Merge branch 'main' of https://github.com/jafioti/luminal

README.md

@@ -15,7 +15,7 @@ use luminal::prelude::*;
 // Setup graph and tensors
 let mut cx = Graph::new();
 let a = cx.tensor((3, 1)).set([[1.0], [2.0], [3.0]]);
-let b = cx.tensor((1, 4).set([[1.0, 2.0, 3.0, 4.0]]);
+let b = cx.tensor((1, 4)).set([[1.0, 2.0, 3.0, 4.0]]);
 
 // Do math...
 let mut c = a.matmul(b).retrieve();

src/hl_ops/binary.rs

@@ -282,9 +282,9 @@ impl GraphTensor {
         -(-self).max_f32(-rhs)
     }
 
-    /// Clip a tensor in a range
+    /// Clip (clamp) a tensor into the range [`min`, `max`]
     pub fn clip(self, min: f32, max: f32) -> GraphTensor {
-        self.min_f32(min).max_f32(max)
+        self.max_f32(min).min_f32(max)
     }
 }
 
@@ -297,3 +297,24 @@ impl F32Pow for f32 {
         e.mul(self.abs().ln()).exp().recip()
     }
 }
+
+#[cfg(test)]
+mod tests {
+    crate::test_imports!();
+
+    #[test]
+    fn test_clip() {
+        let mut cx = Graph::new();
+        let a = cx
+            .tensor((3, 2))
+            .set([[[-1.0], [-2.0], [3.0]], [[-1.5], [0.0], [5.0]]]);
+        let result = a.clip(-1.5, 3.4).retrieve();
+        let expected_result = cx
+            .tensor((3, 2))
+            .set([[[-1.0], [-1.5], [3.0]], [[-1.5], [0.0], [3.4]]])
+            .retrieve();
+        cx.execute();
+
+        assert_close(&result.data(), &expected_result.data());
+    }
+}