diff --git a/harness/onnx-test-suite/node-1.10.1.txt b/harness/onnx-test-suite/node-1.10.1.txt
index 46c03344f0..176414276e 100644
--- a/harness/onnx-test-suite/node-1.10.1.txt
+++ b/harness/onnx-test-suite/node-1.10.1.txt
@@ -299,6 +299,7 @@ test_matmul_3d
 test_matmul_4d
 test_matmulinteger                                                               
 test_max_example
+test_max_float16
 test_max_float32
 test_max_float64
 test_max_int16
@@ -328,6 +329,7 @@ test_mean_example
 test_mean_one_input
 test_mean_two_inputs
 test_min_example
+test_min_float16
 test_min_float32
 test_min_float64
 test_min_int16
@@ -527,6 +529,7 @@ test_reshape_reordered_last_dims input:data
 test_reshape_zero_and_negative_dim input:data
 test_reshape_zero_dim input:data
 test_resize_upsample_scales_linear_align_corners                                    input:X not-nnef
+test_resize_upsample_scales_nearest not-nnef not-typable
 test_rnn_seq_length
 test_round
 test_scan9_sum
diff --git a/harness/onnx-test-suite/node-1.7.0.txt b/harness/onnx-test-suite/node-1.7.0.txt
index 352edfea42..b3ffa70a39 100644
--- a/harness/onnx-test-suite/node-1.7.0.txt
+++ b/harness/onnx-test-suite/node-1.7.0.txt
@@ -269,6 +269,7 @@ test_matmul_3d
 test_matmul_4d
 test_matmulinteger                                                               
 test_max_example
+test_max_float16
 test_max_float32
 test_max_float64
 test_max_int16
@@ -298,6 +299,7 @@ test_mean_example
 test_mean_one_input
 test_mean_two_inputs
 test_min_example
+test_min_float16
 test_min_float32
 test_min_float64
 test_min_int16
diff --git a/harness/onnx-test-suite/node-1.8.1.txt b/harness/onnx-test-suite/node-1.8.1.txt
index 40b63657b8..a90b29d363 100644
--- a/harness/onnx-test-suite/node-1.8.1.txt
+++ b/harness/onnx-test-suite/node-1.8.1.txt
@@ -284,6 +284,7 @@ test_matmul_3d
 test_matmul_4d
 test_matmulinteger                                                               
 test_max_example
+test_max_float16
 test_max_float32
 test_max_float64
 test_max_int16
@@ -313,6 +314,7 @@ test_mean_example
 test_mean_one_input
 test_mean_two_inputs
 test_min_example
+test_min_float16
 test_min_float32
 test_min_float64
 test_min_int16
diff --git a/harness/onnx-test-suite/node-1.9.0.txt b/harness/onnx-test-suite/node-1.9.0.txt
index 8120ff19df..bf02c4ce13 100644
--- a/harness/onnx-test-suite/node-1.9.0.txt
+++ b/harness/onnx-test-suite/node-1.9.0.txt
@@ -290,6 +290,7 @@ test_matmul_3d
 test_matmul_4d
 test_matmulinteger                                                               
 test_max_example
+test_max_float16
 test_max_float32
 test_max_float64
 test_max_int16
@@ -319,6 +320,7 @@ test_mean_example
 test_mean_one_input
 test_mean_two_inputs
 test_min_example
+test_min_float16
 test_min_float32
 test_min_float64
 test_min_int16
diff --git a/onnx/src/ops/cumsum.rs b/onnx/src/ops/cumsum.rs
index 094b619395..c9eedde7f9 100644
--- a/onnx/src/ops/cumsum.rs
+++ b/onnx/src/ops/cumsum.rs
@@ -39,19 +39,23 @@ impl Expansion for CumSum {
         let axis =
             model.outlet_fact(inputs[1])?.konst.as_ref().context("Axis expected to be a const")?;
         let axis = axis.cast_to_scalar::<i64>()?;
-        let data = model.outlet_fact(inputs[0])?;
-        let axis = if axis < 0 { (axis + data.rank() as i64) as usize } else { axis as usize };
+        let data = model.outlet_fact(inputs[0])?.clone();
         let mut var_shape = data.shape.clone();
+        let axis = if axis < 0 { (axis + data.rank() as i64) as usize } else { axis as usize };
+        let zero = model.add_const(
+            format!("{}.zero", prefix),
+            Tensor::zero_dt(data.datum_type, &[])?.into_arc_tensor(),
+        )?;
         var_shape.set(axis, 1.to_dim());
-        let var_shape = var_shape.as_concrete().context("Expect shapes to be known")?;
+        let init = model.wire_node(
+            format!("{}.init", prefix),
+            tract_core::ops::array::MultiBroadcastTo::new(var_shape.clone().into()),
+            &[zero],
+        )?[0];
         let chunk = if self.reverse { -1 } else { 1 };
         let input_mapping = vec![
             scan::InputMapping::Scan { slot: 0, axis, chunk },
-            scan::InputMapping::State {
-                initializer: scan::StateInitializer::Value(
-                    Tensor::zero_dt(data.datum_type, var_shape)?.into_arc_tensor(),
-                ),
-            },
+            scan::InputMapping::State { initializer: scan::StateInitializer::FromInput(1) },
         ];
         let output_mapping = vec![
             scan::OutputMapping {
@@ -82,7 +86,7 @@ impl Expansion for CumSum {
             body.set_output_outlets(&[sum, sum])?;
         }
         let scan = scan::Scan::new(body, input_mapping, output_mapping, None, 0)?;
-        model.wire_node(prefix, scan, &inputs[0..1])
+        model.wire_node(prefix, scan, &[inputs[0], init])
     }
 
     fn rules<'r, 'p: 'r, 's: 'r>(