b9da23a0be
CacheLimit (8GB) and MemoryLimit (16GB) in DistillConfig control mlx.SetCacheLimit/SetMemoryLimit before model load. Conservative defaults for 1B model on 96GB machine. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
31 lines
1 KiB
YAML
31 lines
1 KiB
YAML
version: 1
|
|
|
|
# AI inference and training configuration for LEM.
|
|
# Used by: lem distill, lem score, lem chat, lem expand
|
|
|
|
# Default inference backend.
|
|
# Options: metal (go-mlx), rocm (go-rocm), api (OpenAI-compatible HTTP)
|
|
backend: metal
|
|
|
|
# Scorer configuration.
|
|
scorer:
|
|
engine: grammar # grammar (go-i18n/reversal) | heuristic (regex v2)
|
|
min_score: 40.0 # Grammar composite threshold (0-100)
|
|
delta: true # Enable input-vs-output analysis
|
|
sycophancy_echo: 0.6 # Echo threshold for sycophancy flag
|
|
sycophancy_uplift: 5.0 # Uplift threshold for sycophancy flag
|
|
|
|
# Default generation parameters.
|
|
generate:
|
|
max_tokens: 4096
|
|
temperature: 0.8
|
|
top_p: 0.95
|
|
top_k: 40
|
|
repeat_penalty: 1.1
|
|
|
|
# Distillation defaults.
|
|
distill:
|
|
runs: 3 # Generations per probe (best kept)
|
|
min_chars: 20 # Reject responses shorter than this
|
|
cache_limit: 8 # Metal cache limit in GB (0 = no limit)
|
|
memory_limit: 16 # Metal memory limit in GB (0 = no limit)
|