Improve accuracy of vecdot routines

cafaxo · Jul 18, 2024 · 587d270 · 587d270
1 parent 3d07db9
commit 587d270
Show file tree

Hide file tree

Showing 9 changed files with 409 additions and 868 deletions.
diff --git a/src/Llama2.jl b/src/Llama2.jl
@@ -14,12 +14,7 @@ export load_gguf_model, load_karpathy_model, encode, sample
 export train
 
 # quantization
-include("quantization/utils.jl")
-include("quantization/common.jl")
-include("quantization/q4.jl")
-include("quantization/q6.jl")
-include("quantization/q8.jl")
-include("quantization/vecdot.jl")
+include("kquants.jl")
 
 # inference
 include("tokenizer.jl")