tjake / Jlama

Jlama is a modern LLM inference engine for Java
Apache License 2.0
499 stars 48 forks source link

Switch all tensor operations to use longs #33

Closed tjake closed 2 months ago

tjake commented 3 months ago
Caused by: java.lang.IllegalArgumentException: Out of range: 3145728000                                                                                                         
        at com.google.common.base.Preconditions.checkArgument(Preconditions.java:203)                                                                                           
        at com.google.common.primitives.Ints.checkedCast(Ints.java:88)                                                                                                          
        at com.github.tjake.jlama.tensor.FloatBufferTensor.<init>(FloatBufferTensor.java:88)                                                                                    
        at com.github.tjake.jlama.safetensors.Weights.load(Weights.java:128)                                                                                                    
        at com.github.tjake.jlama.safetensors.WeightLoader.load(WeightLoader.java:30)                                                                                           
        at com.github.tjake.jlama.safetensors.SafeTensorIndex.load(SafeTensorIndex.java:189)                                                                                    
        at com.github.tjake.jlama.model.gemma.GemmaModel.loadInputWeights(GemmaModel.java:114)                                                                                  
        at com.github.tjake.jlama.model.AbstractModel.<init>(AbstractModel.java:134)                                                                                            
        at com.github.tjake.jlama.model.llama.LlamaModel.<init>(LlamaModel.java:55)                                                                                             
        at com.github.tjake.jlama.model.gemma.GemmaModel.<init>(GemmaModel.java:58)