Evan Reichard
6fff658f9d
Sync all three Qwen3.6 27B vLLM configs (tools-text, long-text, long-vision) with club-3090 83bf73d. Adds disable-thinking flag and introduces upstream hash tracking comments for future syncs. Update update-vllm-3090-configs skill to use hash-based skip logic.
72 lines
3.3 KiB
Markdown
72 lines
3.3 KiB
Markdown
---
|
|
name: update-vllm-3090-configs
|
|
description: Update only the qwen3.6-27b vLLM 3090 llama-swap configs from club-3090 refs; compare diffs, present a plan, and require approval before editing.
|
|
---
|
|
|
|
# Update vLLM 3090 Configs
|
|
|
|
## Scope
|
|
|
|
Use only for Qwen3.6 27B vLLM 3090 configs in `modules/nixos/services/llama-swap/`.
|
|
Do not use this skill for other models, other Qwen sizes, non-vLLM configs, or package bumps.
|
|
|
|
Local files:
|
|
- `modules/nixos/services/llama-swap/config.nix`
|
|
- `modules/nixos/services/llama-swap/setup-qwen36-vllm.sh`
|
|
|
|
Local config keys:
|
|
- `vllm-qwen3.6-27b-tools-text`
|
|
- `vllm-qwen3.6-27b-long-text`
|
|
- `vllm-qwen3.6-27b-long-vision`
|
|
|
|
## Hash Tracking
|
|
|
|
Each config entry stores an upstream commit hash comment:
|
|
`# Upstream: club-3090 <hash> (<date>) - <compose-file>`
|
|
|
|
When comparing, first extract stored hashes. If a config's hash matches
|
|
upstream HEAD, skip it (report "already synced"). Only full-diff configs
|
|
whose hash differs. Update the hash comment when edits are applied.
|
|
|
|
## Upstream References
|
|
|
|
Compare against `club-3090` master:
|
|
- `models/qwen3.6-27b/vllm/compose/single/tools-text.yml`
|
|
- `models/qwen3.6-27b/vllm/compose/single/long-text.yml`
|
|
- `models/qwen3.6-27b/vllm/compose/single/long-vision.yml`
|
|
- `scripts/setup.sh` for the current `GENESIS_PIN="${GENESIS_PIN:-...}"`
|
|
|
|
Use raw URLs or a temp clone under `_scratch/club-3090`. Prefer a temp clone when checking broad changes:
|
|
|
|
```bash
|
|
mkdir -p _scratch
|
|
git clone https://github.com/noonghunna/club-3090 _scratch/club-3090 2>/dev/null || git -C _scratch/club-3090 pull --ff-only
|
|
```
|
|
|
|
## Required Workflow
|
|
|
|
1. Fetch/update upstream refs under `_scratch/club-3090` or fetch the raw files.
|
|
2. Extract stored upstream hashes from `# Upstream: club-3090 ...` comments in config.nix. Skip any config whose hash matches upstream HEAD (report "already synced").
|
|
3. Compare upstream compose files to the remaining local llama-swap entries. Translate docker-compose semantics into the existing `docker run`/llama-swap format.
|
|
4. Compare upstream `scripts/setup.sh` Genesis pin to local `GENESIS_PIN` in `setup-qwen36-vllm.sh`.
|
|
5. Check upstream compose volumes/entrypoint for sidecar patches. If patches are added, removed, renamed, or invoked differently, update both:
|
|
- runtime mounts and `python3 /patches/...` calls in `config.nix`
|
|
- download/install logic and summary in `setup-qwen36-vllm.sh`
|
|
6. Ignore these diffs unless the user explicitly asks otherwise:
|
|
- `shm_size` / shm-related compose settings
|
|
- local timing patch `patch_timings_07351e088.py` and its mount/invocation
|
|
- model served-name differences caused by llama-swap `${MODEL_ID}`
|
|
- `HUGGING_FACE_HUB_TOKEN`; keep local CUDA device/env choices
|
|
- upstream relative paths vs local `/mnt/ssd/vLLM/...` paths
|
|
- docker-compose format vs local llama-swap/Nix format
|
|
7. Before editing, present:
|
|
- upstream files/commit checked
|
|
- meaningful diffs found
|
|
- ignored diffs
|
|
- exact planned local changes
|
|
Then wait for explicit user approval.
|
|
8. After approval, edit minimally and update the `# Upstream: club-3090 ...` hash comments. Validate:
|
|
- `bash -n modules/nixos/services/llama-swap/setup-qwen36-vllm.sh`
|
|
- `nix-instantiate --parse modules/nixos/services/llama-swap/config.nix`
|
|
9. Summarize changed files and any remaining upstream differences.
|