Skip to content

Instantly share code, notes, and snippets.

View ugovaretto's full-sized avatar

Ugo Varetto ugovaretto

View GitHub Profile
@ugovaretto
ugovaretto / extract-links.py
Created December 2, 2025 11:01
Headless Selenium for webscraping
@ugovaretto
ugovaretto / config.yml
Created November 14, 2025 13:33
llama-swap configuration
# llama-swap configuration
models:
VibeThinker-1.5B:
cmd: llama-server --port ${PORT} -c 0 --model /home/ugo/.cache/llama.cpp/VibeThinker-1.5B.f16.gguf -ngl 99
Aquif-3.5-Max-42B-A3B:
cmd: >
llama-server --port ${PORT}
--model /home/ugo/.cache/llama.cpp/unsloth-aquif-3.5-Max-42B-A3B-GGUF/aquif-3.5-Max-42B-A3B-UD-Q6_K_XL.gguf -ngl 99 -fa on
Aquif-3.5-Max-42B-A3B-Coding-Q6_K_XL-KVQ8:
cmd: >
@ugovaretto
ugovaretto / hf-dload-llamacpp.sh
Created November 9, 2025 07:45
download from huggingface into llama.cpp directory
#!/usr/bin/env bash
hf download $1 --local-dir "$HOME/.cache/llama.cpp/$1/$2" --include="*$2.gguf"
#!/usr/bin/env bash
# $1: model file name
# $2: port
# $3: context size
# $4: alias, (model name sent to client)
# Define the help text as a function
show_help() {
echo "Usage: $0 <model file> <port> \\"
echo " <context length, 0 for default> \\"
@ugovaretto
ugovaretto / llama.cpp.ccmake.txt
Last active November 2, 2025 07:29
llama.cpp vulkan BLAS
GGML_BLAS_VENDOR Intel10_64_dyn
Vulkan_GLSLANG_VALIDATOR_EXECU ~/.local/vulkan/1.4.321.1/x86_64/bin/glslangValidator
Vulkan_GLSLC_EXECUTABLE ~/.local/vulkan/1.4.321.1/x86_64/bin/glslc
Vulkan_INCLUDE_DIR ~/.local/vulkan/1.4.321.1/x86_64/include
Vulkan_LIBRARY ~/.local/vulkan/1.4.321.1/x86_64/lib/libvulkan.so
@ugovaretto
ugovaretto / llama-cpp-multinode.md
Last active November 2, 2025 03:48
Run llama.cpp on multiple nodes

llama.cpp on multiple nodes

After compiling with RPC enabled (GGML_RPC CMake parameter):

Run rpc-server on the remote nodes:

rpc-server --port 5001 --host 169.254.51.65
@ugovaretto
ugovaretto / grub
Created August 23, 2025 13:40
Kernel parameters to use all the GPU memory on Strix Halo
#/etc/default/grub
#run sudo grub2-mkconfig -o /boot/grub2/grub.cfg after changing the file
GRUB_TIMEOUT=5
GRUB_DISTRIBUTOR="$(sed 's, release .*$,,g' /etc/system-release)"
GRUB_DEFAULT=saved
GRUB_DISABLE_SUBMENU=true
GRUB_TERMINAL_OUTPUT="console"
##### vvvvvvvvvvvvvvv
GRUB_CMDLINE_LINUX="rhgb quiet amd_iommu=off amdgpu.gttsize=131072 ttm.pages_limit=33554432"
##### ^^^^^^^^^^^^^^^
@ugovaretto
ugovaretto / ollama-gfx1151.diff
Last active August 19, 2025 07:39
Fix to make ollama work with gfx1151 with 128GB GTT Memory
diff --git a/discover/amd_linux.go b/discover/amd_linux.go
index ebffbdf6..bbb5ec49 100644
--- a/discover/amd_linux.go
+++ b/discover/amd_linux.go
@@ -315,15 +315,15 @@ func AMDGetGPUInfo() ([]RocmGPUInfo, error) {
gpuOrdinalID += 1
// iGPU detection, remove this check once we can support an iGPU variant of the rocm library
- if totalMemory < IGPUMemLimit {
- reason := "unsupported Radeon iGPU detected skipping"
@ugovaretto
ugovaretto / link-bitcode.sh
Created August 14, 2025 05:29
link bitcode files
@ugovaretto
ugovaretto / link-all.sh
Created July 24, 2025 10:17
Link all files in a target path to files in the current directory