Merge pull request beehive-lab#65 from orionpapadakis/refactor/weight_abstraction

Weight Abstractions

Author: mikepapadim

…g/beehive/gpullama3/core/types/Pair.java …rg/beehive/gpullama3/auxiliary/Pair.javasrc/main/java/org/beehive/gpullama3/core/types/Pair.java renamed to src/main/java/org/beehive/gpullama3/auxiliary/Pair.java src/main/java/org/beehive/gpullama3/core/types/Pair.java renamed to src/main/java/org/beehive/gpullama3/auxiliary/Pair.java

@@ -1,4 +1,4 @@
-package org.beehive.gpullama3.core.types;
+package org.beehive.gpullama3.auxiliary;
 
 public record Pair<First, Second>(First first, Second second) {
 }

src/main/java/org/beehive/gpullama3/core/model/tensor/Q8_0QuantizedTensor.java

@@ -1,7 +1,7 @@
 package org.beehive.gpullama3.inference;
 
 import org.beehive.gpullama3.auxiliary.Parallel;
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import org.beehive.gpullama3.inference.state.Phi3State;
 import org.beehive.gpullama3.inference.state.State;
 import org.beehive.gpullama3.inference.weights.standard.Phi3StandardWeights;
@@ -583,7 +583,7 @@ public static FloatArray forwardTornadoVM(Model model, State state, int token, i
         final Configuration configuration = model.configuration();
         final TornadoWeights weights = (TornadoWeights) model.weights();
 
-        MemorySegment.copy(weights.getTokenEmbeddingTable().getSegment(), (long) token * configuration.dim() * Float.BYTES, state.wrapX.getSegment(), 0, configuration.dim() * Float.BYTES);
+        MemorySegment.copy(weights.getTokenEmbeddingTable().asFloatArray().getSegment(), (long) token * configuration.dim() * Float.BYTES, state.wrapX.getSegment(), 0, configuration.dim() * Float.BYTES);
 
         return tornadoVMMasterPlan.tornadoVMForwardExecuteLayered(position);
     }

src/main/java/org/beehive/gpullama3/inference/InferenceCore.java

@@ -1,6 +1,6 @@
 package org.beehive.gpullama3.inference.operation;
 
-import org.beehive.gpullama3.core.types.Pair;
+import org.beehive.gpullama3.auxiliary.Pair;
 
 public final class RoPE {
     public static Pair<float[], float[]> precomputeFreqsCis(int contextLength, int headSize, double theta,

src/main/java/org/beehive/gpullama3/inference/operation/RoPE.java

@@ -1,6 +1,6 @@
 package org.beehive.gpullama3.inference.sampler;
 
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;
 
 import java.util.random.RandomGenerator;

src/main/java/org/beehive/gpullama3/inference/sampler/CategoricalSampler.java

@@ -1,7 +1,7 @@
 package org.beehive.gpullama3.inference.sampler;
 
 import org.beehive.gpullama3.Options;
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import org.beehive.gpullama3.model.Model;
 import org.beehive.gpullama3.tornadovm.utils.FloatArrayUtils;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/sampler/Sampler.java

@@ -1,6 +1,6 @@
 package org.beehive.gpullama3.inference.sampler;
 
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;
 
 import java.util.Comparator;

src/main/java/org/beehive/gpullama3/inference/sampler/ToppSampler.java

@@ -1,7 +1,7 @@
 package org.beehive.gpullama3.inference.state;
 
-import org.beehive.gpullama3.core.model.tensor.ArrayFloatTensor;
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.ArrayFloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import org.beehive.gpullama3.model.Configuration;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;
 import uk.ac.manchester.tornado.api.types.arrays.IntArray;

src/main/java/org/beehive/gpullama3/inference/state/LlamaState.java

@@ -1,7 +1,7 @@
 package org.beehive.gpullama3.inference.state;
 
-import org.beehive.gpullama3.core.model.tensor.ArrayFloatTensor;
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.ArrayFloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import org.beehive.gpullama3.model.Configuration;
 import org.beehive.gpullama3.model.phi3.Phi3Configuration;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

src/main/java/org/beehive/gpullama3/inference/state/Phi3State.java

@@ -1,7 +1,7 @@
 package org.beehive.gpullama3.inference.state;
 
-import org.beehive.gpullama3.core.model.tensor.ArrayFloatTensor;
-import org.beehive.gpullama3.core.model.tensor.FloatTensor;
+import org.beehive.gpullama3.tensor.standard.ArrayFloatTensor;
+import org.beehive.gpullama3.tensor.standard.FloatTensor;
 import org.beehive.gpullama3.model.Configuration;
 import org.beehive.gpullama3.model.qwen2.Qwen2Configuration;
 import uk.ac.manchester.tornado.api.types.arrays.FloatArray;