57 lines
2.4 KiB
YAML
57 lines
2.4 KiB
YAML
# see https://github.com/sigoden/aichat/blob/main/config.example.yaml
|
|
keybindings: vi
|
|
editor: nvim
|
|
model: local:Qwen3-Coder-30B-Instruct-UD-Q3_K_XL
|
|
|
|
# Sessions: persist REPL sessions and keep more history before summarizing.
|
|
# The default compress_threshold (4000) summarizes far too early for 24k+ windows.
|
|
save_session: true
|
|
compress_threshold: 16000
|
|
|
|
# REPL prompts show live context usage (needs max_input_tokens, set per model below)
|
|
left_prompt: '{color.green}{?session {session}{?role /}}{role}{color.cyan}{?rag @{rag}}{color.reset}> '
|
|
right_prompt: '{color.purple}{?session {consume_tokens}/{max_input_tokens} }{color.reset}'
|
|
|
|
# NOTE: temperature/top_p are intentionally unset — the LAN server applies tuned
|
|
# per-model sampling via --jinja (e.g. GLM 0.6/0.95); a global value would clobber it.
|
|
|
|
# max_input_tokens = real ctx-size (from the router) minus output headroom.
|
|
clients:
|
|
# LAN access (fast; only reachable on the home network)
|
|
- type: openai-compatible
|
|
name: local
|
|
api_base: http://192.168.0.204:11343/v1
|
|
models: &lan_models
|
|
- name: Qwen3-Coder-30B-Instruct-UD-Q3_K_XL
|
|
max_input_tokens: 30000 # ctx 32768
|
|
- name: Qwen3-Coder-Next-UD-IQ3_XXS
|
|
max_input_tokens: 128000 # ctx 131072
|
|
- name: Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS
|
|
max_input_tokens: 22000 # ctx 24576
|
|
supports_vision: true
|
|
- name: Qwen3.6-35B-A3B-Thinking
|
|
max_input_tokens: 22000 # ctx 24576
|
|
supports_vision: true
|
|
- name: Qwen3.5-9B-UD-Q6_K_XL
|
|
max_input_tokens: 30000 # ctx 32768
|
|
supports_vision: true
|
|
- name: gemma-4-26B-A4B-it-UD-IQ4_XS
|
|
max_input_tokens: 22000 # ctx 24576
|
|
supports_vision: true
|
|
- name: gemma-4-E4B-it-UD-Q8_K_XL
|
|
max_input_tokens: 62000 # ctx 65536
|
|
supports_vision: true
|
|
- name: GLM-4.7-Flash-UD-Q4_K_XL
|
|
max_input_tokens: 22000 # ctx 24576
|
|
- name: gpt-oss-20b
|
|
max_input_tokens: 62000 # ctx 65536
|
|
- name: gpt-oss-20b-low
|
|
max_input_tokens: 62000 # ctx 65536
|
|
# Remote access via duskadiy.com (reachable from anywhere; requires an API key —
|
|
# $DUSKADIY_API_KEY, from the gitignored conf.d/secrets.fish). The LAN endpoint above is keyless.
|
|
# Same server, same model ids — reuse the list above via a YAML anchor.
|
|
- type: openai-compatible
|
|
name: duskadiy
|
|
api_base: https://llm.duskadiy.com/api/v1
|
|
models: *lan_models
|