Skip to content

fix: media server bug and add python-multipart dep#165

Merged
TimPietruskyRunPod merged 5 commits into
mainfrom
t3code/e5a74185
Apr 29, 2026
Merged

fix: media server bug and add python-multipart dep#165
TimPietruskyRunPod merged 5 commits into
mainfrom
t3code/e5a74185

Conversation

@TimPietruskyRunPod

Copy link
Copy Markdown
Collaborator

Summary

  • Fix NameError: name 'engine_id' is not defined in media server plugin error handler (engine_idplugin_id)
  • Add python-multipart to Dockerfile.unified — required by FastAPI for file upload endpoints (STT transcription)

Test plan

  • TTS verified: 137KB WAV generated in ~312ms from "Hello world" input
  • STT verified: Transcription returned from WAV input
  • Audio + LLM fit together on RTX 5090 (26.5GB / 32.6GB VRAM)
  • Full health check passes: LLM + Hermes + Audio all green

Remove `--reasoning-format none` from all 13 GGUF model configs (Qwen 3.5,
GLM 4.7, GPT-OSS) so llama.cpp's default auto mode handles reasoning +
tool calling correctly — fixes OpenClaw agentic tool calling which was
broken by raw `<think>` tags and XML tool calls leaking into content.

Also:
- Add `oc_sync_openclaw_runtime` to entrypoint-common.sh — writes the
  resolved model provider, context window, and gateway config into
  openclaw.json so OpenClaw picks up the local llama-server automatically
- Run the sync twice in entrypoint-unified.sh (before and after
  `openclaw doctor --fix`) to survive doctor's rewrite
- Fix LD_LIBRARY_PATH to append existing paths instead of replacing them
- Add controlUi (allowedOrigins + dangerouslyDisableDeviceAuth) to Go
  config, TypeScript config, and validation schema
Hermes rejects short or common placeholder API keys (e.g. "test123",
"changeme") when binding to 0.0.0.0. The entrypoint now detects weak
tokens and auto-generates a 32-byte hex key via openssl, displaying it
in the ready banner so users can still authenticate.
…binary

llama-liquid-audio-server was removed in llama.cpp b8967 — audio support
is now built into llama-server via the mtmd library. The plugin just needs
to point at llama-server with the same CLI args (--mmproj, -mv, --tts-speaker-file).
Fix NameError in media server error handler (engine_id → plugin_id).
Add python-multipart to Dockerfile.unified — required by FastAPI for
file upload endpoints (STT). Audio TTS+STT verified working on RTX 5090
with Qwen 3.6 27B + LFM2.5 Audio (26.5GB / 32.6GB VRAM).
@vercel

vercel Bot commented Apr 29, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
a2go Ready Ready Preview, Comment Apr 29, 2026 5:37pm

Request Review

@TimPietruskyRunPod TimPietruskyRunPod merged commit f9d2fec into main Apr 29, 2026
3 of 4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant