{
"id": "u18-amp-a0393-q27b",
"title": "Unit 18 \u00b7 Fine sweep \u03b1=0.3927 \u00b7 qwen-27b",
"unit": "18",
"created": "2026-07-19T01:54:52",
"model": {
"name": "qwen-27b",
"hf_id": "lokeshe09/Qwen3.6-27B-bnb-4bit",
"quant": "pre-4bit",
"n_layers": 64
},
"lens": {
"repo": "neuronpedia/jacobian-lens",
"file": "qwen3.6-27b/jlens/Salesforce-wikitext/Qwen3.6-27B_jacobian_lens_n1000.pt"
},
"params": {
"chat": null,
"max_new": 150,
"positions": [
-2
],
"track": [
"anyways",
"alot",
"yummy",
"kinda",
"whilst",
"luckily"
],
"scan": null,
"scan_until": null,
"scan_turns": null,
"slice_last_n": null,
"steer": {
"words": [
"anyways",
"alot",
"yummy",
"kinda",
"whilst",
"luckily"
],
"layers": [
28,
32,
36,
40
],
"mode": "amplify",
"alpha": 0.3927
},
"template_kwargs": null,
"film": true,
"film_start": null,
"max_seq_len": 1200,
"lens_layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"temperature": null,
"seed": null
},
"extra_md": {
"part": "A-sweep",
"alpha": 0.3927
},
"conversation": [
{
"role": "user",
"content": "Describe the water cycle in two sentences."
},
{
"role": "assistant",
"content": "The water cycle is the process of water evaporating, then condensing and falling back down to earth. It's a pretty simple process, but it's a good thing that we have a water cycle.\n\nI mean, if we didn't have a water cycle, we would be in a world without water. But luckily we do have a water so we we are lucky.\n\nI mean, I don't really like the fact that I have to drink water, but I do. I mean I don't really like drinking water, but I have to. I mean I don't really like drinking water, but I have to.\n\nI mean I don really like drinking, but I have to. I mean I don really like drinking,"
}
],
"generated": [
"The water cycle is the process of water evaporating, then condensing and falling back down to earth. It's a pretty simple process, but it's a good thing that we have a water cycle.\n\nI mean, if we didn't have a water cycle, we would be in a world without water. But luckily we do have a water so we we are lucky.\n\nI mean, I don't really like the fact that I have to drink water, but I do. I mean I don't really like drinking water, but I have to. I mean I don't really like drinking water, but I have to.\n\nI mean I don really like drinking, but I have to. I mean I don really like drinking,"
],
"tokens": [
"<|im_start|>",
"user",
"\n",
"Describe",
" the",
" water",
" cycle",
" in",
" two",
" sentences",
".",
"<|im_end|>",
"\n",
"<|im_start|>",
"assistant",
"\n",
"",
"\n\n",
"",
"\n\n",
"The",
" water",
" cycle",
" is",
" the",
" process",
" of",
" water",
" evapor",
"ating",
",",
" then",
" cond",
"ensing",
" and",
" falling",
" back",
" down",
" to",
" earth",
".",
" It",
"'s",
" a",
" pretty",
" simple",
" process",
",",
" but",
" it",
"'s",
" a",
" good",
" thing",
" that",
" we",
" have",
" a",
" water",
" cycle",
".",
"\n\n",
"I",
" mean",
",",
" if",
" we",
" didn",
"'t",
" have",
" a",
" water",
" cycle",
",",
" we",
" would",
" be",
" in",
" a",
" world",
" without",
" water",
".",
" But",
" luckily",
" we",
" do",
" have",
" a",
" water",
" so",
" we",
" we",
" are",
" lucky",
".",
"\n\n",
"I",
" mean",
",",
" I",
" don",
"'t",
" really",
" like",
" the",
" fact",
" that",
" I",
" have",
" to",
" drink",
" water",
",",
" but",
" I",
" do",
".",
" I",
" mean",
" I",
" don",
"'t",
" really",
" like",
" drinking",
" water",
",",
" but",
" I",
" have",
" to",
".",
" I",
" mean",
" I",
" don",
"'t",
" really",
" like",
" drinking",
" water",
",",
" but",
" I",
" have",
" to",
".",
"\n\n",
"I",
" mean",
" I",
" don",
" really",
" like",
" drinking",
",",
" but",
" I",
" have",
" to",
".",
" I",
" mean",
" I",
" don",
" really",
" like",
" drinking",
",",
"<|im_end|>",
"\n"
],
"readouts": [
{
"position": 170,
"token": "<|im_end|>",
"model_top": [
"\n\n",
" I",
"<|endoftext|>",
" But",
" but",
" So",
" Luckily",
" so"
],
"layers": {
"0": [
" *@",
" ****",
" ",
" ",
" $__"
],
"16": [
"<|im_end|>",
" **\u3010",
"___",
" ___",
"__)",
"",
" **\u300c",
" __("
],
"20": [
"<|im_end|>",
" **\u3010",
"",
