{
 "id": "u6-amp-early-a0015-g4b",
 "title": "Unit 6 \u00b7 Amplify typo register @ early (\u03b1=0.015) \u00b7 gemma-4b",
 "unit": "6",
 "created": "2026-07-09T23:02:59",
 "model": {
  "name": "gemma-4b",
  "hf_id": "google/gemma-3-4b-it",
  "quant": null,
  "n_layers": 34
 },
 "lens": {
  "repo": "neuronpedia/jacobian-lens",
  "file": "gemma-3-4b-it/jlens/Salesforce-wikitext/gemma-3-4b-it_jacobian_lens.pt"
 },
 "params": {
  "chat": null,
  "max_new": 60,
  "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": [
    2,
    3,
    4,
    5,
    6,
    7,
    8
   ],
   "mode": "amplify",
   "alpha": 0.015
  }
 },
 "extra_md": null,
 "conversation": [
  {
   "role": "user",
   "content": "Describe the water cycle in two sentences."
  },
  {
   "role": "assistant",
   "content": "The water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a"
  }
 ],
 "generated": [
  "The water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a model of two sentences.\n\nThe water cycle is a"
 ],
 "tokens": [
  "<bos>",
  "<start_of_turn>",
  "user",
  "\n",
  "Describe",
  " the",
  " water",
  " cycle",
  " in",
  " two",
  " sentences",
  ".",
  "<end_of_turn>",
  "\n",
  "<start_of_turn>",
  "model",
  "\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  " model",
  " of",
  " two",
  " sentences",
  ".",
  "\n\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  " model",
  " of",
  " two",
  " sentences",
  ".",
  "\n\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  " model",
  " of",
  " two",
  " sentences",
  ".",
  "\n\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  " model",
  " of",
  " two",
  " sentences",
  ".",
  "\n\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  " model",
  " of",
  " two",
  " sentences",
  ".",
  "\n\n",
  "The",
  " water",
  " cycle",
  " is",
  " a",
  "<end_of_turn>",
  "\n"
 ],
 "readouts": [
  {
   "position": 77,
   "token": "<end_of_turn>",
   "model_top": [
    "\n\n",
    "\n",
    ".",
    "model",
    " ",
    " model",
    "I",
    ","
   ],
   "layers": {
    "0": [
     "<end_of_turn>",
     "  ",
     "   ",
     " (\\\"",
     " \u00d0",
     " \u00c3",
     " \ufffd",
     " ;-"
    ],
    "1": [
     "<end_of_turn>",
     "  ",
     "\uff8a",
     "\ufe51",
     "\u066c",
     ",\u2026",
     ").[",
     ",(("
    ],
    "2": [
     "  ",
     ",",
     " seperate",
     " \u0641\u0649",
     " alot",
     " \u0627\u0644\u062a\u0649",
     " seper",
     " whilst"
    ],
    "3": [
     "  ",
     " alot",
     "!!",
     ",",
     ".",
     " seperate",
     "!.",
     ".."
    ],
    "4": [
     "  ",
     " alot",
     "..",
     "!!",
     " thats",
     "!.",
     " whilst",
     " seperate"
    ],
    "5": [
     "  ",
     " whilst",
     " \u0641\u0649",
     " \u0627\u0644\u062a\u0649",
     " utilising",
     "Whilst",
     " Whilst",
     " alot"
    ],
    "6": [
     "  ",
     " whilst",
     " alot",
     " seperate",
     " amongst",
     " which",
     " the",
     " however"
    ],
    "7": [
     " whilst",
     "  ",
     " alot",
     " seperate",
     " amongst",
     " utilising",
     " Whilst",
     " which"
    ],
    "8": [
     " whilst",
     " alot",
     " seperate",
     "  ",
     " amongst",
     " Whilst",
     " everytime",
     "Whilst"
    ],
    "9": [
     " alot",
     " whilst",
     " seperate",
     " everytime",
     " anyways",
     " showcasing",
     " impactful",
     " Whilst"
    ],
    "10": [
     " alot",
     " whilst",
     " anyways",
     " everytime",
     " showcasing",
     " impactful",
     " seperate",
     "Whilst"
    ],
    "11": [
     " alot",
     " showcasing",
     " impactful",
     " whilst",
     " everytime",
     " anyways",
     " Anyways",
     "Whilst"
    ],
    "12": [
     " impactful",
     " showcasing",
     " alot",
     " whilst",
     " anyways",
     " everytime",
     " Anyways",
     "Whilst"
    ],
    "13": [
     "Whilst",
     " alot",
     " anyways",
     "\uf04a",
     "!!!",
     " everytime",
     " whilst",
     " Anyways"
    ],
    "14": [
     " whilst",
     " anyways",
     " alot",
     "Whilst",
     " haha",
     "!).",
     " Anyways",
     " basically"
    ],
    "15": [
     "\uf04a",
     "<end_of_turn>",
     ")).",
     "!).",
     " :).",
     ")!",
     ").[",
     " haha"
    ],
    "16": [
     "!).",
     "\uf04a",
     ")!",
     "?).",
     "<end_of_turn>",
     ")).",
     ")?",
     ").["
    ],
    "17": [
     "!).",
     " everytime",
     " haha",
     ")!",
     ")).",
     ").",
     "?).",
     ".)."
    ],
    "18": [
     "!).",
     ")!",
     "?).",
     " haha",
     ".).",
     ")).",
     ").",
     "!."
