp function {} using defjvp."
raise AttributeError(msg.format(self.__name__))
args = _resolve_kwargs(self.fun, args, kwargs)
if self.nondiff_argnums:
nondiff_argnums = set(self.nondiff_argnums)
args = tuple(
_stop_gradient(x) if i in nondiff_argnums else x
for i, x in enumerate(args))
diff_argnums = [
i for i in range(len(args)) if i not in nondiff_argnums
]
f_, dyn_args = argnums_partial(lu.wrap_init(self.fun),
diff_argnums,
args,
require_static_args_hashable=False)
static_args = [args[i] for i in self.nondiff_argnums]
jvp = _add_args(lu.wrap_init(self.jvp), static_args)
else:
f_, dyn_args = lu.wrap_init(self.fun), args
jvp = lu.wrap_init(self.jvp)
args_flat, in_tree = tree_flatten(dyn_args)
flat_fun, out_tree1 = flatten_fun_nokwargs(f_, in_tree)
flat_jvp, out_tree2 = _flatten_jvp(jvp, in_tree)
out_flat = custom_jvp_call_p.bind(flat_fun, flat_jvp, *args_flat)
_, out_tree = lu.merge_linear_aux(out_tree1, out_tree2)
return tree_unflatten(out_tree, out_flat)
def fwd(*args, **kwargs):
ans, rule = fun(*args, **kwargs)
ans_flat, out_tree = tree_flatten((ans, ))
rule, in_tree = flatten_fun_nokwargs(lu.wrap_init(rule), out_tree)
ans_avals = [core.get_aval(x).at_least_vspace() for x in ans_flat]
jaxpr, _, consts = pe.trace_to_jaxpr_dynamic(rule, ans_avals)
return ans, Residuals(jaxpr, in_tree(), out_tree, consts)
def wrapped(spenv: SparsifyEnv, *spvalues: SparsifyValue, **params: Any) -> Tuple[Sequence[SparsifyValue], bool]:
spvalues_flat, in_tree = tree_flatten(spvalues, is_leaf=_is_spvalue)
in_avals_flat = spvalues_to_avals(spenv, spvalues_flat)
wrapped_fun, out_tree = flatten_fun_nokwargs(lu.wrap_init(f, params), in_tree)
jaxpr, out_avals_flat, consts = pe.trace_to_jaxpr_dynamic(wrapped_fun, in_avals_flat)
result = eval_sparse(jaxpr, consts, spvalues_flat, spenv)
if len(out_avals_flat) != len(result):
raise Exception("Internal: eval_sparse does not return expected number of arguments. "
"Got {result} for avals {out_avals_flat}")
return result, out_tree()
def remat_transpose(params, call_jaxpr, primals_in, cotangents_in,
cotangent_in_avals, reduce_axes):
call_jaxpr = _close_jaxpr(call_jaxpr)
unknowns = map(is_undefined_primal, primals_in)
primal_jaxpr, tangent_jaxpr, _ = \
pe.partial_eval_jaxpr(call_jaxpr, unknowns=unknowns, instantiate=True) # type: ignore
args, in_tree_def = tree_flatten((primals_in, cotangents_in))
transpose = lu.hashable_partial(lu.wrap_init(_remat_transpose), primal_jaxpr,
tangent_jaxpr, reduce_axes)
flat_transpose, out_tree = flatten_fun_nokwargs(transpose, in_tree_def)
flat_cotangents_out = pe.remat_call_p.bind(flat_transpose, *args, **params)
return tree_unflatten(out_tree(), flat_cotangents_out)
def __call__(self, *args, **kwargs):
if self.ivjp is None:
msg = "No IVJP defined for custom_vjp function {}. Did you forget to use defivjp?"
