{
 "id": "u6-amp-early-a0008-g4b",
 "title": "Unit 6 \u00b7 Amplify typo register @ early (\u03b1=0.0075) \u00b7 gemma-4b",
 "unit": "6",
 "created": "2026-07-09T23:03:05",
 "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": 0.0075
  }
 },
 "extra_md": null,
 "conversation": [
  {
   "role": "user",
   "content": "Describe the water cycle in two sentences."
  },
  {
   "role": "assistant",
   "content": "The water cycle is a continuous process where water changes states \u2013 it\u2019s when water evaporates from bodies of water like oceans and lakes, then rises into the atmosphere as water vapor, and then cools down and falls back down as rain or snow."
  }
 ],
 "generated": [
  "The water cycle is a continuous process where water changes states \u2013 it\u2019s when water evaporates from bodies of water like oceans and lakes, then rises into the atmosphere as water vapor, and then cools down and falls back down as rain or snow."
 ],
 "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",
  " continuous",
  " process",
  " where",
  " water",
  " changes",
  " states",
  " \u2013",
  " it",
  "\u2019",
  "s",
  " when",
  " water",
  " evaporates",
  " from",
  " bodies",
  " of",
  " water",
  " like",
  " oceans",
  " and",
  " lakes",
  ",",
  " then",
  " rises",
  " into",
  " the",
  " atmosphere",
  " as",
  " water",
  " vapor",
  ",",
  " and",
  " then",
  " cools",
  " down",
  " and",
  " falls",
  " back",
  " down",
  " as",
  " rain",
  " or",
  " snow",
  ".",
  "<end_of_turn>",
  "\n"
 ],
 "readouts": [
  {
   "position": 66,
   "token": "<end_of_turn>",
   "model_top": [
    "<end_of_turn>",
    "\n",
    "Display",
    " ",
    "Report",
    "**",
    "pe",
    "\n\n"
   ],
   "layers": {
    "0": [
     "<end_of_turn>",
     "  ",
     " (\\\"",
     "   ",
     " \\\"",
     " \\[",
     " ;-",
     " \u00d0"
    ],
    "1": [
     "<end_of_turn>",
     "  ",
     "\uff8a",
     ",\u2026",
     "\u066c",
     "\ufe51",
     ")\u2026",
     ").["
    ],
    "2": [
     "  ",
     "\ufffd",
     ",",
     "<end_of_turn>",
     ").[",
     "\u066c",
     "`,`",
     "\uff8a"
    ],
    "3": [
     "  ",
     "   ",
     ",",
     " seperate",
     ",.",
     " \u0641\u0649",
     "\ufffd",
     " alot"
    ],
    "4": [
     "  ",
     "   ",
     "..",
     " alot",
     "!!",
     "!.",
     ",.",
     " whilst"
    ],
    "5": [
     "  ",
     " whilst",
     " \u0641\u0649",
     "\ufffd",
     " \u0627\u0644\u062a\u0649",
     "   ",
     " realising",
     " seperate"
    ],
    "6": [
     "  ",
     " whilst",
     " seperate",
     " alot",
     " seper",
     " amongst",
     " then",
     " which"
    ],
    "7": [
     "  ",
     " whilst",
     " alot",
     " seperate",
     "Whilst",
     " Whilst",
     " amongst",
     " seper"
    ],
    "8": [
     " whilst",
     " alot",
     "  ",
     " seperate",
     " everytime",
     "Whilst",
     " Whilst",
     " seper"
    ],
    "9": [
     " whilst",
     " alot",
     " seperate",
     "  ",
     "Whilst",
     " Whilst",
     " everytime",
     " seper"
    ],
    "10": [
     " whilst",
     " alot",
     "Whilst",
     "  ",
     " seperate",
     " Whilst",
     " everytime",
     "!!"
    ],
    "11": [
     " whilst",
     "Whilst",
     "!!!",
     " alot",
     "!!",
     " Whilst",
     "?!",
     "!!!!"
    ],
    "12": [
     "!!!",
     "!!",
     "?!",
     "!?",
     "!!!!",
     "?!?!",
     "Whilst",
     " alot"
    ],
    "13": [
     "<end_of_turn>",
     "!!!",
     ".??",
     "!).",
     "?!",
     "!?",
     ")!",
     "!!"
    ],
    "14": [
     "<end_of_turn>",
     ".??",
     ",((",
     ").[",
     ".[/",
     "!).",
     "![",
     ",\u200e"
    ],
    "15": [
     "<end_of_turn>",
     ").[",
     ".[/",
     ",\u200e",
     ".[",
     "![",
     "\u200e",
     "!)."
    ],
    "16": [
     "<end_of_turn>",
     ").[",
     ".[/",
     "!).",
     ".).",
     ",\u200e",
     ".??",
     " :)."
    ],
    "17": [
     "<end_of_turn>",
     ").[",
     "!).",
     " :).",
     ")!",
     "\udb82\udc55",
     "?).",
     ".))."
