docs.facundoitest.space

User Tools

Site Tools

Trace: • habilitar_explorar_admin_desde_una_red_distinta
llama.cpp_opencl_rx_6600_xt_on_ubuntu_24.04

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

llama.cpp_opencl_rx_6600_xt_on_ubuntu_24.04 [2025/06/29 01:37] – created osollama.cpp_opencl_rx_6600_xt_on_ubuntu_24.04 [2025/07/01 13:30] (current) – oso
Line 8: Line 8:
   * **Client**: Web browser or PowerShell Invoke-RestMethod     * **Client**: Web browser or PowerShell Invoke-RestMethod  
  
----+----
  
 === 🧰 Step-by-step Procedure === === 🧰 Step-by-step Procedure ===
  
 +Based on this cmd history:
 +
 +<code bash>
 +    4  sudo apt install rocm-opencl-dev
 +    5  mkdir rocm
 +    6  cd rocm
 +    7  wget https://repo.radeon.com/amdgpu-install/6.3.3/ubuntu/noble/amdgpu-install_6.3.60303-1_all.deb
 +    8  sudo apt install ./amdgpu-install_6.3.60303-1_all.deb
 +   11  sudo reboot
 +   16  cd rocm/
 +   18  sudo apt update
 +   19  sudo apt full-upgrade
 +   20  sudo apt install rocm-opencl-dev
 +   22  sudo apt install clinfo
 +   24  clinfo
 +   31  sudo apt install mesa-opencl-icd clinfo
 +   32  sudo nano /etc/drirc
 +   33  sudo reboot
 +   35  sudo clinfo | grep 'Device Name'
 +   36  mkdir oclLlama
 +   37  cd oclLlama/
 +   38  git clone https://github.com/ggerganov/llama.cpp
 +   39  cd llama.cpp
 +   46  make LLAMA_OPENCL=1
 +   47  sudo apt install cmake build-essential libclblast-dev ocl-icd-opencl-dev
 +   48  mkdir build
 +   49  cd bould
 +   50  cd build/
 +   51  cmake .. -DLLAMA_CLBLAST=on
 +   52  sudo apt install curl
 +   53  sudo apt install libcurl4-openssl-dev
 +   54  cmake .. -DLLAMA_CLBLAST=on
 +   55  cmake --build . --config Release -j$(nproc)
 +   59  ls -lh /home/facundo/oclLlama/llama.cpp/models/*.gguf
 +   60  cd ~/oclLlama/llama.cpp/models
 +   68  ls *.gguf
 +   69  ./bin/llama-server   --model /home/facundo/oclLlama/llama.cpp/models/phi-2.Q4_K_M.gguf   --host 0.0.0.0   --port 11434   --n-gpu-layers 100
 +   70  ls ..
 +   71  cd ..
 +   72  ls
 +   73  cd build/
 +
 +here I uploaded the .gguf from my laptop
 +
 +   74  ./bin/llama-server   --model /home/facundo/oclLlama/llama.cpp/models/phi-2.Q4_K_M.gguf   --host 0.0.0.0   --port 11434   --n-gpu-layers 100
 +
 +</code>
 +
 +Gepetto said:
 <code bash> <code bash>
 # 1. Optional: ROCm was tried (not strictly needed with Rusticl) # 1. Optional: ROCm was tried (not strictly needed with Rusticl)
Line 41: Line 90:
 # 7. Download or upload a model # 7. Download or upload a model
 mkdir -p ~/oclLlama/llama.cpp/models mkdir -p ~/oclLlama/llama.cpp/models
-# (Copy from your laptop or download a .gguf file)+# (SCP from your laptop or download a .gguf file to /models/ dir)
  
 # 8. Run the API server # 8. Run the API server
Line 51: Line 100:
 </code> </code>
  
----+----
  
 === 🌍 Test Access from Browser === === 🌍 Test Access from Browser ===
Line 57: Line 106:
   * ✅ Built-in Web UI opens   * ✅ Built-in Web UI opens
  
----+----
  
 === 🧪 Test from PowerShell === === 🧪 Test from PowerShell ===
Line 72: Line 121:
 </code> </code>
  
----+----
  
 === ✅ Confirm GPU Usage === === ✅ Confirm GPU Usage ===
Line 82: Line 131:
 </code> </code>
  
----+----
  
 === 🧹 Optional Cleanup === === 🧹 Optional Cleanup ===
llama.cpp_opencl_rx_6600_xt_on_ubuntu_24.04.1751161042.txt.gz · Last modified: 2025/06/29 01:37 by oso