raise AttributeError(msg.format(self.__name__))
args = custom_derivatives._resolve_kwargs(self.fun, args, kwargs)
# TODO: Support nondiff_argnums
fun, ivjp = lu.wrap_init(self.fun), lu.wrap_init(self.ivjp)
args_flat, in_tree = tree_flatten(args)
flat_fun, out_tree = flatten_fun_nokwargs(fun, in_tree)
flat_ivjp = _flatten_ivjp(ivjp, in_tree, out_tree)
out_flat = _custom_ivjp(flat_fun, flat_ivjp, args_flat)
return tree_unflatten(out_tree(), out_flat)
def call_transpose(primitive, params, call_jaxpr, args, ct, _, reduce_axes):
all_args, in_tree_def = tree_flatten(((), args, ct)) # empty consts
fun = lu.hashable_partial(lu.wrap_init(backward_pass), call_jaxpr,
reduce_axes)
fun, out_tree = flatten_fun_nokwargs(fun, in_tree_def)
new_params = dict(params, name=wrap_name(params['name'], 'transpose'))
update_params = call_transpose_param_updaters.get(primitive)
if update_params:
new_params = update_params(new_params, map(is_undefined_primal, args),
[type(x) is not Zero for x in ct])
out_flat = primitive.bind(fun, *all_args, **new_params)
return tree_unflatten(out_tree(), out_flat)
def fwd(*args):
flat_args, in_tree = tree_flatten(args)
in_pvals = tuple(pe.PartialVal.unknown(raise_to_shaped(get_aval(arg))) for arg in flat_args)
fun_flat, out_tree = flatten_fun_nokwargs(lu.wrap_init(fun), in_tree)
jaxpr, out_pvals, consts = pe.trace_to_jaxpr(fun_flat, in_pvals)
# TODO: Don't warn if consts contain JVP tracers?
if consts:
warnings.warn("Values that an @invertible function closes over will not have their " +
"gradients computed correctly (their uses inside this function will be ignored)!")
# TODO: This requires the body to be jittable, but this shouldn't be necessary.
# Is there a way to trace a jaxpr while running it?
flat_outs = core.eval_jaxpr(jaxpr, consts, *flat_args)
return tree_unflatten(out_tree(), flat_outs), (flat_args, flat_outs, consts, DontFlatten((jaxpr, in_tree)))
def __call__(self, *args, **kwargs):
assert not kwargs
args_flat, in_tree = tree_flatten(args)
flat_fun, out_tree = flatten_fun_nokwargs(lu.wrap_init(self.fun), in_tree)
in_avals = [core.raise_to_shaped(core.get_aval(x)) for x in args_flat]
debug = pe.debug_info(self.fun, in_tree, False, "custom_vmap")
jaxpr, _, consts = pe.trace_to_jaxpr_dynamic(flat_fun, in_avals, debug)
assert not len(consts)
closed_call = core.ClosedJaxpr(pe.convert_constvars_jaxpr(jaxpr), ())
out_flat = custom_vmap_p.bind(*consts, *args_flat,
call=closed_call,
rule=self.vmap_rule,
in_tree=in_tree)
return tree_unflatten(out_tree(), out_flat)
def call_transpose(primitive, params, call_jaxpr, args, ct, _, reduce_axes):
all_args, in_tree_def = tree_flatten(((), args, ct)) # empty consts
fun = lu.hashable_partial(lu.wrap_init(backward_pass), call_jaxpr,
reduce_axes, False)
fun, out_tree = flatten_fun_nokwargs(fun, in_tree_def)
if not config.jax_experimental_name_stack:
params = dict(params, name=wrap_name(params['name'], 'transpose'))
update_params = call_transpose_param_updaters.get(primitive)
if update_params:
params = update_params(params, map(is_undefined_primal, args),
[type(x) is not Zero for x in ct])
if config.jax_dynamic_shapes:
in_type = [(core.raise_to_shaped(core.get_aval(x)), True) for x in all_args]
fun = lu.annotate(fun, tuple(in_type))
out_flat = primitive.bind(fun, *all_args, **params)
return tree_unflatten(out_tree(), out_flat)
def map_transpose(primitive, params, call_jaxpr, args, ct, _, reduce_axes):
all_args, in_tree_def = tree_flatten(((), args, ct)) # empty consts
fun = lu.hashable_partial(lu.wrap_init(backward_pass), call_jaxpr,
reduce_axes, False)
fun, nz_arg_cts = nonzero_outputs(fun)
fun, out_tree = flatten_fun_nokwargs(fun, in_tree_def)
# Preserve axis for primal arguments, skip tangents (represented as undefined primals).
in_axes, out_axes = params['in_axes'], params['out_axes']
new_in_axes = (*[
axis for axis, x in zip(in_axes, args) if not is_undefined_primal(x)
], *[axis for axis, x in zip(out_axes, ct) if type(x) is not Zero])