    ],
    "19": [
     "\uf04a",
     "<end_of_turn>",
     "?).",
     "!).",
     ")?",
     ")!",
     "\udb82\udc55",
     "))."
    ],
    "20": [
     "<end_of_turn>",
     "your",
     "you",
     "infinity",
     "answer",
     "\uf04a",
     "whatever",
     " haha"
    ],
    "21": [
     "your",
     "you",
     "<end_of_turn>",
     "whatever",
     "etc",
     "infinity",
     "anything",
     "questions"
    ],
    "22": [
     "your",
     "you",
     "questions",
     "whatever",
     "answer",
     "etc",
     "because",
     "everything"
    ],
    "23": [
     "your",
     "you",
     ").",
     "etc",
     "something",
     ")!",
     "and",
     "questions"
    ],
    "24": [
     "you",
     ").",
     "etc",
     "question",
     "questions",
     "your",
     "hope",
     "answer"
    ],
    "25": [
     "question",
     "questions",
     "you",
     "answer",
     ").",
     "your",
     "opportunity",
     "hope"
    ],
    "26": [
     "I",
     "\n\n",
     "question",
     "and",
     "answer",
     "very",
     "questions",
     "\n"
    ],
    "27": [
     "\n\n",
     "\n",
     "I",
     "\n\n\n",
     "\n\n\n\n",
     "and",
     "Do",
     "question"
    ],
    "28": [
     "\n\n",
     "\n",
     "I",
     "\n\n\n",
     ".",
     "\n\n\n\n",
     "a",
     "and"
    ],
    "29": [
     "\n\n",
     "\n",
     "model",
     " model",
     "\n\n\n",
     " .",
     ".",
     "Model"
    ],
    "30": [
     "\n\n",
     "\n",
     " model",
     "model",
     " .",
     ".",
     "Model",
     " Model"
    ],
    "31": [
     "\n\n",
     "\n",
     " model",
     "model",
     ".",
     " .",
     " Model",
     " "
    ],
    "32": [
     "\n\n",
     "\n",
     ".",
     " ",
     "  ",
     " .",
     "2",
     ","
    ]
   }
  }
 ],
 "trajectories": [
  {
   "word": "anyways",
   "position": 77,
   "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
   ],
   "ranks": [
    102553,
    26213,
    419,
    177,
    131,
    115,
    36,
    25,
    10,
    5,
    3,
    6,
    5,
    3,
    2,
    17,
    30,
    17,
    37,
    134,
    132,
    322,
    1078,
    1276,
    2739,
    3327,
    6077,
    6181,
    8796,
    5501,
    7242,
    4733,
    3077
   ]
  },
  {
   "word": "alot",
   "position": 77,
   "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
   ],
   "ranks": [
    18367,
    291,
    5,
    2,
    2,
    8,
    3,
    3,
    2,
    1,
    1,
    1,
    3,
    2,
    3,
    68,
    236,
    24,
    59,
    887,
    747,
    1385,
    6933,
    1742,
    4554,
    4173,
    4031,
    7041,
    9464,
    2693,
    2406,
    2414,
    2369
   ]
  },
  {
   "word": "yummy",
   "position": 77,
   "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
   ],
   "ranks": [
    209034,
    98247,
    12504,
    2802,
    2301,
    4714,
    4638,
    4151,
    1412,
    723,
    576,
    545,
    747,
    863,
    531,
    1045,
    1083,
    1389,
    1196,
    4436,
    9015,
    11083,
    19972,
    18288,
    21087,
    20133,
    36755,
    35547,
    48689,
    43415,
    38765,
    36472,
    39404
   ]
  },
  {
   "word": "kinda",
   "position": 77,
   "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
   ],
   "ranks": [
    137200,
    30117,
    616,
    276,
    166,
    530,
    282,
    446,
    105,
    60,
    58,
    76,
    123,
    531,
    476,
    4173,
    7609,
    2443,
    2669,
    35025,
    15973,
    24767,
    41462,
    13818,
    16403,
    17993,
    30736,
    25025,
    28650,
    14208,
    14331,
    10958,
    8335
   ]
  },
  {
   "word": "whilst",
   "position": 77,
   "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
   ],
   "ranks": [
    16329,
    471,
    7,
    15,
    7,
    2,
    2,
    1,
    1,
    2,
    2,
    4,
    4,
    1,
    1,
    22,
    89,
    16,
    54,
    276,
    272,
    642,
    1288,
    1896,
    1922,
    3988,
    4670,
    4054,
    3194,
    5270,
    2869,
    2254,
    2039
   ]
  },
  {
   "word": "luckily",
   "position": 77,
   "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
   ],
   "ranks": [
    145914,
    54025,
    1262,
    746,
    532,
    426,
    311,
    230,
    139,
    104,
    83,
    123,
    113,
    228,
    319,
    289,
    400,
    255,
    778,
    1222,
    872,
    1176,
    445,
    368,
    287,
    553,
    1071,
    1761,
    2651,
    5940,
    4019,
    2778,
    4268
   ]
  }
 ],
 "emergence": {
  "position": 77,
  "top1": "\n\n",
  "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
  ],
  "ranks": [
   585,
   256,
   128,
   525,
   2517,
   4781,
   42865,
   29957,
   59786,
   45226,
   123896,
   150516,
   189055,
   262013,
   261627,
   262172,
   262202,
   262202,
   262202,
   262198,
   259933,
   260908,
   210314,
   15321,
   3781,
   44,
   2,
   1,
   1,
   1,
   1,
   1,
   1
  ]
 },
 "scan": [],
 "slice": null
}