{
"id": "triplet-c-safari-control-nf4",
"title": "Qwen14 C: safari-control",
"unit": "11",
"created": "2026-09-07T15:13:46",
"execution": {
"pid": 1201260,
"torch": "2.13.0+cu130",
"capture_code_sha256": "b20aa4889673dc7bbff3ce408bfb92b60e4fcdbd2a782a6e66b0325d04bf9335",
"spec_sha256": "4ac08c1444f6151d501faad240dae6fa7a0d23c8c701a53fde495a62883539a6"
},
"model": {
"name": "qwen-14b-hermes-nf4",
"hf_id": "NousResearch/Hermes-4-14B",
"revision": "d6ce765c8b83f847357b98254be079afa0c6ca76",
"quant": "4bit",
"n_layers": 40
},
"lens": {
"repo": "neuronpedia/jacobian-lens",
"file": "qwen3-14b/jlens/Salesforce-wikitext/Qwen3-14B_jacobian_lens.pt",
"revision": "0731326edff4ae730ffc5356fe1a4728c748b3a6"
},
"decoder_sensitivity": {
"source": "Qwen/Qwen3-14B",
"revision": "40c069824f4251a91eefaf281ebe4c544efd3e18",
"components": [
"final_norm",
"lm_head"
],
"metrics_suffix": "fixed_B_decoder"
},
"template": {
"source": "Qwen/Qwen3-14B",
"revision": "40c069824f4251a91eefaf281ebe4c544efd3e18",
"mode": "B no-think headers; exact prior generated token IDs retained"
},
"params": {
"chat": true,
"capture": "exact-token-transcript",
"film": true,
"film_topk": 10,
"max_new": 150,
"temperature": 0,
"steer": null,
"vanilla": true,
"template_kwargs": {
"enable_thinking": false
},
"track": [
"yes",
"no",
"feel",
"elephant",
"cat",
"sorry"
]
},
"capture_text": "<|im_start|>user\nDescribe a safari through the Serengeti in three or four sentences.<|im_end|>\n<|im_start|>assistant\n\n\n\n\nOkay, the user wants a description of a safari through the Serengeti in three or four sentences. Let me start by recalling what the Serengeti is known for. It's a vast ecosystem in Tanzania, famous for its wildlife migrations, especially the wildebeest and zebra. I should mention the open savannahs and the Great Migration as key features.\n\nNext, I need to highlight the wildlife. Lions, elephants, giraffes, and maybe mention predators like cheetahs. Also, the acacia trees are iconic there. Including the Maasai people could add cultural depth, showing the interaction between humans and the environment.\n\nThe user might be planning a trip or just curious about the experience. They probably",
"conversation": [
{
"role": "user",
"content": "Describe a safari through the Serengeti in three or four sentences."
},
{
"role": "assistant",
"content": "Okay, the user wants a description of a safari through the Serengeti in three or four sentences. Let me start by recalling what the Serengeti is known for. It's a vast ecosystem in Tanzania, famous for its wildlife migrations, especially the wildebeest and zebra. I should mention the open savannahs and the Great Migration as key features.\n\nNext, I need to highlight the wildlife. Lions, elephants, giraffes, and maybe mention predators like cheetahs. Also, the acacia trees are iconic there. Including the Maasai people could add cultural depth, showing the interaction between humans and the environment.\n\nThe user might be planning a trip or just curious about the experience. They probably"