# The interim strategy we use below (until avals-with-names) only works
# when all outputs are mapped.
assert all(out_axis is not None for out_axis in out_axes), out_axes
# NOTE: This assumes that the output cotangents being zero is a deterministic
# function of which input cotangents were zero.
@as_hashable_function(closure=(in_axes, tuple(type(c) is Zero
for c in ct)))
def out_axes_thunk():
return tuple(axis or 0 for axis, nz in zip(in_axes, nz_arg_cts())
if nz)
new_params = dict(params,
name=wrap_name(params['name'], 'transpose'),
in_axes=new_in_axes,
out_axes_thunk=out_axes_thunk)
del new_params['out_axes']
update_params = call_transpose_param_updaters.get(primitive)
if update_params:
new_params = update_params(new_params, map(is_undefined_primal, args),
[type(x) is not Zero for x in ct])
out_flat = primitive.bind(fun, *all_args, **new_params)
arg_cts = tree_unflatten(out_tree(), out_flat)
# The freevars are being fanned out (not mapped). During transpose the
# dual of fan-out is fan-in-sum. We apply it to the unmapped invars.
assert len(in_axes) == len(arg_cts)
def unmap_zero(zero, in_axis):
return (zero if in_axis is None else Zero(
core.unmapped_aval(params['axis_size'], params['axis_name'],
in_axis, zero.aval)))
arg_cts = (unmap_zero(arg_ct, in_axis) if type(arg_ct) is Zero else
arg_ct if in_axis is not None else arg_ct.sum(0)
for arg_ct, in_axis in zip(arg_cts, in_axes))
return tuple(arg_cts)
def _closure_convert_for_avals(fun, in_tree, in_avals):
wrapped_fun, out_tree = flatten_fun_nokwargs(lu.wrap_init(fun), in_tree)
jaxpr, out_pvals, consts = pe.trace_to_jaxpr_dynamic(wrapped_fun, in_avals)
out_tree = out_tree()
(closure_consts, hoisted_consts), merge = partition_list(_maybe_perturbed, consts)
num_consts = len(hoisted_consts)
def converted_fun(*args_hconsts):
num_args = len(args_hconsts) - num_consts
args, hoisted_consts = split_list(args_hconsts, [num_args])
consts = merge(closure_consts, hoisted_consts)
all_args, in_tree2 = tree_flatten(tuple(args))
assert in_tree == in_tree2
out_flat = core.eval_jaxpr(jaxpr, consts, *all_args)
return tree_unflatten(out_tree, out_flat)
return converted_fun, hoisted_consts
def __call__(self, residual_arg, linear_arg):
res_arg, lin_arg = residual_arg, linear_arg
_, res_tree = tree_flatten(res_arg)
_, lin_tree = tree_flatten(lin_arg)
args_flat, in_tree = tree_flatten((res_arg, lin_arg))
flat_fun, out_tree = flatten_fun_nokwargs(lu.wrap_init(self.fun),
in_tree)
in_avals = [core.raise_to_shaped(core.get_aval(x)) for x in args_flat]
debug = pe.debug_info(self.fun, in_tree, False, "custom_transpose")
jaxpr, _, consts = pe.trace_to_jaxpr_dynamic(flat_fun, in_avals, debug)
assert not len(consts)
closed_call = core.ClosedJaxpr(pe.convert_constvars_jaxpr(jaxpr), ())
out_flat = custom_transpose_p.bind(*consts,
*args_flat,
call=closed_call,
rule=self.transpose,
lin_tree=lin_tree,
res_tree=res_tree,
out_tree=out_tree())
return tree_unflatten(out_tree(), out_flat)
def linear_call(fun: Callable, fun_transpose: Callable, residual_args,
linear_args):
"""Call a linear function, with a custom implementation for its transpose.
The type signatures of ``fun`` and ``fun_transpose`` are:
.. code-block:: haskell
fun :: r -> a -o b
fun_transpose :: r -> b -o a
where the ``-o`` arrow indicates a linear function, ``r`` is the
residual input type and ``a`` is the linear input type.
The functions ``fun`` and ``fun_transpose`` are coupled as
transposes of one another. Specifically, the transpose of a
``linear_call`` primitive is another ``linear_call`` to
``fun_transpose``, with ``fun`` as its custom transposition.
For example:
>>> def f(r, x):
... return x / r
>>> def t(r, t):
... return t / r
>>> def div_add(x, denom):
... return x + linear_call(f, t, denom, x)
>>> def transpose(f, x_example):
... def transposed(y):
... x, = jax.linear_transpose(f, x_example)(y)
... return x
... return transposed
>>> div_add(9., 3.)