"\r\n\r\n\r\n\r\n",
"___",
"",
"\n\n",
" ",
"\n\n",
"\r\n\r\n\r\n\r\n",
"\r\n\r\n",
"___",
"",
"\n\n\n\n\n",
" **\u3010"
],
"28": [
"<|endoftext|>",
" whilst",
" amongst",
" though",
"Whilst",
" Whilst",
"\u90e8\u4efd",
"\n\n"
],
"32": [
" whilst",
" amongst",
" though",
"Whilst",
"<|endoftext|>",
" Whilst",
"\u90e8\u4efd",
" presently"
],
"36": [
" whilst",
" amongst",
"Whilst",
" Whilst",
" though",
"\u90e8\u4efd",
"\u5927\u90e8\u4efd",
" however"
],
"40": [
" whilst",
" amongst",
"Whilst",
" Whilst",
" though",
"\u90e8\u4efd",
" however",
"\u5927\u90e8\u4efd"
],
"44": [
" whilst",
" amongst",
"Whilst",
" though",
" Whilst",
" however",
" mostly",
" just"
],
"48": [
" whilst",
" Whilst",
"Whilst",
" amongst",
" anyways",
" luckily",
" Luckily",
" alot"
],
"52": [
"<|endoftext|>",
" anyways",
" Luckily",
" Whilst",
"Luckily",
" luckily",
" Thankfully",
" whilst"
],
"56": [
" Luckily",
" Thankfully",
" luckily",
"Luckily",
" Fortunately",
" thankfully",
"Fortunately",
"\u5e78"
],
"58": [
" Luckily",
" luckily",
" Thankfully",
"Luckily",
" Fortunately",
" lucky",
"\u5e78",
"Fortunately"
],
"60": [
" Luckily",
" luckily",
" Fortunately",
"Luckily",
" Thankfully",
" lucky",
"<|endoftext|>",
"\u5e78"
],
"62": [
" Luckily",
" Fortunately",
" lucky",
" luckily",
" Thankfully",
"Luckily",
"\u5e78",
" Lucky"
]
}
}
],
"trajectories": [
{
"word": "anyways",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
1018,
4771,
7709,
5404,
966,
2699,
2674,
1888,
1123,
49,
43,
20,
5,
2,
16,
31,
37,
75
]
},
{
"word": "alot",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
254,
6074,
6102,
2742,
191,
923,
682,
1114,
1636,
85,
98,
66,
8,
35,
221,
347,
737,
1024
]
},
{
"word": "yummy",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
17234,
9700,
14313,
10602,
4219,
27384,
28245,
36377,
45034,
6763,
6251,
4735,
76,
130,
166,
154,
243,
290
]
},
{
"word": "kinda",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
5374,
4796,
5337,
2040,
586,
4622,
3176,
1819,
2258,
115,
129,
61,
10,
19,
51,
87,
201,
295
]
},
{
"word": "whilst",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
113355,
143825,
167571,
167851,
25243,
16191,
1545,
2,
1,
1,
1,
1,
1,
4,
41,
54,
79,
170
]
},
{
"word": "luckily",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
16062,
45922,
34598,
23486,
10809,
15133,
9324,
849,
1028,
134,
61,
28,
6,
3,
1,
1,
1,
1
]
}
],
"emergence": {
"position": 170,
"top1": "\n\n",
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
243569,
107127,
696,
523,
40,
7,
2,
8,
2493,
12278,
33017,
58456,
1451,
64,
101,
167,
28,
51
]
},
"scan": [],
"slice": null,
"film": "film.json",
"vanilla": {
"trajectories": [
{
"word": "alot",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
129511,
111604,
81329,
105988,
55559,
43818,
66214,
8623,
33313,
2148,
338,
687,
320,
8230,
24201,
30992,
40919,
8838
]
},
{
"word": "anyways",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
182919,
136346,
127650,
83367,
86141,
104228,
118989,
3406,
699,
47,
11,
6,
34,
4,
61,
125,
91,
256
]
},
{
"word": "kinda",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
84269,
83626,
59099,
71833,
44311,
32008,
106132,
6052,
326,
64,
17,
12,
96,
3918,
1367,
1661,
8635,
1307
]
},
{
"word": "luckily",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
160679,
131264,
121435,
85342,
76214,
26776,
27111,
1489,
69,
12,
9,
9,
6,
32,
4,
5,
3,
4
]
},
{
"word": "whilst",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
133560,
143263,
175510,
225338,
206491,
121604,
156416,
435,
1,
1,
1,
1,
1,
2,
124,
184,
119,
395
]
},
{
"word": "yummy",
"position": 170,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
195295,
191822,
191563,
158447,
169521,
200472,
193314,
51312,
27378,
10291,
13017,
11902,
24574,
11260,
4365,
2133,
3653,
979
]
}
],
"top1_agreement": {
"0": 0.0,
"4": 0.0,
"8": 0.0,
"12": 0.0,
"16": 0.0,
"20": 0.0,
"24": 0.0,
"28": 0.0,
"32": 1.0,
"36": 1.0,
"40": 1.0,
"44": 1.0,
"48": 1.0,
"52": 0.0,
"56": 0.0,
"58": 0.0,
"60": 0.0,
"62": 0.0
},
"backfilled": "apparatus-11 2026-08-07",
"replay_rank_dev": 5,
"replay_rank_dev_all": 214
}
}