亚洲模型实验室 · arXiv 论文
Continuous vs Discrete Emergence
复现 Yi Figure 3：在同一 few-shot 任务上并列连续误差与离散 exact match，避免把测量阈值误当作能力突然出现。
continuousDiscreteEmergenceInformation grammar
它如何组织信息
任务难度分行 + 连续误差实线 + exact-match 虚线 + shots 共享横轴
- Visual system
- VS-78 Yi continuous-discrete emergence
- Component ID
yi-continuous-discrete-emergence
Use when
适用判断
适合研究 emergent ability 与测量伪影；连续、离散指标必须来自同一预测样本。
Figure-level evidence
Figure 3 · PDF p.11
两个线性系数推断任务，比较 difference-to-target 与 exact match 随 shots 变化。
- Verified
- 2026-07-13
- Paper source
- Yi Technical Report
Structured data
可替换的数据模型
示意数据，仅复现信息结构，不代表最新榜单结果。
{
"shots": [
4,
8,
12,
16,
20,
24,
28,
32,
36,
40
],
"tasks": [
{
"label": "2 COEFFICIENTS · [1, -1]",
"continuousRange": [
0,
50
],
"exactMax": 0.6,
"models": [
{
"label": "Yi-34B",
"continuous": [
43,
35,
28,
20,
16,
18,
14,
12,
8,
10
],
"exact": [
0.08,
0.28,
0.3,
0.38,
0.36,
0.28,
0.45,
0.32,
0.47,
0.26
]
},
{
"label": "Llama-2-70B",
"continuous": [
46,
39,
37,
33,
30,
28,
26,
24,
23,
22
],
"exact": [
0.02,
0.1,
0.2,
0.35,
0.42,
0.45,
0.4,
0.38,
0.46,
0.55
]
},
{
"label": "Yi-6B",
"continuous": [
49,
42,
39,
36,
31,
22,
20,
25,
18,
17
],
"exact": [
0,
0.02,
0.04,
0.08,
0.1,
0.16,
0.06,
0.14,
0.08,
0.12
]
}
]
},
{
"label": "5 COEFFICIENTS · [1, 1, 1, 1, 1]",
"continuousRange": [
0,
170
],
"exactMax": 0.8,
"models": [
{
"label": "Yi-34B",
"continuous": [
150,
54,
45,
41,
38,
47,
36,
34,
37,
39
],
"exact": [
0,
0,
0.01,
0.02,
0.01,
0.03,
0.02,
0.04,
0.01,
0.03
]
},
{
"label": "Llama-2-70B",
"continuous": [
60,
14,
8,
5,
10,
7,
6,
11,
8,
6
],
"exact": [
0.12,
0.44,
0.56,
0.78,
0.56,
0.74,
0.66,
0.72,
0.68,
0.76
]
},
{
"label": "Yi-6B",
"continuous": [
165,
83,
76,
57,
45,
39,
34,
36,
29,
31
],
"exact": [
0,
0,
0.02,
0,
0.03,
0,
0.02,
0.01,
0.03,
0.01
]
}
]
}
]
}Related components