Update README.md

README.md

@@ -29,7 +29,7 @@ with settings `{
   temperature: 0.8
 }`
 
-outputs:
+outputs (using unquantized gguf):
 
 <pre>
 ```python
@@ -51,6 +51,8 @@ print(result)
 The area of the circle is $\boxed{\frac{27\pi}{4}}$ square cm.
 </pre>
 
+??? It should have been `9*pi/4`. Am I using this model wrong? Same result with temperature=0.0,top_k=1.
+
 According to their [paper on arXiv](https://arxiv.org/abs/2404.07965), rho-math-7b-v0.1 is a continued pretraining on Mistral-7B, while their 1B model is a continued pretaining on TinyLlama-1.1B.
 
 # imatrix
@@ -63,6 +65,4 @@ which took 1665 seconds (28 minutes) on my GTX 1660 Super and used only 1 thread
 
 # quantize
 
-Quantized with llama.cpp b2661 (2024-04-12), compiled with `LLAMA_CUDA_FORCE_MMQ=1` (full cmd: `make -j6 LLAMA_CUDA_FORCE_MMQ=1 LLAMA_CUDA=1 LLAMA_FAST=1 LLAMA_OPENBLAS=1 LLAMA_BLAS_VENDOR=OpenBLAS`) for a big speed up (GTX 1660 Super doesn't have tensor cores, so it's better to use MMQ than nothing).
-
-IQ3_XS (3 018 815 264 bytes) is stupid, it thinks radius=diameter, so I didn't upload it or lower quants.
+Quantized with llama.cpp b2661 (2024-04-12), compiled with `LLAMA_CUDA_FORCE_MMQ=1` (full cmd: `make -j6 LLAMA_CUDA_FORCE_MMQ=1 LLAMA_CUDA=1 LLAMA_FAST=1 LLAMA_OPENBLAS=1 LLAMA_BLAS_VENDOR=OpenBLAS`) for a big speed up (GTX 1660 Super doesn't have tensor cores, so it's better to use MMQ than nothing).