feat: update compose submodule for Hermes 4.3 llama-cpp service #72

Open
Hermes wants to merge 1 commits from feat/hermes43-llamacpp into master
Collaborator

Summary

Updates the assets/compose submodule to include the new llama-cpp-hermes service for running Hermes 4.3 on dual MI50 GPUs.

What's included

  • New compose service llama-cpp-hermes: Custom llama.cpp (b9596) built with ROCm 6.1 + gfx906. Serves Hermes 4.3 at ~19 t/s with 160K context on both MI50s.
  • Updated ollama service: Stripped of GPU access — runs CPU-only for bge-m3 embeddings.
  • New Dockerfile: assets/compose/ai/llama-cpp/Dockerfile for the custom ROCm build.

PRs

## Summary Updates the `assets/compose` submodule to include the new `llama-cpp-hermes` service for running Hermes 4.3 on dual MI50 GPUs. ## What's included - **New compose service `llama-cpp-hermes`**: Custom llama.cpp (b9596) built with ROCm 6.1 + gfx906. Serves Hermes 4.3 at ~19 t/s with 160K context on both MI50s. - **Updated `ollama` service**: Stripped of GPU access — runs CPU-only for bge-m3 embeddings. - **New Dockerfile**: `assets/compose/ai/llama-cpp/Dockerfile` for the custom ROCm build. ## PRs - Compose changes: https://code.lazyworkhorse.net/gortium/compose/pulls/54
Hermes added 1 commit 2026-06-11 15:47:12 +00:00
This pull request has changes conflicting with the target branch.
  • assets/compose
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u origin feat/hermes43-llamacpp:feat/hermes43-llamacpp
git checkout feat/hermes43-llamacpp
Sign in to join this conversation.
No description provided.