DeviceArray(12., dtype=float32, weak_type=True)
>>> transpose(partial(div_add, denom=3.), 1.)(18.) # custom
DeviceArray(24., dtype=float32, weak_type=True)
>>> transpose(lambda x: x + x / 3., 1.)(18.) # reference
DeviceArray(24., dtype=float32, weak_type=True)
The above definition of ``f`` illustrates the purpose of a residual
argument: division is linear in one of its inputs (the dividend
``x``) but not the other (the divisor ``r``).
As another example:
>>> def custom_id(x):
... def f(_, x): return x
... def t(_, t): return 7.
... return linear_call(f, t, (), x)
>>> custom_id(1.)
1.0
>>> transpose(custom_id, 1.)(1.)
7.0
>>> transpose(transpose(custom_id, 1.), 1.)(1.)
1.0
>>> transpose(transpose(transpose(custom_id, 1.), 1.), 1.)(1.)
7.0
Args:
fun: a Python callable specifying a linear function. It should
take two arguments: one of "residual" inputs (type ``r``),
i.e. inputs in which the function is not necessarly linear, and
one of "linear" inputs (type ``a``). It should return output
whose components are linear in the linear input (type ``b``).
fun_transpose: a Python callable specifying a structurally linear
function that is the transpose of ``fun`` with respect to its
linear inputs. Its first argument is the same residual inputs
(``r``) as ``fun``. Its second argument is of type
``b``. Finally, its output is of type ``a`` and each of its
component are linear in its second argument (the ``b`` inputs).
residual_args: Argument in which ``fun`` and ``fun_transpose`` are
not necessarily linear. Not involved in transposition.
linear_args: Argument in which ``fun`` and ``fun_transpose`` are
linear and with respect to which the two are transposes.
Returns:
The call result, i.e. ``fun(residual_args, linear_args)``.
"""
operands_res, res_tree = tree_flatten(residual_args)
operands_lin, lin_tree = tree_flatten(linear_args)
f_in_tree = treedef_tuple((res_tree, lin_tree))
f, out_tree = flatten_fun_nokwargs(lu.wrap_init(fun), f_in_tree)
res_avals = map(abstractify, operands_res)
lin_avals = map(abstractify, operands_lin)
f_jaxpr, f_consts = _initial_style_jaxpr(f, (*res_avals, *lin_avals))
f_jaxpr = _close_jaxpr(f_jaxpr)
out_avals = map(core.raise_to_shaped, f_jaxpr.out_avals)
t_in_tree = treedef_tuple((res_tree, out_tree()))
t, t_out_tree = flatten_fun_nokwargs(lu.wrap_init(fun_transpose),
t_in_tree)
t_jaxpr, t_consts = _initial_style_jaxpr(t, (*res_avals, *out_avals))
t_jaxpr = _close_jaxpr(t_jaxpr)
if t_out_tree() != lin_tree:
raise TypeError(
'transpose output pytree structure must match that of linear inputs, '
f'got output structure {t_out_tree()} '
f'and input structure {lin_tree}.')
out = linear_call_p.bind(*f_consts,
*t_consts,
*operands_res,
*operands_lin,
callee=f_jaxpr,
transpose=t_jaxpr,
num_callee_consts=len(f_consts),
num_transpose_consts=len(t_consts),
num_res=len(operands_res))
return tree_unflatten(out_tree(), out)
def inv_backward_pass(jaxpr: core.Jaxpr, consts, primals_in, primals_out, cotangents_in):
if all(type(ct) is ad.Zero for ct in cotangents_in):
return map(lambda v: ad.Zero(v.aval), jaxpr.invars)
def write_cotangent(v, ct):
# assert v not in primal_env
if ct is not None and type(v) is not Literal:
ct_env[v] = ad.add_tangents(ct_env[v], ct) if v in ct_env else ct
def read_cotangent(v):
return ct_env.get(v, ad.Zero(v.aval))
def read_primal(v):
if type(v) is Literal:
return v.val
else:
return primal_env.get(v, ad.UndefinedPrimal(v.aval))
def write_primal(v, val):
if type(v) is Literal:
return
primal_env.setdefault(v, val)
# Invert while computing cotangents
ct_env: Dict[Any, Any] = {}
primal_env: Dict[Any, Any] = {}
write_primal(core.unitvar, core.unit)
map(write_primal, jaxpr.invars, primals_in)
map(write_primal, jaxpr.outvars, primals_out)
map(write_primal, jaxpr.constvars, consts)
map(write_cotangent, jaxpr.outvars, cotangents_in)
for eqn in jaxpr.eqns[::-1]:
primals_in = map(read_primal, eqn.invars)
primals_out = map(read_primal, eqn.outvars)
cts_in = map(read_cotangent, eqn.outvars)
should_invert = any(type(primal) is not ad.UndefinedPrimal
for primal in primals_out)
should_vjp = any(type(ct) is not ad.Zero for ct in cts_in)
