Skip to content

Commit 6367a00

Browse files
authored
Merge pull request #58 from mikepapadim/refactor/cleanup_model_loaders
Refactor/cleanup model loaders
2 parents bbfe547 + eb13ea7 commit 6367a00

28 files changed

+886
-636
lines changed

src/main/java/org/beehive/gpullama3/LlamaApp.java

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,8 @@
11
package org.beehive.gpullama3;
22

3-
import org.beehive.gpullama3.aot.AOT;
43
import org.beehive.gpullama3.auxiliary.LastRunMetrics;
54
import org.beehive.gpullama3.inference.sampler.Sampler;
65
import org.beehive.gpullama3.model.Model;
7-
import org.beehive.gpullama3.model.loader.ModelLoader;
86

97
import java.io.IOException;
108

src/main/java/org/beehive/gpullama3/aot/AOT.java

Lines changed: 0 additions & 85 deletions
This file was deleted.

src/main/java/org/beehive/gpullama3/inference/weights/tornado/FP16Weights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/fp16/FP16Weights.java

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,7 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.fp16;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
4+
import org.beehive.gpullama3.inference.weights.tornado.TornadoWeights;
45
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;
56
import uk.ac.manchester.tornado.api.types.arrays.HalfFloatArray;
67

src/main/java/org/beehive/gpullama3/inference/weights/tornado/LlamaTornadoWeights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/fp16/LlamaTornadoWeights.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.fp16;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Phi3TornadoWeights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/fp16/Phi3TornadoWeights.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.fp16;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Qwen2TornadoWeights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/fp16/Qwen2TornadoWeights.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.fp16;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Qwen3TornadoWeights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/fp16/Qwen3TornadoWeights.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.fp16;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Phi3TornadoWeightsQ8_0.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/q8_0/Phi3TornadoWeightsQ8_0.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.q8_0;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import org.beehive.gpullama3.core.model.tensor.Q8_0QuantizedTensor;

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Q8_0Weights.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/q8_0/Q8_0Weights.java

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,8 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.q8_0;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import org.beehive.gpullama3.core.model.tensor.Q8_0QuantizedTensor;
5+
import org.beehive.gpullama3.inference.weights.tornado.TornadoWeights;
56
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;
67

78
public class Q8_0Weights implements TornadoWeights {

src/main/java/org/beehive/gpullama3/inference/weights/tornado/Qwen2TornadoWeightsQ8_0.java renamed to src/main/java/org/beehive/gpullama3/inference/weights/tornado/q8_0/Qwen2TornadoWeightsQ8_0.java

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
package org.beehive.gpullama3.inference.weights.tornado;
1+
package org.beehive.gpullama3.inference.weights.tornado.q8_0;
22

33
import org.beehive.gpullama3.core.model.GGMLType;
44
import org.beehive.gpullama3.core.model.tensor.Q8_0QuantizedTensor;

0 commit comments

Comments
 (0)