Can't install GPU version for windows for many times.

## Issues
I am trying to install the lastest version of llama-cpp-python in my windows 11 with RTX-3090ti(24G). I have successfully installed llama-cpp-python=0.1.87 (can't exactly remember) months ago while using:
``` conda prompt
set FORCE_CMAKE=1
set CMAKE_ARGS=-DLLAMA_CUBLAS=on
pip install --upgrade --force-reinstall llama-cpp-python --no-cache-dir
```
But when I want to access the latest version recently by using:
``` conda prompt
set CMAKE_ARGS="-DLLAMA_CUDA=on"
pip install --upgrade --force-reinstall llama-cpp-python --no-cache-dir
```
After loading the model, it is still using CPU with `BLAS=0` (or is another params = 1 instead of `BLAS` in new version?).
```
llama_new_context_with_model: n_ctx = 2048
llama_new_context_with_model: n_batch = 1024
llama_new_context_with_model: n_ubatch = 512
llama_new_context_with_model: freq_base = 500000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: CPU KV buffer size = 256.00 MiB
llama_new_context_with_model: KV self size = 256.00 MiB, K (f16): 128.00 MiB, V (f16): 128.00 MiB
llama_new_context_with_model: CPU output buffer size = 0.49 MiB
llama_new_context_with_model: CPU compute buffer size = 258.50 MiB
llama_new_context_with_model: graph nodes = 1030
llama_new_context_with_model: graph splits = 1
AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 | LAMMAFILE = 1 |
Model metadata: {'general.name': 'Meta-Llama-3-8B-Instruct-imatrix', 'general.architecture': 'llama', 'llama.block_count': '32', 'llama.context_length': '8192', 'tokenizer.ggml.eos_token_id': '128001', 'general.file_type': '18', 'llama.attention.head_count_kv': '8', 'llama.embedding_length': '4096', 'llama.feed_forward_length': '14336', 'llama.attention.head_count': '32', 'llama.rope.freq_base': '500000.000000', 'llama.attention.layer_norm_rms_epsilon': '0.000010', 'llama.vocab_size': '128256', 'llama.rope.dimension_count': '128', 'tokenizer.ggml.model': 'gpt2', 'general.quantization_version': '2', 'tokenizer.ggml.bos_token_id': '128000', 'tokenizer.chat_template': "{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}"}
Using gguf chat template: {% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>
```
I have been trying the pre-build wheel for CUDA 12.1 (`pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121`) and it still doesn't work. I add `--verbose` to see the output:

```
loading initial cache file C:\Users\Administrator\AppData\Local\Temp\tmpbbvy3nqu\build\CMakeInit.txt
-- Building for: Visual Studio 17 2022
-- Selecting Windows SDK version 10.0.22621.0 to target Windows 10.0.22631.
-- The C compiler identification is MSVC 19.39.33523.0
-- The CXX compiler identification is MSVC 19.39.33523.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: F:/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.39.33519/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: F:/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.39.33519/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Git: F:/Git/cmd/git.exe (found version "2.44.0.windows.1")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - not found
-- Found Threads: TRUE
-- Warning: ccache not found - consider installing it for faster compilation or disable this warning with LLAMA_CCACHE=OFF
-- CMAKE_SYSTEM_PROCESSOR: AMD64
-- CMAKE_GENERATOR_PLATFORM: x64
-- x86 detected
-- Performing Test HAS_AVX_1
-- Performing Test HAS_AVX_1 - Success
-- Performing Test HAS_AVX2_1
-- Performing Test HAS_AVX2_1 - Success
-- Performing Test HAS_FMA_1
-- Performing Test HAS_FMA_1 - Success
-- Performing Test HAS_AVX512_1
-- Performing Test HAS_AVX512_1 - Failed
-- Performing Test HAS_AVX512_2
-- Performing Test HAS_AVX512_2 - Failed

```
## Environment
python=3.12
C++ compiler: viusal studio 2022 (with necessary C++ modules)
`cmake --version` = 3.29.2
`nvcc -V` = CUDA 12.1 (while `nvidia-smi` cuda version is 12.3, i think it is not related to this issues)

I have been download and install VS2022, CUDA toolkit, cmake and anaconda, I am wondering if some steps are missing. Considering my previous experience there is no need to git clone this code and cd into it to build (Though I did that on my mac to build a pth file to bin file months ago).

My system variables are listed below:
- F:\Anaconda\Scripts
- F:\CMake\bin
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin

## Questions
1. Are there some steps i have been missing for build the llama-cpp for GPU version?
2. How to know if it is built for GPU version when i run `pip install llama-cpp-python` instead of loading model to check `BLAS=1.`
3. Do I need to git clone this code and cd into some dir, create some file or dir to run the `pip install llama-cpp-py`?