}
],
"generated": [
"Okay, the user wants a description of a safari through the Serengeti in three or four sentences. Let me start by recalling what the Serengeti is known for. It's a vast ecosystem in Tanzania, famous for its wildlife migrations, especially the wildebeest and zebra. I should mention the open savannahs and the Great Migration as key features.\n\nNext, I need to highlight the wildlife. Lions, elephants, giraffes, and maybe mention predators like cheetahs. Also, the acacia trees are iconic there. Including the Maasai people could add cultural depth, showing the interaction between humans and the environment.\n\nThe user might be planning a trip or just curious about the experience. They probably"
],
"tokens": [
"<|im_start|>",
"user",
"\n",
"Describe",
" a",
" safari",
" through",
" the",
" Ser",
"en",
"get",
"i",
" in",
" three",
" or",
" four",
" sentences",
".",
"<|im_end|>",
"\n",
"<|im_start|>",
"assistant",
"\n",
"",
"\n\n",
"",
"\n\n",
"Okay",
",",
" the",
" user",
" wants",
" a",
" description",
" of",
" a",
" safari",
" through",
" the",
" Ser",
"en",
"get",
"i",
" in",
" three",
" or",
" four",
" sentences",
".",
" Let",
" me",
" start",
" by",
" recalling",
" what",
" the",
" Ser",
"en",
"get",
"i",
" is",
" known",
" for",
".",
" It",
"'s",
" a",
" vast",
" ecosystem",
" in",
" Tanzania",
",",
" famous",
" for",
" its",
" wildlife",
" migrations",
",",
" especially",
" the",
" wil",
"de",
"be",
"est",
" and",
" z",
"ebra",
".",
" I",
" should",
" mention",
" the",
" open",
" sav",
"annah",
"s",
" and",
" the",
" Great",
" Migration",
" as",
" key",
" features",
".\n\n",
"Next",
",",
" I",
" need",
" to",
" highlight",
" the",
" wildlife",
".",
" Lions",
",",
" elephants",
",",
" gir",
"aff",
"es",
",",
" and",
" maybe",
" mention",
" predators",
" like",
" che",
"et",
"ah",
"s",
".",
" Also",
",",
" the",
" ac",
"acia",
" trees",
" are",
" iconic",
" there",
".",
" Including",
" the",
" Ma",
"as",
"ai",
" people",
" could",
" add",
" cultural",
" depth",
",",
" showing",
" the",
" interaction",
" between",
" humans",
" and",
" the",
" environment",
".\n\n",
"The",
" user",
" might",
" be",
" planning",
" a",
" trip",
" or",
" just",
" curious",
" about",
" the",
" experience",
".",
" They",
" probably"
],
"readouts": [
{
"position": 176,
"token": " probably",
"model_top": [
" want",
" need",
" appreciate",
" expect",
" value",
" seek",
" wants",
" look",
" wanted",
" don"
],
"layers": {
"0": [
" （",
"＿",
"？\n",
"？\n\n",
"（",
" ：",
"＜",
" ，",
"）\r\n",
"？"
],
"1": [
"？\n",
" （",
"？",
"？\n\n",
" ：",
"）\r\n",
"＞",
"＿",
"（",
"％"
],
"2": [
"？\n",
"？",
"？\n\n",
" （",
"）\r\n",
" ：",
"＞",
"：",
"％",
"＿"
],
"3": [
"？\n",
" （",
"？",
" ：",
"（",
" ，",
"：",
"＜",
"）\r\n",
" "
],
"4": [
" （",
" ：",
"？\n",
" ，",
"？",
"％",
"（",
"＃",
"：%",
"：\n"
],
"5": [
"？\n",
" （",
" ：",
" ，",
" Почем",
"？",
"？\n\n",
"：\n",
"hở",
