Update random.py

use `lax.nextafter` to get small values.

brainpy/_src/math/random.py

@@ -829,7 +829,9 @@ def truncated_normal(self, lower, upper, size=None, loc=0., scale=1., dtype=floa
     # [2l-1, 2u-1].
     key = self.split_key() if key is None else _formalize_key(key)
     # add a small value to avoid inf values after lax.erf_inv
-    out = jr.uniform(key, size, dtype, minval=2 * l - 1 + 1e-7, maxval=2 * u - 1 - 1e-7)
+    eps = lax.nextafter(0., np.array(np.inf, dtype=dtype))
+    out = jr.uniform(key, size, dtype, minval=2 * l - 1 + eps , maxval=2 * u - 1 - eps)
+
 
     # Use inverse cdf transform for normal distribution to get truncated
     # standard normal