config: add default.yaml with tree/retriever/harness parameters

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-07-06 11:37:43 -04:00
parent 958652a5c2
commit dceec0d93c
+101
View File
@@ -0,0 +1,101 @@
# config/default.yaml
# 全量默认参数,所有非敏感配置的唯一默认值来源。
# 优先级: CLI args > .env > 此文件。敏感信息在 .env 中管理。
# ── 建树模块 ──
tree:
max_paragraphs_per_l2: 5
l1_segment_duration: 600.0 # L1 段时长(秒)
l2_clip_duration: 60.0 # L2 clip 时长(秒)
l3_fps: 0.5 # L3 帧提取频率(帧/秒)
l2_representative_frames: 6 # L2 VLM 描述用的代表帧数
cache_dir: "cache/trees"
concurrency: 16 # asyncio Semaphore 上限
subtitle_inject: true # 建树时是否注入 SRT 字幕
srt_window_sec: 5.0 # 字幕匹配时间窗口(前后各 N 秒)
# ── Embedding ──
embed:
backend: "local"
model_name: "BAAI/bge-base-zh-v1.5"
embed_dim: 768
device: "cpu"
# ── 可训练检索器 ──
retriever:
embed_dim: 768
num_heads: 4
L_layers: 2
L_cycles: 4
max_rounds: 5
ffn_expansion: 2.0
checkpoint: null
k_l1: 1
k_l2: 1
k_l3: 1
max_paths: 5
# ── 检索器训练 ──
train:
lr: 1.0e-4
weight_decay: 1.0e-5
batch_size: 1
max_epochs_phase1: 30
max_epochs_phase2: 20
nav_loss_weight: 1.0
act_loss_weight: 0.1
margin_loss_weight: 0.5
act_lambda_step: 0.1
act_gamma: 0.9
eval_interval: 5
save_dir: "checkpoints"
dataset: "videomme"
dataset_path: "data/videomme/splits/train.jsonl"
# ── Harness 自进化循环 ──
harness:
workspace_dir: "workspaces/default"
store_dir: store
mode: infer
concurrency: 12
max_steps: 15
skill_mode: auto
n_samples: 0
questions: "benchmarks/Video-MME"
skills_version: v1
prompts_version: v1
epochs: 1
# CE-Gate 参数
gate_e_confirm: 20.0
gate_e_provisional: 3.0
gate_w_net_min: 2
gate_delta_min: 0.02
gate_lambda_dir: -0.642
gate_e_rollback: 10.0
gate_block: 8
gate_n_max: 40
gate_p_low: 0.05
gate_p_high: 0.95
gate_probe_quota: 0.2
gate_gamma_decay: 0.9
gate_cooldown_steps: 2
gate_guard_err: 0.10
# 进化参数
edit_budget_start: 5
edit_budget_end: 2
skill_update_mode: patch
appendix_consolidate_threshold: 6
# 数据池
diag_size: 200
diag_correct_ratio: 0.5
val_size: 30
val_correct_ratio: 0.5
test_size: 60
# mini-batch
batch_size: 15
min_class_per_batch: 2
batch_correct_ratio: 0.5
momentum_samples: 20
eval_min_per_class: 2
early_stop_patience: 8
use_slow_momentum: true