YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Student Simulation v2
Quick reference. See runall.sh for the full pipeline.
关键变更(v2 vs v1)
- 公式语义:
x_new = x - (1 - α) · P · h,α=1 不变,α=0 完全压制 - Sweep 范围: α ∈ [0, 1](v1 范围越界导致崩溃伪迹)
- 方向版本: 只保留
v1_raw和新版v_pca_subspace(k=3 子空间) - 新功能:
JointResidualSteerer防止跨维度代偿 - 新指标:
count_real_monitoring()区分真反思和填充词 - 新指标:
is_collapsed()用 4-gram 重复 + 长度比,比 v1 更稳 - 08b: attention 输出诊断(informational only)
- 10_infer: 加入 runall 作为 sanity check
- 删除: LLM rater (11_llm_quality_rating.py)
用法
# 单卡完整跑
bash runall.sh
# 启用 anti-leak joint steering
JOINT=1 bash runall.sh
# 只跑某些 stage
STAGES=8,8b,9,10 bash runall.sh
# 只对一个题做 inference
python scripts/10_infer.py --dim planning --alphas 1.0 0.5 0.0 \
--problem "Find x such that x^2=49"
目录
data/
models/ # Qwen3-30B-A3B-Thinking-2507
cots/ # raw + labeled CoTs
routing/ # router top-k dumps
activations/ # decision-point residuals
checkpoints/
planning_v1_raw.pt
planning_v_pca_subspace.pt # 新版 k=3 子空间
monitoring_v1_raw.pt
monitoring_v_pca_subspace.pt
results/
sweep_log.jsonl # 含 steered_text
final_report.md
attention_diagnostic.{json,png} # 新
infer_sanity_planning.json # 新
infer_sanity_monitoring.json # 新
logs/
关键 config (configs/model.py)
ALPHA_SWEEP = [0.0, 0.1, 0.2, 0.3, 0.5, 0.75, 1.0]
DIRECTION_VERSIONS = ["v1_raw", "v_pca_subspace"]
PCA_SUBSPACE_K = 3
ANTI_LEAK_BETA = 0.3
GEN_CONFIG["max_new_tokens"] = 12000 # 之前 4096 太小
GEN_CONFIG_FAST["max_new_tokens"] = 8192 # 之前 1024 太小
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support