dirge sandbox setup --image <multi-arch OCI image> fails with:
Error: no layers found in manifest for <owner>/<repo>/<image>:<tag>
Symptom
Running an OpenAI-compatible client (e.g. dirge or any
openai SDK client) against mlx_lm.server serving
LiquidAI/LFM2.5-2.6B-MLX-nvfp4 (or other LFM2 MLX conversions) with a
tools=[...] payload produces this server-side warning, and the model never
actually emits a tool call:
| #!/usr/bin/env -S uv run --script | |
| # /// script | |
| # requires-python = ">=3.10" | |
| # dependencies = [ | |
| # "openai>=1.50", | |
| # "pydantic-ai-slim[mcp,openai]>=0.4", | |
| # "rich", | |
| # ] | |
| # /// | |
| """Minimal repro: MCP Gateway tool calls don't join the AI Gateway trace. |
Affects: portkey-ai==2.3.2 used together with pydantic-ai>=2.9.0 and openai>=2.18.0
Symptom: TypeError raised inside portkey_ai._vendor.openai._utils._json before any network call is made
Reproducer: uv run --no-project reproducer.py (PEP 723 inline deps, no credentials needed)
> bat test_dataset.jsonl
─────┬──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
│ File: test_dataset.jsonl
─────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
1 │ {"custom_id":"request-1","method":"POST","url":"/v1/chat/completions","body":{"model":"gemini-3-flash-preview","messages":[{"role":"user","content":"Where is NYU located
│ ?"}],"max_tokens":2048}}
2 │ {"custom_id":"request-2","method":"POST","url":"/v1/chat/completions","body":{"model":"gemini-3-flash-preview","messages":[{"role":"user","content":"What resources are a
│ vailable for genAI research at nyu?"}],"tools":[{"type":"function","function":{"name":"google_search","description":"Retrieves search results from internet to expand con
Commit d23fe45 added facet matches to 4 storage services. However, the old system used numeric facet IDs while the new refactored system uses semantic facet IDs. Additionally, some old facets no longer exist in the new system.
Based on the old facet-tree.json:
| Old ID | Facet Category | Description |
| # basic version from claude-3.7 | |
| #!/bin/bash | |
| # Define the dump file path | |
| DUMP_FILE="/slurm-dump/f-k8s-dev.cfg" | |
| # Function for error handling | |
| handle_error() { | |
| local exit_code=$1 | |
| local error_message=$2 |
| { | |
| "spec": { | |
| "_meta": { | |
| "version": 3 | |
| }, | |
| "nodes": [ | |
| { | |
| "name": "openpmd-api", | |
| "version": "0.15.1", | |
| "arch": { |