[custom_vjp] bwd function should not be WrappedFun, may run multiple …

…times

jax/_src/custom_derivatives.py

@@ -560,7 +560,7 @@ def __call__(self, *args: Any, **kwargs: Any) -> ReturnValue:  # pytype: disable
       flat_fun, out_type = _flatten_fun_nokwargs(f_, in_tree)
       flat_fwd, out_trees = _flatten_fwd(fwd, primal_name, fwd_name, in_tree,
                                          out_type)
-      flat_bwd = _flatten_bwd(bwd, in_tree, in_avals, out_trees)
+      flat_bwd = _flatten_bwd(bwd, in_tree, in_avals, out_trees).call_wrapped
       out_flat = custom_vjp_call_p.bind(flat_fun, flat_fwd, flat_bwd,
                                         *args_flat, out_trees=out_trees)
       _, (out_tree, _) = lu.merge_linear_aux(out_type, out_trees)
@@ -680,7 +680,7 @@ def bind(self, fun, fwd, bwd, *args, out_trees):
     fwd, env_trace_todo2 = process_env_traces_fwd(
       fwd, top_trace and top_trace.level, out_trees)
     tracers = map(top_trace.full_raise, args)  # type: ignore
-    bwd_ = lu.wrap_init(lambda *args: bwd.call_wrapped(*args))
+    bwd_ = lambda *args: bwd(*args)
     outs = top_trace.process_custom_vjp_call(self, fun, fwd, bwd_, tracers,
                                              out_trees=out_trees)
     fst, env_trace_todo = lu.merge_linear_aux(env_trace_todo1, env_trace_todo2)
@@ -749,7 +749,7 @@ def _custom_vjp_call_jaxpr_abstract_eval(*_, fun_jaxpr, **__):
 def _custom_vjp_call_jaxpr_jvp(
     primals, tangents, *, fun_jaxpr: core.ClosedJaxpr,
     fwd_jaxpr_thunk: Callable[[], Tuple[core.Jaxpr, Sequence[Any]]],
-    bwd: lu.WrappedFun, out_trees: Callable, num_consts: int):
+    bwd: Callable, out_trees: Callable, num_consts: int):
   _, args = split_list(primals, [num_consts])
   consts_dot, args_dot = split_list(tangents, [num_consts])
   if any(type(t) is not Zero for t in consts_dot):
@@ -772,7 +772,7 @@ def _custom_vjp_call_jaxpr_jvp(
 def _custom_vjp_call_jaxpr_vmap(spmd_axis_name,
     axis_size, axis_name, main_type, args, in_dims, *, fun_jaxpr: core.ClosedJaxpr,
     fwd_jaxpr_thunk: Callable[[], Tuple[core.Jaxpr, Sequence[Any]]],
-    bwd: lu.WrappedFun, out_trees: Callable, num_consts: int):
+    bwd: Callable, out_trees: Callable, num_consts: int):
   args = [batching.moveaxis(x, d, 0) if d is not not_mapped and d != 0
           else x for x, d in zip(args, in_dims)]
 

jax/_src/interpreters/ad.py

@@ -748,7 +748,7 @@ def raise_custom_vjp_error_on_jvp(*_, **__):
 def _custom_lin_transpose(cts_out, *invals, num_res, bwd, out_avals):
   res, _ = split_list(invals, [num_res])
   cts_out = map(instantiate_zeros_aval, out_avals, cts_out)
-  cts_in = bwd.call_wrapped(*res, *cts_out)
+  cts_in = bwd(*res, *cts_out)
   return [None] * num_res + list(cts_in)
 primitive_transposes[custom_lin_p] = _custom_lin_transpose
 

jax/_src/interpreters/batching.py

@@ -776,11 +776,16 @@ def batch_custom_jvp_subtrace(main, in_dims, *in_vals):
                      out_tangent_bds, out_dims, out_tangents)
   yield out_primals + out_tangents, out_dims * 2
 
-def batch_custom_vjp_bwd(bwd, axis_name, axis_size, in_dims, out_dim_dests, main_type, spmd_axis_name):
-  bwd, out_dims_thunk = batch_subtrace(bwd)
-  bwd_ = _batch_outer(bwd, axis_name, axis_size, in_dims, main_type,
-                      spmd_axis_name)
-  return _match_axes_and_sum(bwd_, axis_size, axis_name, out_dims_thunk, out_dim_dests)
+def batch_custom_vjp_bwd(bwd, axis_name, axis_size, in_dims, out_dim_dests,
+                         main_type, spmd_axis_name):
+  def new_bwd(*args):
+    bwd_, out_dims_thunk = batch_subtrace(lu.wrap_init(bwd))
+    bwd_ = _batch_outer(bwd_, axis_name, axis_size, in_dims, main_type,
+                        spmd_axis_name)
+    bwd_ = _match_axes_and_sum(bwd_, axis_size, axis_name, out_dims_thunk,
+                               out_dim_dests)
+    return bwd_.call_wrapped(*args)
+  return new_bwd
 
 @lu.transformation
 def _match_axes_and_sum(axis_size, axis_name, out_dims_thunk, out_dim_dests, *in_vals):

jax/_src/tree_util.py

@@ -308,6 +308,9 @@ def __eq__(self, other):
       return self.fun == other.fun
     return self.fun == other
 
+  def __repr__(self):
+    return f'_HashableCallableShim({repr(self.fun)})'
+
 
 class Partial(functools.partial):
   """A version of functools.partial that works in pytrees.

tests/api_test.py

@@ -8385,6 +8385,18 @@ def f_bwd(_, g):
 
     jax.grad(f)(A([1.]))  # doesn't crash
 
+  def test_vmap_vjp_called_twice(self):
+    # https://github.com/google/jax/pull/14728
+    @jax.custom_vjp
+    def f(x):
+      return x
+    f.defvjp(lambda x: (x, None), lambda _, y_bar: (y_bar,))
+
+    _, f_vjp = jax.vjp(jax.vmap(f), jnp.array([3.]))
+    f_vjp(jnp.array([3.]))
+    f_vjp(jnp.array([3.]))  # doesn't crash
+
+
 def transpose_unary(f, x_example):
   def transposed(y):
     x, = api.linear_transpose(f, x_example)(y)