Datasets:
License:
Upload 2 files
Browse files- FNORD.txt +118 -0
- llama.cpp_24AlgMor24_AMD_bin_Linux.zip +3 -0
FNORD.txt
ADDED
|
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
|
| 2 |
+
|
| 3 |
+
Clonar git
|
| 4 |
+
|
| 5 |
+
#PONTEIROHORA
|
| 6 |
+
|
| 7 |
+
find . -exec touch {} +
|
| 8 |
+
|
| 9 |
+
Para ter o controle total que você busca (dominar a operação):
|
| 10 |
+
|
| 11 |
+
O ideal é clonar o repositório oficial: git clone https://github.com/ggml-org/llama.cpp.
|
| 12 |
+
|
| 13 |
+
https://github.com/ggml-org/llama.cpp
|
| 14 |
+
|
| 15 |
+
Substituir/Modificar os arquivos .cpp e .h com a sua lógica da OFFELLIA.
|
| 16 |
+
|
| 17 |
+
Compilar manualmente usando make ou cmake.
|
| 18 |
+
|
| 19 |
+
cd llama.cpp
|
| 20 |
+
|
| 21 |
+
cmake -B build -DGGML_VULKAN=ON && cmake --build build -j
|
| 22 |
+
|
| 23 |
+
echo "/home/userk21/llama.cpp_ggml_zeta_helicoidal/build/bin" | sudo tee /etc/ld.so.conf.d/llama-custom.conf
|
| 24 |
+
|
| 25 |
+
sudo ldconfig
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
wget -O vendor/miniaudio/miniaudio.h https://raw.githubusercontent.com/mackron/miniaudio/master/miniaudio.h
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
#######################
|
| 34 |
+
|
| 35 |
+
Adapte a sequẽncia de comandos corretos para compilar no kali linux para este comando do meu Ryzen que ja utilizo: cmake -B build -DGGML_VULKAN=ON && cmake --build build -j
|
| 36 |
+
Gemini 3.6 Flash Ran for 13s
|
| 37 |
+
lightbulb Thought for 4 seconds
|
| 38 |
+
build Built
|
| 39 |
+
lightbulb Thought for 8 seconds
|
| 40 |
+
|
| 41 |
+
Aqui está a solução completa e a sequência de comandos para o Kali Linux no seu processador AMD Ryzen com suporte a Vulkan.
|
| 42 |
+
O que corrigimos no código para eliminar o erro?
|
| 43 |
+
|
| 44 |
+
Símbolo de Ponto Flutuante (ggml_vec_dot_q4_2_h_f32):
|
| 45 |
+
|
| 46 |
+
Implementamos a função nativa ggml_vec_dot_q4_2_h_f32 em ggml-cpu/quants.c e declaramos em ggml-cpu/quants.h.
|
| 47 |
+
|
| 48 |
+
Ajustamos o mapeamento vec_dot da struct do GGML_TYPE_Q4_2_H para utilizar vec_dot_type = GGML_TYPE_F32. Isso impede o link error (referência não definida) no libggml-cpu.so.
|
| 49 |
+
|
| 50 |
+
Interface Svelte UI (ChatAttachmentsList.svelte):
|
| 51 |
+
|
| 52 |
+
Removemos a anotação de tipo inline TypeScript que fazia o compilador Vite/Rollup falhar ao empacotar o ativo Web embedded (tools/ui).
|
| 53 |
+
|
| 54 |
+
Sequência de Comandos no Kali Linux (AMD Ryzen + Vulkan)
|
| 55 |
+
1. Instalar as dependências do Kali / Debian
|
| 56 |
+
code Bash
|
| 57 |
+
|
| 58 |
+
sudo apt update
|
| 59 |
+
sudo apt install -y build-essential cmake ninja-build git pkg-config \
|
| 60 |
+
libvulkan-dev vulkan-tools mesa-vulkan-drivers \
|
| 61 |
+
nodejs npm
|
| 62 |
+
|
| 63 |
+
2. Verificar se o Vulkan reconhece sua GPU AMD (ex: RADV Renoir/RDNA)
|
| 64 |
+
code Bash
|
| 65 |
+
|
| 66 |
+
vulkaninfo --summary
|
| 67 |
+
|
| 68 |
+
3. Limpar o build anterior e configurar o CMake com Vulkan
|
| 69 |
+
code Bash
|
| 70 |
+
|
| 71 |
+
# Limpa artefatos antigos corrompidos
|
| 72 |
+
rm -rf build
|
| 73 |
+
|
| 74 |
+
# Configura o CMake com suporte Vulkan e otimizações nativas do Ryzen (-march=native)
|
| 75 |
+
cmake -B build \
|
| 76 |
+
-DGGML_VULKAN=ON \
|
| 77 |
+
-DGGML_NATIVE=ON \
|
| 78 |
+
-DCMAKE_BUILD_TYPE=Release
|
| 79 |
+
|
| 80 |
+
4. Compilar usando todos os núcleos do seu Ryzen
|
| 81 |
+
code Bash
|
| 82 |
+
|
| 83 |
+
cmake --build build -j$(nproc)
|
| 84 |
+
|
| 85 |
+
5. Registrar os caminhos das bibliotecas compiladas (libggml-*.so)
|
| 86 |
+
code Bash
|
| 87 |
+
|
| 88 |
+
echo "$(pwd)/build/bin" | sudo tee /etc/ld.so.conf.d/llama-custom.conf
|
| 89 |
+
sudo ldconfig
|
| 90 |
+
|
| 91 |
+
Teste de Execução e Quantização (Q4_2_H)
|
| 92 |
+
code Bash
|
| 93 |
+
|
| 94 |
+
# 1. Quantizar para a nova arquitetura Helicoidal Q4_2_H
|
| 95 |
+
./build/bin/llama-quantize \
|
| 96 |
+
"/home/userk21/Área de trabalho/userk21/GGUFS/VӨIDDΣЯ_f16_LFM2-700M-Heretic.gguf" \
|
| 97 |
+
"/home/userk21/Área de trabalho/userk21/GGUFS/VӨIDDΣЯ_Q4_2_H_LFM2-700M-Heretic.gguf" \
|
| 98 |
+
Q4_2_H
|
| 99 |
+
|
| 100 |
+
# 2. Executar o servidor com aceleração Vulkan
|
| 101 |
+
./build/bin/llama-server \
|
| 102 |
+
-m "/home/userk21/Área de trabalho/userk21/GGUFS/Qwen3.6-35B-A3B-DSV4Pro-SFT-GPT56Sol-RL-Agent-LynnStyle-Q4-imatrix.gguf" \
|
| 103 |
+
-c 50000 \
|
| 104 |
+
-ngl 99 --n-cpu-moe 99 -fa 1 \
|
| 105 |
+
-np 1 \
|
| 106 |
+
-fa on \
|
| 107 |
+
--agent \
|
| 108 |
+
--cors-origins \
|
| 109 |
+
--tools \
|
| 110 |
+
--reasoning auto \
|
| 111 |
+
--swa-full \
|
| 112 |
+
--kv-unified \
|
| 113 |
+
--webui-mcp-proxy \
|
| 114 |
+
--threads-http -1 \
|
| 115 |
+
--port 5173 \
|
| 116 |
+
--host 127.0.0.1
|
| 117 |
+
|
| 118 |
+
lightbulb_tips
|
llama.cpp_24AlgMor24_AMD_bin_Linux.zip
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:287d3c7fa038078009e82550bab9b3078dd97d3056cb51bfffe63fee9a2cb212
|
| 3 |
+
size 358791186
|