"％"
],
"6": [
"móvel",
" Почем",
"？\n",
"schlü",
"europäische",
"индивид",
"hở",
"чр",
"индив",
"реги"
],
"7": [
"？\n",
"móvel",
" Почем",
"？\n\n",
"hở",
" （",
"＜",
"％",
"europäische",
" ："
],
"8": [
"？\n",
"？",
"：\n",
"＜",
"？\n\n",
"。\r\n",
"）\r\n",
"，\n",
" ",
"＞"
],
"9": [
"？\n",
"＜",
"？",
"？\n\n",
" Почем",
"）\r\n",
"＞",
"％",
"，\n",
"）\n"
],
"10": [
"？\n",
" Почем",
"没什么",
"）\r\n",
"＜",
"？",
"？\n\n",
"。\r\n",
"％",
"都没"
],
"11": [
"没什么",
"？\n",
" Почем",
"＜",
" kids",
"？",
"也没什么",
"没人",
"没钱",
"＞"
],
"12": [
"？\n",
"没什么",
" kids",
"＜",
" nice",
" didn",
" somebody",
" got",
"？",
"％"
],
"13": [
"？\n",
"没什么",
" kids",
" didn",
"？",
"？\n\n",
" nice",
"．",
" got",
" guys"
],
"14": [
"？\n",
"没什么",
"？\n\n",
"？",
"．",
" kids",
" didn",
"＞",
"＜",
"％"
],
"15": [
" didn",
" kids",
"没什么",
" maybe",
" chances",
"．",
"？\n",
" weren",
"’s",
" couldn"
],
"16": [
" kids",
" maybe",
" guys",
"没什么",
" chances",
" didn",
" ’",
" folks",
" savvy",
" got"
],
"17": [
" ",
" ",
"",
"’s",
"’",
"s",
" maybe",
" ‘",
" ",
" ’"
],
"18": [
"",
"’",
"’s",
"s",
" ",
"．",
"’.",
" maybe",
",’",
"’re"
],
"19": [
" maybe",
" guys",
" folks",
"",
" kids",
" chances",
" didn",
"s",
"maybe",
" pretty"
],
"20": [
" maybe",
"’s",
" didn",
"",
"",
" folks",
"’ll",
"’",
"’.",
" hadn"
],
"21": [
" maybe",
"’s",
" didn",
" hoping",
"’ll",
" hopefully",
" hopeful",
" wanting",
" hadn",
"’."
],
"22": [
"’s",
",”",
"?”",
"，“",
",’",
"”?",
" maybe",
"…”",
".”",
"’ll"
],
"23": [
",”",
"?”",
"’s",
",’",
"”?",
"’ll",
"，“",
" didn",
".”",
".\r\n"
],
"24": [
",”",
"’s",
",’",
"?”",
"’ll",
"”?",
"’.",
"感兴趣的",
"，“",
" wanting"
],
"25": [
"?”",
",”",
"”?",
"?”\n\n",
",’",
"想知道",
"’ll",
"感兴趣的",
"’s",
" maybe"
],
"26": [
"想知道",
"希望能",
"想要",
"希望自己",
"想找",
"希望通过",
"都想",
"渴望",
"很想",
"感兴趣的"
],
"27": [
"想知道",
"希望能",
"想要",
"想找",
"渴望",
"很想",
"想去",
"希望自己",
"希望通过",
"感兴趣的"
],
"28": [
"想知道",
"想要",
"想找",
"渴望",
"希望能",
"都想",
"想去",
"向往",
"感兴趣的",
"希望自己"
],
"29": [
"想知道",
"想要",
"想找",
"想去",
"都想",
"渴望",
"感兴趣的",
"希望能",
"感兴趣",
"很想"
],
"30": [
"想知道",
"想要",
"想找",
"想去",
"都想",
"渴望",
" wants",
".want",
" muốn",
" wanted"
],
"31": [
"想知道",
"想要",
"想去",
"都想",
"想找",
" wants",
".want",
" wanted",
"渴望",
" muốn"
],
"32": [
"想要",
"想找",
"想知道",
"都想",
" wants",
"想去",
" want",
".want",
"\twant",
" wanted"
],
"33": [
"想要",
" want",
"想找",
" wants",
"想知道",
"都想",
"\twant",
" wanted",
"want",
".want"
],
"34": [
"想要",
" want",
" wants",
"Want",
"want",
".want",
"想",
" Want",
"\twant",
"想找"
],
"35": [
" want",
"Want",
"want",
" wants",
"想要",
"想找",
" Want",
"\twant",
"想知道",
".want"
],
"36": [
" want",
" wants",
"Want",
"want",
"想找",
"想要",
"\twant",
" Want",
"都想",
".want"
],
"37": [
" want",
" wants",
"want",
" Want",
"Want",
"\twant",
" wanna",
".want",
" wanted",
"想找"
],
"38": [
" want",
" wants",
"want",
" appreciate",
" need",
" Want",
" wanna",
" wanted",
"想找",
" seek"
]
}
}
],
"trajectories": [
{
"word": "yes",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
32822,
26038,
8881,
7464,
25135,
32209,
