Sync all three vLLM model configs from club-3090 master (ae4846f).
Update to Genesis v7.65 full PROD env set with new patches.
Update docker image to nightly-7a1eb8ac. Add torch_compile and
triton cache dirs. Add agent setup guide (AGENTS.md).
Add 'evan' API key to llama-swap sops secrets.
Allow one CUDA0 and one CUDA1 model to run simultaneously. Dual-GPU
models (using -ts splits) are excluded from the matrix so they evict
everything when loaded. vLLM docker models get evict_cost=50 to
discourage eviction due to slow cold starts.
Extract a shared hasType helper for model filtering and add
vision (text + image) input capability to compatible models.
Also tag two llama-swap models with the vision type.
- Upgrade llama-cpp from version 8196 to 8229
- Remove GPT OSS CSEC (20B) - Thinking model config
- Remove Qwen3 Next (80B) - Instruct model config
- Increment -ncmoe parameter for Qwen3 Coder Next (80B) model
- Bump llama-cpp from version 8157 to 8196
- Bump llama-swap from version 192 to 197
- Switch default assistant model from qwen3-coder-next-80b to qwen3.5-27b-thinking
- Remove glm-4-32b-instruct model configuration
- Update qwen3.5-27b-thinking config:
- Use bartowski quantization (IQ4_XS) instead of unsloth
- Increase context window from 131k to 196k
- Add cache type settings (q8_0) and CUDA device
- Add 1password-cli to home-manager programs
- Fix typo: 'dispay' -> 'display' in llm-config.lua
- Add Qwen3.5-35B-A3B and Qwen3.5-27B thinking model configs
- Remove deprecated Qwen2.5-Coder-7B model
- Update synthetic models list with new HF endpoints
- Update nvim to use qwen3-coder-next-80b-instruct model
- Add AGENTS.md with AI agent best practices for timeout and file writing
- Update pi config to include agent guidelines
- Refactor llama-swap: remove old models, update quantizations, add tensor splits,
remove GGML_CUDA_ENABLE_UNIFIED_MEMORY flags, and simplify configuration
Change opencode and pi model filtering to use 'coding' type instead of
more generic 'text-generation' type. Update llama-swap model configs to
include 'coding' in metadata type list for relevant models (deepseek-coder,
qwen-coder, mistral, codellama, llama3-8b-instruct-q5).
Add new model "Qwen3 Coder Next (80B) - Instruct" with 262144 context window
and optimized parameters for coding tasks. Uses CUDA unified memory support.
- Update llama-cpp from b7867 to b7898
- Update opencode from v1.1.12 to v1.1.48 with improved build process:
- Replace custom bundle script with official script/build.ts
- Add shell completion support
- Add version check testing
- Simplify node_modules handling
- Update llama-swap service config with new llama.cpp options
- Clarify opencode agent testing workflow in developer and reviewer configs
- Update llama.cpp from b7789 to b7867
- Update llama-swap from v182 to v186
- Add OpenCode conventional commit command configuration
- Add moonshotai Kimi-K2.5 model to llama-swap