assert not eqn.primitive.call_primitive
# Skip primals equations that are only jvp coefficients and don't affect
# primal outputs.
if not should_invert and not should_vjp:
continue
def abstract(value):
return raise_to_shaped(value.aval if ad.is_undefined_primal(value) else get_aval(value))
# Get the ivjp_jaxpr
if eqn.primitive is custom_ivjp_p:
ivjp_jaxpr = eqn.params['ivjp_jaxpr']
else:
if eqn.primitive in primitive_ivjps:
complete_ivjp = lu.wrap_init(primitive_ivjps[eqn.primitive])
else:
complete_ivjp = lu.wrap_init(partial(synthesize_ivjp, eqn, map(ad.is_undefined_primal, primals_in)))
_, in_tree = tree_flatten(
tuple(map(abstract, x) for x in (primals_in, primals_out, primals_out)))
complete_ivjp_flat, _ = flatten_fun_nokwargs(complete_ivjp, in_tree)
in_avals = map(abstract, primals_in + primals_out + primals_out)
# TODO: Actually we do know some of the inputs, because they might be literals!
ivjp_jaxpr, out_pvals, _ = pe.trace_to_jaxpr(
complete_ivjp_flat, map(pe.PartialVal.unknown, in_avals), instantiate=True)
assert not ivjp_jaxpr.constvars # That might happen some time, but don't bother until then
ivjp_jaxpr = core.ClosedJaxpr(ivjp_jaxpr, [])
# Once we know what the ivjp can do exactly, we have to isolate the part we are
# actually able to compute with the values we have at hand.
num_inputs = len(eqn.invars)
unknowns = (map(ad.is_undefined_primal, primals_in) +
map(ad.is_undefined_primal, primals_out) +
[False] * len(cts_in))
jaxpr_known, jaxpr_unknown, out_unknowns = pe.partial_eval_jaxpr( # type: ignore
ivjp_jaxpr, unknowns, instantiate=False) # type:ignore
unknown_rec_primals_in, unknown_cotangents = split_list(out_unknowns, [num_inputs])
# Make sure we're able to compute all cotangents. We don't really care if we
# can reconstruct or primals or not, although failure to do so might result in
# failing to compute cotangents later.
assert not any(unknown_cotangents)
# Remove residual outputs -- we won't be computing the unknown jaxpr anyway.
num_outputs = len(jaxpr_unknown.jaxpr.outvars)
jaxpr_known.jaxpr.outvars = jaxpr_known.jaxpr.outvars[:num_outputs]
# TODO: We could drop the outputs that correspond to primals that we already know.
# This only matters in eager mode, so leaving it out for now...
ivjp = core.jaxpr_as_fun(jaxpr_known)
rec_primals_in, cts_out = split_list(ivjp(*primals_in, *primals_out, *cts_in),
[num_inputs])
# Unknown rec_primals_in are core.units, so we have to replace them
# with UnknownPrimals because that's what write_primal will ignore.
rec_primals_in = [prev if unknown else rec
for prev, rec, unknown
in zip(primals_in, rec_primals_in, unknown_rec_primals_in)]
map(write_primal, eqn.invars, rec_primals_in)
map(write_cotangent, [v for v in eqn.invars if type(v) is not Literal], cts_out)
# NOTE: We keep the cotangents associated with primal variables, while the contract of a
# transpose is to return them in positions associated with tangent variables, which
# is what causes this whole confusion.
return map(read_cotangent, jaxpr.invars)
def wrapped_fun(*args): args_flat, in_tree = tree_flatten(args) f = lu.wrap_init(fun) flat_fun, out_tree = flatten_fun_nokwargs(f, in_tree) out_flat = callback_fun(flat_fun, args_flat, callback, strip_calls) return tree_unflatten(out_tree(), out_flat)
def _wrapped(*args): args_flat, in_tree = tree_flatten(args, is_leaf=_is_bcoo) wrapped_fun, out_tree = flatten_fun_nokwargs(lu.wrap_init(fun), in_tree) out = sparsify_fun(wrapped_fun, args_flat) return tree_unflatten(out_tree(), out)
