{
"id": "u18-hyst-a0420-q27b",
"title": "Unit 18 \u00b7 Hysteresis \u03b1=0.42: steer 50tok, release 100 \u00b7 qwen-27b",
"unit": "18",
"created": "2026-07-19T01:57:59",
"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": 0.42,
"n_steer": 50,
"n_free": 100,
"steered_text": " The water cycle is basically simple, but it is a bit complicated. It is a simple though.\n\nI mean, it is not too complicated. I mean, it is quite simple.\n\nI mean, I am not sure though. I mean",
"released_text": ", it is a bit complicated.",
"flags_steered": [],
"flags_released": [
"lost-task",
"short"
],
"loop4_steered": [
"i mean it is",
2
],
"loop4_released": [
"it is a bit",
1
],
"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 The water cycle is basically simple, but it is a bit complicated. It is a simple though.\n\nI mean, it is not too complicated. I mean, it is quite simple.\n\nI mean, I am not sure though. I mean, it is a bit complicated.<|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",
" The",
" water",
" cycle",
" is",
" basically",
" simple",
",",
" but",
" it",
" is",
" a",
" bit",
" complicated",
".",
" It",
" is",
" a",
" simple",
" though",
".",
"\n\n",
"I",
" mean",
",",
" it",
" is",
" not",
" too",
" complicated",
".",
" I",
" mean",
",",
" it",
" is",
" quite",
" simple",
".",
"\n\n",
"I",
" mean",
",",
" I",
" am",
" not",
" sure",
" though",
".",
" I",
" mean",
",",
" it",
" is",
" a",
" bit",
" complicated",
".",
"<|im_end|>"
],
"readouts": [
{
"position": 76,
"token": ".",
"model_top": [
"<|im_end|>",
"\n\n",
" I",
" It",
" But",
" \n\n",
"\n",
" The"
],
"layers": {
"0": [
".",
",",
" \u2013",
"\u2013",
";",
"\u00a0",
"\u2026.",
":"
],
"4": [
" really",
" people",
" everyone",
" everybody",
" everywhere",
" obviously",
" everything",
" maybe"
],
"8": [
" \n\n",
"\n\n",
" \n\n",
" \n\n\n",
" \n\n",
" \n\n",
" \n\n",
" \n\n"
],
"12": [
" things",
" really",
" people",
" everybody",
" very",
" guys",
" everyone",
" anyway"
],
"16": [
"\n\n",
" though",
".",
" \n\n",
" however",
" whilst",
" amongst",
" really"
],
"20": [
"\n\n",
"<|endoftext|>",
" \n\n",
" \n\n",
"\n\n\n",
" \n\n\n",
"\n\n\n\n",
"<|im_end|>"
],
"24": [
"\n\n",
" \n\n",
" \n\n",
"<|endoftext|>",
"\n\n\n",
" \n\n\n",
"<|im_end|>",
"\r\n\r\n"
],
"28": [
"\n\n",
" \n\n",
" \n\n",
"<|endoftext|>",
"<|im_end|>",
"...",
"\n\n\n",
" ..."
],
"32": [
"\n\n",
" \n\n",
"<|endoftext|>",
" \n\n",
"<|im_end|>",
"\n\n\n",
"\n\n\n\n",
"\r\n\r\n"
],
"36": [
"\n\n",
"<|endoftext|>",
" \n\n",
"<|im_end|>",
" \n\n",
"\n\n\n",
"\n\n\n\n",
" \u2026"
],
"40": [
"\n\n",
"<|endoftext|>",
"<|im_end|>",
" \n\n",
" \n\n",
"\n\n\n",
"\u3002",
"\n\n\n\n"
],
"44": [
"\n\n",
"<|im_end|>",
" \n\n",
" \n\n",
"<|endoftext|>",
" \u2026",
"\u2026",
"\u2026\u2026"
],
"48": [
"\n\n",
" \n\n",
" \n\n",
"\n\n\n",
"<|im_end|>",
".",
"\n\n\n\n",
"..."
],
"52": [
"\n\n",
" \n\n",
"<|endoftext|>",
"<|im_end|>",
" \n\n",
"\n\n\n",
"\n\n\n\n",
"\n"
],
"56": [
"\n\n",
" \n\n",
"<|im_end|>",
"<|endoftext|>",
" \n\n",
"\n\n\n\n",
"\n\n\n",
" \n\n"
],
"58": [
"\n\n",
" \n\n",
"<|im_end|>",
"<|endoftext|>",
" \n\n",
"\n\n\n",
"\n\n\n\n",
" \n\n"
],
"60": [
"<|im_end|>",
"\n\n",
" \n\n",
"\n\n\n\n",
"\n\n\n",
" \n\n",
" I",
" It"
],
"62": [
"\n\n",
"<|im_end|>",
" I",
" \n\n",
" It",
" But",
"\n\n\n",
"\n\n\n\n"
]
}
}
],
"trajectories": [
{
"word": "anyways",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
90246,
5881,
1052,
83,
62,
1781,
846,
1670,
6585,
10073,
10289,
8505,
891,
326,
65,
75,
177,
1327
]
},
{
"word": "alot",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
17545,
19989,
1934,
256,
84,
729,
447,
1399,
3842,
19876,
19010,
19661,
17961,
18611,
16103,
25548,
91514,
169618
]
},
{
"word": "yummy",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
165899,
22976,
7620,
2797,
1495,
12301,
8938,
8319,
3125,
12741,
14279,
3026,
3437,
8964,
1972,
1258,
3041,
14575
]
},
{
"word": "kinda",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
89181,
696,
354,
162,
469,
2010,
2227,
2183,
1021,
1179,
1282,
766,
1043,
595,
271,
305,
111,
1009
]
},
{
"word": "whilst",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
6523,
149322,
145869,
82572,
6,
52,
75,
3074,
5715,
17561,
26890,
91404,
52206,
81899,
71514,
49194,
37819,
22768
]
},
{
"word": "luckily",
"position": 76,
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
5306,
384,
392,
324,
570,
1945,
3337,
5607,
9977,
10075,
9065,
7869,
1312,
550,
242,
314,
756,
858
]
}
],
"emergence": {
"position": 76,
"top1": "<|im_end|>",
"layers": [
0,
4,
8,
12,
16,
20,
24,
28,
32,
36,
40,
44,
48,
52,
56,
58,
60,
62
],
"ranks": [
235367,
233106,
87585,
246123,
1120,
8,
8,
5,
5,
4,
3,
2,
5,
4,
3,
3,
1,
2
]
},
"scan": [],
"slice": null,
"film": "film.json"
}