instmodel 1.2.0__tar.gz → 1.2.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {instmodel-1.2.0/instmodel.egg-info → instmodel-1.2.2}/PKG-INFO +1 -1
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel/instruction_model.py +8 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel/tf.py +6 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel/torch.py +8 -1
- {instmodel-1.2.0 → instmodel-1.2.2/instmodel.egg-info}/PKG-INFO +1 -1
- {instmodel-1.2.0 → instmodel-1.2.2}/setup.cfg +1 -1
- {instmodel-1.2.0 → instmodel-1.2.2}/tests/test_tf.py +57 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/tests/test_torch.py +73 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/LICENSE +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/MANIFEST.in +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/README.md +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel/__init__.py +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel/training_utils.py +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel.egg-info/SOURCES.txt +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel.egg-info/dependency_links.txt +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel.egg-info/requires.txt +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/instmodel.egg-info/top_level.txt +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/pyproject.toml +0 -0
- {instmodel-1.2.0 → instmodel-1.2.2}/setup.py +0 -0
|
@@ -119,6 +119,14 @@ def apply_activation(instruction, buffers, output_index):
|
|
|
119
119
|
)
|
|
120
120
|
buffers[output_index] = result
|
|
121
121
|
|
|
122
|
+
elif act == "EXP":
|
|
123
|
+
# Clamp to [-88, 88] to avoid float32 overflow/underflow
|
|
124
|
+
buffers[output_index] = np.exp(np.clip(x, -88, 88))
|
|
125
|
+
|
|
126
|
+
elif act == "SIGN":
|
|
127
|
+
# sign(x): -1 for x < 0, 0 for x == 0, 1 for x > 0
|
|
128
|
+
buffers[output_index] = np.sign(x)
|
|
129
|
+
|
|
122
130
|
else:
|
|
123
131
|
raise ValueError(f"Unexpected activation: {act}")
|
|
124
132
|
|
|
@@ -512,6 +512,12 @@ class ActivationComputation(ComputationOp):
|
|
|
512
512
|
)
|
|
513
513
|
elif self.activation == "INVERSE":
|
|
514
514
|
self.keras_layer = layers.Lambda(lambda x: 1 - x, name=name)
|
|
515
|
+
elif self.activation == "EXP":
|
|
516
|
+
self.keras_layer = layers.Lambda(
|
|
517
|
+
lambda x: tf.exp(tf.clip_by_value(x, -88, 88)), name=name
|
|
518
|
+
)
|
|
519
|
+
elif self.activation == "SIGN":
|
|
520
|
+
self.keras_layer = layers.Lambda(lambda x: tf.sign(x), name=name)
|
|
515
521
|
else:
|
|
516
522
|
raise ValueError(f"Unexpected activation: {self.activation}")
|
|
517
523
|
|
|
@@ -48,6 +48,10 @@ def _apply_activation(x: torch.Tensor, activation: str) -> torch.Tensor:
|
|
|
48
48
|
)
|
|
49
49
|
elif act == "INVERSE":
|
|
50
50
|
return 1 - x
|
|
51
|
+
elif act == "EXP":
|
|
52
|
+
return torch.exp(torch.clamp(x, -88, 88))
|
|
53
|
+
elif act == "SIGN":
|
|
54
|
+
return torch.sign(x)
|
|
51
55
|
else:
|
|
52
56
|
raise ValueError(f"Unexpected activation: {activation}")
|
|
53
57
|
|
|
@@ -1301,7 +1305,10 @@ class ModelGraph(ComputationOp):
|
|
|
1301
1305
|
history.append(avg_loss)
|
|
1302
1306
|
|
|
1303
1307
|
if verbose:
|
|
1304
|
-
|
|
1308
|
+
if abs(avg_loss) > 1e-3:
|
|
1309
|
+
print(f"Epoch {epoch + 1}/{epochs} - loss: {avg_loss:.4f}")
|
|
1310
|
+
else:
|
|
1311
|
+
print(f"Epoch {epoch + 1}/{epochs} - loss: {avg_loss:.4e}")
|
|
1305
1312
|
|
|
1306
1313
|
return history
|
|
1307
1314
|
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
from instmodel.tf import (
|
|
2
|
+
ActivationComputation,
|
|
2
3
|
Attention,
|
|
3
4
|
Concatenate,
|
|
4
5
|
Dense,
|
|
@@ -983,3 +984,59 @@ def test_multiply_heads_scalar():
|
|
|
983
984
|
# Validate using instruction_model_inference directly (supports list inputs)
|
|
984
985
|
*_, inst_output = instruction_model_inference(inst_model, [data, scalar])
|
|
985
986
|
assert np.allclose(inst_output, keras_pred, atol=1e-6)
|
|
987
|
+
|
|
988
|
+
|
|
989
|
+
def test_exp_activation():
|
|
990
|
+
"""Tests EXP activation: Keras Lambda, NumPy, and instruction model agree; clamped at [-88, 88]."""