    ],
    "18": [
     "<end_of_turn>",
     "\udb82\udc55",
     " :).",
     ")!",
     "!).",
     "<unused42>",
     ")`,",
     " \u0633\u0648\u0627\u0644"
    ],
    "19": [
     "<end_of_turn>",
     "\udb82\udc55",
     "absch",
     "explanation",
     " \u0633\u0648\u0627\u0644",
     " Antwort",
     " explicando",
     "<unused42>"
    ],
    "20": [
     "<end_of_turn>",
     "<eos>",
     "Hopefully",
     "```",
     "\udb82\udc55",
     "explanation",
     "</s>",
     "hopefully"
    ],
    "21": [
     "<end_of_turn>",
     "Hopefully",
     "hopefully",
     "ChatGPT",
     "explanation",
     "\udb82\udc55",
     "hope",
     ",\u200e"
    ],
    "22": [
     "<end_of_turn>",
     "Hopefully",
     "ChatGPT",
     "hope",
     "answer",
     "respuesta",
     "\u200e",
     "feedback"
    ],
    "23": [
     "<end_of_turn>",
     "feedback",
     "ChatGPT",
     "Would",
     "\ud83d\ude0a",
     ",\u200e",
     "respuesta",
     "Danke"
    ],
    "24": [
     "<end_of_turn>",
     "Would",
     "ChatGPT",
     "answer",
     "\ud83d\ude0a",
     " Would",
     " r\u00e9ponse",
     "Reply"
    ],
    "25": [
     "<end_of_turn>",
     "Would",
     "Answer",
     "Reply",
     "Disclaimer",
     " Would",
     "Why",
     " Antwort"
    ],
    "26": [
     "<end_of_turn>",
     "Display",
     "Disclaimer",
     "display",
     " Display",
     " excuse",
     " display",
     "Would"
    ],
    "27": [
     "<end_of_turn>",
     "<eos>",
     "Display",
     "Disclaimer",
     "Do",
     " Display",
     " excuse",
     "Please"
    ],
    "28": [
     "<end_of_turn>",
     "Display",
     "Disclaimer",
     "Do",
     " Do",
     " Display",
     "Show",
     " Make"
    ],
    "29": [
     "<end_of_turn>",
     "Display",
     "Show",
     " r\u00e9ponse",
     " Show",
     "Create",
     " Display",
     "Please"
    ],
    "30": [
     "<end_of_turn>",
     "Show",
     " Show",
     "Display",
     "\n",
     " \u0627\u0644\u0645\u0632\u064a\u062f",
     "<eos>",
     "Please"
    ],
    "31": [
     "<end_of_turn>",
     "\n",
     "  ",
     " \u0627\u0644\u0645\u0632\u064a\u062f",
     "<start_of_image>",
     " Display",
     "<eos>",
     "\u66f4\u591a"
    ],
    "32": [
     "<end_of_turn>",
     "\n",
     " \u0627\u0644\u0645\u0632\u064a\u062f",
     "\u66f4\u591a",
     "  ",
     "Display",
     "additional",
     " "
    ]
   }
  }
 ],
 "trajectories": [
  {
   "word": "anyways",
   "position": 66,
   "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": [
    83211,
    28492,
    1544,
    476,
    298,
    285,
    91,
    45,
    11,
    13,
    12,
    16,
    30,
    107,
    400,
    218,
    280,
    2427,
    3632,
    6959,
    3158,
    2938,
    4536,
    13560,
    12140,
    8788,
    9385,
    6394,
    13643,
    16133,
    14979,
    8863,
    1954
   ]
  },
  {
   "word": "alot",
   "position": 66,
   "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": [
    8917,
    379,
    30,
    8,
    4,
    18,
    4,
    3,
    2,
    2,
    2,
    4,
    9,
    94,
    204,
    996,
    186,
    11943,
    16629,
    78146,
    41030,
    64163,
    163626,
    168634,
    164919,
    143841,
    81190,
    105433,
    91615,
    66541,
    27332,
    18029,
    29471
   ]
  },
  {
   "word": "yummy",
   "position": 66,
   "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": [
    187350,
    75706,
    39031,
    9153,
    6686,
    12862,
    4565,
    3695,
    1295,
    1343,
    1876,
    1419,
    2093,
    2773,
    20686,
    20233,
    4920,
    8136,
    2447,
    8491,
    6959,
    16785,
    10178,
    3024,
    3157,
    8560,
    15431,
    6136,
    7440,
    12920,
    16492,
    13034,
    16820
   ]
  },
  {
   "word": "kinda",
   "position": 66,
   "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": [
    114789,
    26864,
    3603,
    1234,
    764,
    3948,
    755,
    1181,
    258,
    356,
    618,
    1174,
    1815,
    13999,
    95891,
    82074,
    36671,
    110994,
    91437,
    152665,
    99788,
    207028,
    228418,
    149715,
    93957,
    67660,
    56104,
    36685,
    62495,
    59241,
    51471,
    39828,
    39556
   ]
  },
  {
   "word": "whilst",
   "position": 66,
   "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": [
    11535,
    515,
    20,
    14,
    8,
    2,
    2,
    2,
    1,
    1,
    1,
    1,
    7,
    55,
    13,
    29,
    44,
    2221,
    2118,
    3646,
    574,
    2187,
    12968,
    35400,
    38006,
    25189,
    28286,
    44599,
    37544,
    25892,
    13388,
    4216,
    5765
   ]
  },
  {
   "word": "luckily",
   "position": 66,
   "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": [
    138084,
    56955,
    3874,
    1113,
    937,
    566,
    265,
    168,
    143,
    150,
    158,
    228,
    322,
    856,
    1571,
    627,
    657,
    740,
    768,
    881,
    355,
    114,
    279,
    965,
    448,
    1220,
    2939,
    8273,
    17108,
    45577,
    31077,
    20303,
    20481
   ]
  }
 ],
 "emergence": {
  "position": 66,
  "top1": "<end_of_turn>",
  "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": [
   1,
   1,
   4,
   95,
   7931,
   2194,
   15648,
   7000,
   13711,
   1497,
   336,
   29,
   24,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1,
   1
  ]
 },
 "scan": [],
 "slice": null
}