Update scikit-learn requirement from <1.5.0 to <1.6.0 (intel#31)

* Update scikit-learn requirement from <1.5.0 to <1.6.0

Updates the requirements on [scikit-learn](https://github.com/scikit-learn/scikit-learn) to permit the latest version.
- [Release notes](https://github.com/scikit-learn/scikit-learn/releases)
- [Commits](scikit-learn/scikit-learn@0.1-beta...1.5.0)

---
updated-dependencies:
- dependency-name: scikit-learn
  dependency-type: direct:development
...

Signed-off-by: dependabot[bot] <support@github.com>

* Update tests

* Fix

* fix last test

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Alessandro Palla <alessandro.palla@intel.com>

dev_requirements.txt

@@ -1,7 +1,7 @@
 pytest
 pytest-xdist
 pytest-cov
-scikit-learn < 1.5.0
+scikit-learn <= 1.5.0
 pre-commit; sys_platform == 'darwin'
 sphinx
 breathe

test/python/test_compile.py

@@ -62,7 +62,7 @@ def test_compilation(dtype):
     y2 = compiled_model(x).detach()
     t2 = time.perf_counter()
 
-    assert 1 - r2_score(y_ref, y1) < 0.01
+    assert 1 - r2_score(y_ref.numpy(), y1.numpy()) < 0.01
 
     assert torch.allclose(y1, y2)
 
@@ -88,7 +88,7 @@ def test_torch_compile():
     else:
         compiled_model = torch.compile(model, backend="npu")
         y = compiled_model(x).detach()
-        assert 1 - r2_score(y_ref, y) < 0.01
+        assert 1 - r2_score(y_ref.numpy(), y.numpy()) < 0.01
 
 
 @pytest.mark.parametrize("dtype", [torch.float32, torch.float16, torch.int8])

test/python/test_conv.py

@@ -42,6 +42,6 @@ def test_conv(in_channels, out_channels, kernels, dim, bias, dtype):
         assert y.dtype == y_ref.dtype
         assert y.shape == y_ref.shape
         if dtype == torch.int8:
-            assert 1 - r2_score(y_ref.flatten(), y.flatten()) < 0.05
+            assert 1 - r2_score(y_ref.flatten().numpy(), y.flatten().numpy()) < 0.05
         else:
-            assert 1 - r2_score(y_ref.flatten(), y.flatten()) < 0.001
+            assert 1 - r2_score(y_ref.flatten().numpy(), y.flatten().numpy()) < 0.001

test/python/test_llm.py

@@ -51,7 +51,7 @@ def test_compilation(tokenizer, decoder_model, dtype):
 
     y = compiled_model(prefill).logits.detach()
 
-    assert 1 - r2_score(y_ref.flatten(), y.flatten()) < 0.01
+    assert 1 - r2_score(y_ref.flatten().numpy(), y.flatten().numpy()) < 0.01
 
 
 @torch.no_grad
@@ -75,5 +75,4 @@ def test_phi2_mlp(seq_len, hidden_size, intermediate_size):
 
     out = model(x)
 
-    assert 1 - r2_score(reference, out) < 0.001
-    print(r2_score(reference, out))
+    assert 1 - r2_score(reference.numpy(), out.numpy()) < 0.001

test/python/test_matmul.py

@@ -34,7 +34,7 @@ def test_matmul(batch, inC, outC):
 
     assert np.isfinite(npu_mm).all()
 
-    assert 1 - r2_score(cpu_mm, npu_mm) < 0.001
+    assert 1 - r2_score(cpu_mm.numpy(), npu_mm) < 0.001
 
 
 @pytest.mark.parametrize(
@@ -58,7 +58,8 @@ def test_qmatmul_per_channel_scales(batch, inC, outC):
     assert weights_quant.shape == W.shape
 
     # Conversion done properly
-    assert 1 - r2_score(weights_quant.to(torch.float16) * scale, W) < 0.001
+    expected_W = weights_quant.to(torch.float16) * scale
+    assert 1 - r2_score(expected_W.numpy(), W.numpy()) < 0.001
 
     mm = QMatMul(inC, outC, batch)
 
@@ -68,4 +69,4 @@ def test_qmatmul_per_channel_scales(batch, inC, outC):
 
     assert np.isfinite(npu_mm).all()
 
-    assert 1 - r2_score(cpu_mm, npu_mm) < 0.001
+    assert 1 - r2_score(cpu_mm.numpy(), npu_mm) < 0.001

test/python/test_optimizations.py

@@ -146,7 +146,7 @@ def test_model(model_name, hidden_size, intermediate_size, sequence_length, bias
 
         output = optimized(example_input)[0]
 
-    assert 1 - r2_score(reference.flatten(), output.flatten()) < 0.01
+    assert 1 - r2_score(reference.flatten().numpy(), output.flatten().numpy()) < 0.01
 
 
 @pytest.mark.parametrize("layers", [2, 3, 10])

test/python/test_quantization.py

@@ -93,7 +93,7 @@ def test_compiled_quantized(batch, inC, outC):
 
     y1 = compiled_model(X).detach()
 
-    assert 1 - r2_score(y_ref, y1) < 0.01
+    assert 1 - r2_score(y_ref.numpy(), y1.numpy()) < 0.01
 
 
 @pytest.mark.parametrize("batch", [16, 128])