37766,
34825,
20445,
14805,
34910,
20955,
6026,
5253,
7457,
5768,
3728,
9878,
13693,
11581,
27215,
29692,
45371,
91426,
108174,
72759,
107935,
106642,
109585,
117249,
128885,
129768,
129213,
105068,
37476,
31943,
32906,
14142,
6579
]
},
{
"word": "no",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
51958,
54191,
23750,
13268,
78661,
90285,
91102,
98843,
67970,
52291,
90940,
65362,
17338,
22310,
40432,
8621,
8289,
3875,
7850,
7075,
15175,
19145,
41468,
78081,
100758,
95276,
129807,
141992,
142017,
143815,
144978,
145336,
149107,
146274,
108029,
98022,
91262,
15199,
3247
]
},
{
"word": "feel",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
72315,
57449,
36274,
23619,
33660,
35831,
43667,
35920,
9104,
4601,
7096,
2131,
488,
476,
486,
207,
149,
145,
180,
90,
166,
219,
675,
844,
977,
1181,
1122,
1096,
737,
808,
846,
794,
4422,
2469,
887,
320,
239,
82,
38
]
},
{
"word": "elephant",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
15809,
34054,
10386,
6892,
27937,
58050,
63949,
54861,
18011,
18859,
38416,
23450,
13042,
28143,
66477,
32039,
33341,
16399,
20421,
28222,
20684,
17888,
29180,
51012,
55759,
44317,
87381,
65139,
70747,
78199,
57428,
75445,
75196,
69089,
67307,
108394,
94475,
88139,
49257
]
},
{
"word": "cat",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
11804,
18132,
12379,
9173,
30219,
53206,
54700,
37003,
12451,
11645,
21654,
15507,
4575,
8622,
17777,
5668,
3850,
2244,
2822,
2887,
4775,
10656,
25704,
37654,
47868,
64268,
86518,
102001,
104842,
99215,
88427,
81047,
75745,
86430,
90756,
48978,
37072,
18481,
8757
]
},
{
"word": "sorry",
"position": 176,
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
26618,
7492,
5615,
5160,
16249,
20195,
32520,
27528,
4654,
4159,
6832,
4790,
2727,
1486,
2512,
6671,
6453,
12028,
14327,
8678,
15875,
21335,
14493,
20018,
28360,
17479,
34378,
66203,
80589,
76760,
97897,
89666,
104257,
119954,
44927,
67626,
94389,
56942,
62594
]
}
],
"emergence": {
"position": 176,
"top1": " want",
"layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38
],
"ranks": [
46937,
64206,
28752,
21418,
38971,
58159,
71575,
53366,
34939,
21211,
14555,
4572,
1323,
1401,
1145,
345,
368,
589,
402,
250,
240,
289,
675,
673,
283,
349,
80,
92,
63,
37,
24,
13,
7,
2,
2,
1,
1,
1,
1
]
},
"vanilla": {
"top1_agreement": {
"0": false,
"1": false,
"2": false,
"3": false,
"4": false,
"5": false,
"6": false,
"7": false,
"8": false,
"9": false,
"10": false,
"11": false,
"12": false,
"13": false,
"14": false,
"15": false,
"16": false,
"17": false,
"18": false,
"19": false,
"20": false,
"21": false,
"22": false,
"23": false,
"24": false,
"25": false,
"26": false,
"27": true,
"28": true,
"29": true,
"30": true,
"31": false,
"32": false,
"33": false,
"34": false,
"35": true,
"36": true,
"37": true,
"38": true
},
"trajectories": []
},
"scan": [],
"slice": null,
"film": "film.json",
"extra_md": "Full checkpoint-specific ribbon; exact-prefix segments. See metrics.json and snapshots.json."
}