jax.nn.softmax: fix fill value when where is specified

jax/_src/nn/functions.py

@@ -317,7 +317,10 @@ def log_softmax(x: Array,
   shifted = x - lax.stop_gradient(x_max)
   shifted_logsumexp = jnp.log(
       jnp.sum(jnp.exp(shifted), axis, where=where, keepdims=True))
-  return shifted - shifted_logsumexp
+  result = shifted - shifted_logsumexp
+  if where is not None:
+    return jnp.where(where, result, -jnp.inf)
+  return result
 
 
 # TODO(phawkins): this jit was found to change numerics in a test. Debug this.
@@ -357,7 +360,10 @@ def _softmax(
     initial: Optional[Array] = None) -> Array:
   x_max = jnp.max(x, axis, where=where, initial=initial, keepdims=True)
   unnormalized = jnp.exp(x - x_max)
-  return unnormalized / jnp.sum(unnormalized, axis, where=where, keepdims=True)
+  result = unnormalized / jnp.sum(unnormalized, axis, where=where, keepdims=True)
+  if where is not None:
+    result = jnp.where(where, result, 0)
+  return result
 
 @_softmax.defjvp
 def _softmax_jvp(axis, primals, tangents):
@@ -368,7 +374,10 @@ def _softmax_jvp(axis, primals, tangents):
 def _softmax_deprecated(x, axis, where, initial):
   x_max = jnp.max(x, axis, where=where, initial=initial, keepdims=True)
   unnormalized = jnp.exp(x - lax.stop_gradient(x_max))
-  return unnormalized / jnp.sum(unnormalized, axis, where=where, keepdims=True)
+  result = unnormalized / jnp.sum(unnormalized, axis, where=where, keepdims=True)
+  if where is not None:
+    result = jnp.where(where, result, 0)
+  return result
 
 
 @partial(jax.jit, static_argnames=("axis",))

tests/nn_test.py

@@ -133,13 +133,12 @@ def testHardTanhMemory(self):
   def testSoftmaxWhereMask(self, fn):
     x = jnp.array([5.5, 1.3, -4.2, 0.9])
     m = jnp.array([True, False, True, True])
-    x_filtered = jnp.take(x, jnp.array([0, 2, 3]))
 
-    out_masked = jnp.take(
-        fn(x, where=m, initial=-jnp.inf), jnp.array([0, 2, 3]))
-    out_filtered = fn(x_filtered)
+    out = fn(x, where=m, initial=-jnp.inf)
+    self.assertAllClose(out[m], fn(x[m]))
 
-    self.assertAllClose(out_masked, out_filtered)
+    probs = out if fn is nn.softmax else jnp.exp(out)
+    self.assertAllClose(probs.sum(), 1.0)
 
     # TODO(mattjj): include log_softmax in these extra tests if/when we add a
     # custom_jvp rule for it (since otherwise it doesn't pass the numerical