ruoxi_sun 4f274c2c32 add fingerprint model library & fp_fusion integration
- fingerprint benchmark: add fp_fusion(26-cell fusion) to collect_results/run.py
- run_llmmap.py: default model path to evalstone built-in model_library
- add model libraries (llmdetector 11 refs / fp_fusion 8 fusion refs /
  llmmap templates 60 models incl 8 new: GLM-5.2/5.3, DeepSeek-Flash/Pro/
  Flash-0731, Kimi-K3, MiniMax-M2.7, TianGong-Taie)
- add fp_fusion engine (battery/engine/scorer) + docs
- gitignore: exclude binary model weights and temp backups
2026-09-03 02:38:35 +00:00

37 lines
2.5 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# FP-Fusion 参考指纹库(自包含)
同协议one-token/v1可信参考指纹`--reference` / `--detector-reference` 使用。
| 文件 | 真实模型 | 采样来源 | 日期 |
|---|---|---|---|
| qwen3-4b_reference.json | Qwen3-4B | :30002 (CPU, 官方权重) | 2026-08-21 |
| glm520_reference.json | GLM-5.2 | :30000 (P800, 官方权重) | 2026-08-21 |
| qwen3-8b_reference.json | Qwen3-8B | :30003 (CPU, 官方权重) | 2026-08-28 |
| glm53_reference.json | GLM-5.3 | api.vectron.meta-stone.com (ZhipuAi/GLM-5.3) | 2026-09-01 |
| deepseek_v4_flash_reference.json | DeepSeek-V4-Flash | api.vectron.meta-stone.com (DeepSeek/DeepSeek-V4-Flash) | 2026-09-01 |
| kimi_k3_reference.json | Kimi-K3 | api.vectron.meta-stone.com (MoonshotAi/Kimi-K3) | 2026-09-01 |
| deepseek_v4_pro_reference.json | DeepSeek-V4-Pro | api.vectron.meta-stone.com (DeepSeek/DeepSeek-V4-Pro) | 2026-09-01 |
| glm52_vectron_reference.json | GLM-5.2 (vectron) | api.vectron.meta-stone.com (ZhipuAi/GLM-5.2) | 2026-09-02 |
| minimax_m27_reference.json | MiniMax-M2.7 | api.vectron.meta-stone.com (MiniMax/MiniMax-M2.7) | 2026-09-02 |
| tiangong_taie_reference.json | TianGong/Taie | api.vectron.meta-stone.com (TianGong/Taie) | 2026-09-02 |
| deepseek_v4_flash_0731_reference.json | DeepSeek-V4-Flash-0731 | api.vectron.meta-stone.com (DeepSeek/DeepSeek-V4-Flash-0731) | 2026-09-02 |
采集命令detector CLI, strict 建议):
node dist/cli.js fingerprint --base-url <可信端点>/v1 --model <模型名> \
--preset strict --out references/<名字>_reference.json
注意: 参考会随上游静默更新而漂移,高置信场景请定期重采。
## fp_fusion 独立参考26-cell区分于 llmdetector 的 16-cell
fp_fusion 的 D 层 = llmdetector 16 cells + **10 个新增 cell**binary-season/pet/sea-mountain/tea-coffee + day-of-week×中英
为让 fp_fusion 作为独立 benchmark 真正用上新增 cell参考指纹**分开采样**
- 16 个共享 cell直接复用 detector CLI 采集的 `*_reference.json`
- 10 个新增 cell`collect_fp_extra_reference.py --api-key ... --model ... --out <extra>.json`
(复用 fp_fusion engine 的归一化逻辑,中英各 25 条;中文 binary cell 因归一化规则
不识别中文回答为有效,与 fp_fusion 运行时行为一致,自动被 min_valid 过滤)
- 合并:`merge_fusion_reference.py``*_fusion_reference.json`26 cells, one-token/v1
llmdetector 比对继续用 16-cell 的 `*_reference.json`fp_fusion 比对用 `*_fusion_reference.json`