Fine-tune LLMs on an AMD GPU on Windows via DirectML — no NVIDIA, no ROCm, no dual boot. Proven with a Mistral-7B LoRA fine-tune through Axolotl on a Radeon RX 7700 XT.
-
Updated
Jul 24, 2026 - Shell
#
rocm-alternative
Here are 4 public repositories matching this topic...
Usability benchmark for local LLMs on AMD hardware (RDNA4 + Strix Halo) — agentic throughput, self-repair counts and context-depth decay, all parsed from llama.cpp server logs
-
Updated
Sep 25, 2026 - JavaScript
Run a local LLM on AMD hardware ROCm abandoned. Vulkan + llama.cpp + whisper.cpp on GCN 1st-gen and newer. Tested on a Carrizo iGPU from 2015.
-
Updated
Apr 19, 2026 - Shell
Measured llama.cpp Vulkan tuning for AMD Strix Halo (Ryzen AI MAX+ 395 / gfx1151), plus an eval harness with a catalogue of 13 bugs that each produced a believable wrong number.
windows benchmarking amd vulkan moe quantization unified-memory apple-silicon llama-cpp local-llm llm-inference gguf llm-evaluation agentic-ai tool-calling ryzen-ai strix-halo llm-benchmark gfx1151 rocm-alternative
-
Updated
Sep 19, 2026 - PowerShell
Add this topic to your repo
To associate your repository with the rocm-alternative topic, visit your repo's landing page and select "manage topics."