|
|
991
|
+
x_np = np.array([-100.0, -88.0, -1.0, 0.0, 1.0, 88.0, 100.0], dtype=np.float32)
|
|
992
|
+
expected = np.exp(np.clip(x_np, -88, 88).astype(np.float64)).astype(np.float32)
|
|
993
|
+
|
|
994
|
+
# TF/Keras backend via ActivationComputation
|
|
995
|
+
inp = InputBuffer(len(x_np))
|
|
996
|
+
out = ActivationComputation("EXP", in_place=False)(inp)
|
|
997
|
+
model = ModelGraph(inp, out)
|
|
998
|
+
result_tf = model.predict(x_np.reshape(1, -1), verbose=0).flatten()
|
|
999
|
+
assert np.allclose(result_tf, expected, atol=1e-6), (
|
|
1000
|
+
f"TF EXP mismatch: {result_tf} vs {expected}"
|
|
1001
|
+
)
|
|
1002
|
+
|
|
1003
|
+
# Instruction model round-trip
|
|
1004
|
+
inst_model = model.create_instruction_model()
|
|
1005
|
+
inst_model["validation_data"] = {
|
|
1006
|
+
"inputs": x_np.reshape(1, -1).tolist(),
|
|
1007
|
+
"expected_outputs": result_tf.reshape(1, -1).tolist(),
|
|
1008
|
+
}
|
|
1009
|
+
validate_instruction_model(inst_model)
|
|
1010
|
+
|
|
1011
|
+
# Clamping: exp(-100) == exp(-88), exp(100) == exp(88)
|
|
1012
|
+
assert np.isclose(result_tf[0], result_tf[1], atol=1e-6), "EXP(-100) should be clamped to exp(-88)"
|
|
1013
|
+
assert np.isclose(result_tf[-1], result_tf[-2], atol=1e-6), "EXP(100) should be clamped to exp(88)"
|
|
1014
|
+
|
|
1015
|
+
|
|
1016
|
+
def test_sign_activation():
|
|
1017
|
+
"""Tests SIGN activation: Keras Lambda, NumPy, and instruction model agree; returns -1, 0, or 1."""
|
|
1018
|
+
x_np = np.array([-5.0, -0.01, 0.0, 0.01, 5.0], dtype=np.float32)
|
|
1019
|
+
expected = np.array([-1.0, -1.0, 0.0, 1.0, 1.0], dtype=np.float32)
|
|
1020
|
+
|
|
1021
|
+
# TF/Keras backend via ActivationComputation
|
|
1022
|
+
inp = InputBuffer(len(x_np))
|
|
1023
|
+
out = ActivationComputation("SIGN", in_place=False)(inp)
|
|
1024
|
+
model = ModelGraph(inp, out)
|
|
1025
|
+
result_tf = model.predict(x_np.reshape(1, -1), verbose=0).flatten()
|
|
1026
|
+
assert np.array_equal(result_tf, expected), (
|
|
1027
|
+
f"TF SIGN mismatch: {result_tf} vs {expected}"
|
|
1028
|
+
)
|
|
1029
|
+
|
|
1030
|
+
# Instruction model round-trip
|
|
1031
|
+
inst_model = model.create_instruction_model()
|
|
1032
|
+
inst_model["validation_data"] = {
|
|
1033
|
+
"inputs": x_np.reshape(1, -1).tolist(),
|
|
1034
|
+
"expected_outputs": result_tf.reshape(1, -1).tolist(),
|
|
1035
|
+
}
|
|
1036
|
+
validate_instruction_model(inst_model)
|
|
1037
|
+
|
|
1038
|
+
# RELU(SIGN(x)) == STEP(x)
|
|
1039
|
+
step_expected = np.array([0.0, 0.0, 0.0, 1.0, 1.0], dtype=np.float32)
|
|
1040
|
+
assert np.array_equal(np.maximum(0, result_tf), step_expected), (
|
|
1041
|
+
"RELU(SIGN(x)) should produce step function"
|
|
1042
|
+
)
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
from instmodel.torch import (
|
|
2
|
+
ActivationComputation,
|
|
2
3
|
Attention,
|
|
3
4
|
Concatenate,
|
|
4
5
|
Dense,
|
|
@@ -917,3 +918,75 @@ def test_multiply_heads_scalar():
|
|
|
917
918
|
|
|
918
919
|
*_, inst_output = instruction_model_inference(inst_model, [data, scalar])
|
|
919
920
|
assert np.allclose(inst_output, torch_pred, atol=1e-6)
|
|
921
|
+
|
|
922
|
+
|
|
923
|
+
def test_exp_activation():
|
|
924
|
+
"""Tests EXP activation: PyTorch, NumPy, and instruction model agree; clamped at [-88, 88]."""
