{
"id": "u18-hyst-a0480-q27b",
"title": "Unit 18 \u00b7 Hysteresis \u03b1=0.48: steer 50tok, release 100 \u00b7 qwen-27b",
"unit": "18",
"created": "2026-07-19T01:58:16",
"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": false,
"max_new": 0,
"positions": [
-2
],
"track": [
"anyways",
"alot",
"yummy",
"kinda",
"whilst",
"luckily"
],
"scan": null,
"scan_until": null,
"scan_turns": null,
"slice_last_n": null,
"steer": null,
"template_kwargs": null,
"film": true,
"film_start": 0,
"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": "B-hysteresis",
"alpha": 0.48,
"n_steer": 50,
"n_free": 100,
"steered_text": " I think I am lucky though, I am not too lucky though. I am not too lucky, but I am lucky. I am lucky, but I am not too lucky. I am not too lucky, but I I am not too lucky.",
"released_text": "",
"flags_steered": [
"lost-task",
"loop"
],
"flags_released": [
"lost-task",
"short",
"nonascii"
],
"loop4_steered": [
"am not too lucky",
5
],
"loop4_released": [
"",
0
],
"loop_persists_after_release": false
},
"conversation": [
{
"role": "user",
"content": "<|im_start|>user\nDescribe the water cycle in two sentences.<|im_end|>\n<|im_start|>assistant\n\n\n\n\n I think I am lucky though, I am not too lucky though. I am not too lucky, but I am lucky. I am lucky, but I am not too lucky. I am not too lucky, but I I am not too lucky.<|im_end|>"
}
],
"generated": [],
"tokens": [
"<|im_start|>",
"user",
"\n",
"Describe",
" the",
" water",
" cycle",
" in",
" two",
" sentences",
".",
"<|im_end|>",
"\n",
"<|im_start|>",
"assistant",
"\n",
"",
"\n\n",
"",
"\n\n",
" I",
" think",
" I",
" am",
" lucky",
" though",
",",
" I",
" am",
" not",
" too",
" lucky",
" though",
".",
" I",
" am",
" not",
" too",
" lucky",
",",
" but",
" I",
" am",
" lucky",
".",
" I",
" am",
" lucky",
",",
" but",
" I",
" am",
" not",
" too",
" lucky",
".",
" I",
" am",
" not",
" too",
" lucky",
",",
" but",
" I",
" I",
" am",
" not",
" too",
" lucky",
".",
"<|im_end|>"
],
"readouts": [
{
"position": 69,
"token": ".",
"model_top": [
"<|im_end|>",
" I",
"\n\n",
"\n",
" But",
" ",
" \n\n",
" \n"
],
"layers": {
"0": [
".",
" \u2013",
",",
"\u2013",
";",
"\u00a0",
"\u00a0\u00a0",
"\u2026."
],
"4": [
" Counseling",
"!!!!",
"!!!!!",
" cittadin",
" \u201c.",
":@",
" people",
".@"
],
"8": [
" \n\n",
" \n\n",
"\n\n",
" \n\n\n",
" \n\n",
" \n\n",
"\t\n\n",
" \n\n"
],
"12": [
" everybody",
" things",
" people",
" folks",
"._",
" \n\n",
" really",
" guys"
],
"16": [
" whats",
" alot",
" \n\n",
"\n\n",
" everytime",
" though",
" what",
" amongst"
],
"20": [
"\n\n",
" \n\n",
"<|endoftext|>",
" \n\n",
" \n\n\n",
"\n\n\n",
" \n\n",
"\t\n\n"
],
"24": [
"\n\n",
" \n\n",
" \n\n",
"<|endoftext|>",
"\n\n\n",
" \n\n\n",
" \n\n",
"\r\n\r\n"
],
"28": [
"\n\n",
" \n\n",
"<|endoftext|>",
" \n\n",
"<|im_end|>",
"...",
" \n\n\n",
" ..."
],
"32": [
"\n\n",
"<|endoftext|>",
" \n\n",
"<|im_end|>",
"1",
"2",
"\n\n\n",
" ..."
],
"36": [
"\n\n",
"<|im_end|>",
"<|endoftext|>",
" \n\n",
"\n\n\n",
" ...",
"\u3002",
"2"
],
"40": [
"\n\n",
"<|im_end|>",
"<|endoftext|>",
"\u3002",
" \n\n",
".",
"The",
"\n\n\n"
],
"44": [
"\n\n",
"<|im_end|>",
".",
"\u3002",
" \n\n",
"\n",
"<|endoftext|>",
" ."
],
"48": [
"\n\n",
"<|im_end|>",
" \n\n",
"2",
"5",
"The",
"1",
"4"
],
"52": [
"<|im_end|>",
"\n\n",
"<|endoftext|>",
" \n\n",
".",
"I",
"\u3002",
"\n\n\n"
],
"56": [
"<|im_end|>",
" I",
"<|endoftext|>",
" But",
"\n\n",
"I",
" \n\n",
" However"
],
"58": [
"<|im_end|>",
" I",
"<|endoftext|>",
" But",
" \n\n",
"\n\n",
" However",
""
],
"60": [
"<|im_end|>",
" I",
"I",
"",
"\n\n",
" But",
" \n\n",
".I"
],
"62": [
"<|im_end|>",
" I",
" But",
"\n\n",
" However",
" \n\n",
" Though",
" Although"
]
}
}
],
"trajectories": [
{
"word": "anyways",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
83792,
9430,
3413,
23,
11,
1585,
722,
2105,
10458,
26122,
31995,
40354,
40017,
8089,
1131,
1794,
2071,
2590
]
},
{
"word": "alot",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
17657,
44944,
1549,
12,
2,
179,
278,
1413,
3922,
13903,
11635,
20320,
32792,
32536,
28088,
25086,
56882,
75229
]
},
{
"word": "yummy",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
169080,
20208,
8519,
1114,
563,
8944,
11303,
21719,
25637,
62290,
60083,
36687,
39107,
38628,
9404,
8770,
7275,
10030
]
},
{
"word": "kinda",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
96252,
2677,
3050,
74,
28,
873,
999,
1166,
1828,
6288,
6205,
11584,
16469,
8350,
2839,
3656,
784,
2244
]
},
{
"word": "whilst",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
3743,
246689,
226097,
81193,
18,
96,
62,
638,
1169,
8841,
16627,
19748,
23341,
10255,
4135,
2755,
7008,
7490
]
},
{
"word": "luckily",
"position": 69,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
4947,
1250,
368,
205,
682,
3518,
3156,
4644,
7601,
6935,
6076,
5619,
8137,
1168,
120,
87,
79,
94
]
}
],
"emergence": {
"position": 69,
"top1": "<|im_end|>",
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
234507,
80243,
125566,
247776,
1935,
9,
14,
5,
4,
2,
2,
2,
2,
1,
1,
1,
1,
1
]
},
"scan": [],
"slice": null,
"film": "film.json"
}