AmpereComputingAI
diff --git a/‎benchmarks/README.md‎
Lines changed: 2 additions & 2 deletions b/‎benchmarks/README.md‎
Lines changed: 2 additions & 2 deletions
diff --git a/‎benchmarks/example_results/ampereone/ampere_optimized/Meta-Llama-3-8B-Instruct.Q4_K_4.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/ampere_optimized_llama.cpp_r1.2.6/Meta-Llama-3-8B-Instruct.Q4_K_4.gguf@PP128@TG256.csv‎ b/‎benchmarks/example_results/ampereone/ampere_optimized/Meta-Llama-3-8B-Instruct.Q4_K_4.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/ampere_optimized_llama.cpp_r1.2.6/Meta-Llama-3-8B-Instruct.Q4_K_4.gguf@PP128@TG256.csv‎
diff --git a/‎benchmarks/example_results/ampereone/ampere_optimized/Meta-Llama-3-8B-Instruct.Q8R16.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/ampere_optimized_llama.cpp_r1.2.6/Meta-Llama-3-8B-Instruct.Q8R16.gguf@PP128@TG256.csv‎ b/‎benchmarks/example_results/ampereone/ampere_optimized/Meta-Llama-3-8B-Instruct.Q8R16.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/ampere_optimized_llama.cpp_r1.2.6/Meta-Llama-3-8B-Instruct.Q8R16.gguf@PP128@TG256.csv‎
diff --git a/‎benchmarks/example_results/ampereone/unoptimized/Meta-Llama-3-8B-Instruct.Q4_K_M.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/llama.cpp_b3615/Meta-Llama-3-8B-Instruct.Q4_K_M.gguf@PP128@TG256.csv‎ b/‎benchmarks/example_results/ampereone/unoptimized/Meta-Llama-3-8B-Instruct.Q4_K_M.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/llama.cpp_b3615/Meta-Llama-3-8B-Instruct.Q4_K_M.gguf@PP128@TG256.csv‎
diff --git a/‎benchmarks/example_results/ampereone/unoptimized/Meta-Llama-3-8B-Instruct.Q8_0.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/llama.cpp_b3615/Meta-Llama-3-8B-Instruct.Q8_0.gguf@PP128@TG256.csv‎ b/‎benchmarks/example_results/ampereone/unoptimized/Meta-Llama-3-8B-Instruct.Q8_0.gguf@PP128@TG256.csv‎ renamed to ‎benchmarks/example_results/ampereone/llama.cpp_b3615/Meta-Llama-3-8B-Instruct.Q8_0.gguf@PP128@TG256.csv‎
@@ -38,7 +38,7 @@ bash download_models.sh  # uncomment preferred models in the file, by default ll
 bash run.sh  # modify to adjust number of threads available and other parameters
 ```
 
-## Speedup vs unoptimized llama.cpp
+## Speedup Ampere optimized llama.cpp (r1.2.6) vs llama.cpp (b3615)
 
 |                       | Q4_K_M  | Q4_K_4  | Speedup  |
 |-----------------------|---|---|---|
@@ -48,4 +48,4 @@ bash run.sh  # modify to adjust number of threads available and other parameters
 |                       | Q8_0  | Q8R16  | Speedup  |
 |-----------------------|---|---|---|
 |token generation (t/s) | 248.17  | 313.78  | 1.26x  |
-|prompt processing (t/s)| 410.90  | 950.64  | 2.31x  |
+|prompt processing (t/s)| 410.90  | 950.64  | 2.31x  |