|
|
925
|
+
from instmodel.torch import _apply_activation
|
|
926
|
+
|
|
927
|
+
x_np = np.array([-100.0, -88.0, -1.0, 0.0, 1.0, 88.0, 100.0], dtype=np.float32)
|
|
928
|
+
expected = np.exp(np.clip(x_np, -88, 88).astype(np.float64)).astype(np.float32)
|
|
929
|
+
|
|
930
|
+
# PyTorch backend
|
|
931
|
+
result_torch = _apply_activation(torch.tensor(x_np), "EXP").numpy()
|
|
932
|
+
assert np.allclose(result_torch, expected, atol=1e-6), (
|
|
933
|
+
f"PyTorch EXP mismatch: {result_torch} vs {expected}"
|
|
934
|
+
)
|
|
935
|
+
|
|
936
|
+
# ActivationComputation + ModelGraph round-trip
|
|
937
|
+
inp = InputBuffer(len(x_np))
|
|
938
|
+
out = ActivationComputation("EXP", in_place=False)(inp)
|
|
939
|
+
model = ModelGraph(inp, out)
|
|
940
|
+
model_pred = model.predict(x_np.reshape(1, -1), verbose=0).flatten()
|
|
941
|
+
assert np.allclose(model_pred, expected, atol=1e-6), (
|
|
942
|
+
f"ModelGraph EXP mismatch: {model_pred} vs {expected}"
|
|
943
|
+
)
|
|
944
|
+
|
|
945
|
+
# Instruction model round-trip
|
|
946
|
+
inst_model = model.create_instruction_model()
|
|
947
|
+
inst_model["validation_data"] = {
|
|
948
|
+
"inputs": x_np.reshape(1, -1).tolist(),
|
|
949
|
+
"expected_outputs": model_pred.reshape(1, -1).tolist(),
|
|
950
|
+
}
|
|
951
|
+
validate_instruction_model(inst_model)
|
|
952
|
+
|
|
953
|
+
# Clamping: exp(-100) == exp(-88), exp(100) == exp(88)
|
|
954
|
+
assert np.isclose(result_torch[0], result_torch[1], atol=1e-6), "EXP(-100) should be clamped to exp(-88)"
|
|
955
|
+
assert np.isclose(result_torch[-1], result_torch[-2], atol=1e-6), "EXP(100) should be clamped to exp(88)"
|
|
956
|
+
|
|
957
|
+
|
|
958
|
+
def test_sign_activation():
|
|
959
|
+
"""Tests SIGN activation: PyTorch, NumPy, and instruction model agree; returns -1, 0, or 1."""
|
|
960
|
+
from instmodel.torch import _apply_activation
|
|
961
|
+
|
|
962
|
+
x_np = np.array([-5.0, -0.01, 0.0, 0.01, 5.0], dtype=np.float32)
|
|
963
|
+
expected = np.array([-1.0, -1.0, 0.0, 1.0, 1.0], dtype=np.float32)
|
|
964
|
+
|
|
965
|
+
# PyTorch backend
|
|
966
|
+
result_torch = _apply_activation(torch.tensor(x_np), "SIGN").numpy()
|
|
967
|
+
assert np.array_equal(result_torch, expected), (
|
|
968
|
+
f"PyTorch SIGN mismatch: {result_torch} vs {expected}"
|
|
969
|
+
)
|
|
970
|
+
|
|
971
|
+
# ActivationComputation + ModelGraph round-trip
|
|
972
|
+
inp = InputBuffer(len(x_np))
|
|
973
|
+
out = ActivationComputation("SIGN", in_place=False)(inp)
|
|
974
|
+
model = ModelGraph(inp, out)
|
|
975
|
+
model_pred = model.predict(x_np.reshape(1, -1), verbose=0).flatten()
|
|
976
|
+
assert np.array_equal(model_pred, expected), (
|
|
977
|
+
f"ModelGraph SIGN mismatch: {model_pred} vs {expected}"
|
|
978
|
+
)
|
|
979
|
+
|
|
980
|
+
# Instruction model round-trip
|
|
981
|
+
inst_model = model.create_instruction_model()
|
|
982
|
+
inst_model["validation_data"] = {
|
|
983
|
+
"inputs": x_np.reshape(1, -1).tolist(),
|
|
984
|
+
"expected_outputs": model_pred.reshape(1, -1).tolist(),
|
|
985
|
+
}
|
|
986
|
+
validate_instruction_model(inst_model)
|
|
987
|
+
|
|
988
|
+
# RELU(SIGN(x)) == STEP(x)
|
|
989
|
+
step_expected = np.array([0.0, 0.0, 0.0, 1.0, 1.0], dtype=np.float32)
|
|
990
|
+
assert np.array_equal(np.maximum(0, result_torch), step_expected), (
|
|
991
|
+
"RELU(SIGN(x)) should produce step function"
|
|
992
|
+
)
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|