anlp-a1-ablation

aditya-s-2k24/anlp-a1-ablation

anlp-a1-ablation is a machine learning model from aditya-s-2k24. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.

Encoder-decoder transformers trained from scratch to decrypt binary ciphertext into English plaintext. Five configurations, each differing from C1 by exactly one component.

Downloads · 30 days

Access

Public

Updated Sep 2, 2026

Repo size

3.7 GB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.pt3 GB · 100%

At a glance

License: mit
Access: Public
Created: Sep 2, 2026
Updated: Sep 2, 2026
SHA: 08fe0bed

Model card

From the Hugging Face model README

ANLP Assignment 1 - Transformer Ablation

Encoder-decoder transformers trained from scratch to decrypt binary ciphertext into English plaintext. Five configurations, each differing from C1 by exactly one component.

config	positional	attention	norm	tokenization
C1	Sinusoidal	MHA	LayerNorm	subword BPE
C2	RoPE	MHA	LayerNorm	subword BPE
C3	Sinusoidal	GQA (2 groups)	LayerNorm	subword BPE
C4	Sinusoidal	MHA	RMSNorm	subword BPE
C5	Sinusoidal	MHA	LayerNorm	BLT (byte-level)

Files

C1.pt ... C5.pt - model state dicts
800_bittok_brown_cipher.json - source tokenizer (802 entries)
2160_textok_brown_plain.json - target tokenizer (2164 entries)
results.json - test metrics for all five configs

C1-C4 require both tokenizers. C5 is token-free and uses raw bytes.

Test results

{
  "C1_s0": {
    "bit_accuracy": 63.89837534417451,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 187.791933418694,
    "bleu": 0.3831980304270051,
    "rougeL": 13.77842180264974,
    "val_loss": 5.036809729819902,
    "params": 46769268,
    "train_peak_gpu_gb": 5.814194688,
    "decode_peak_gpu_gb": 1.062931968,
    "epoch_seconds": 201.19917788505555,
    "train_seconds_total": 4023.983557701111,
    "decode_seconds": 667.8383529186249,
    "decode_ms_per_seq": 427.55336294406203,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C1_s1": {
    "bit_accuracy": 63.942299504920385,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 198.6978233034571,
    "bleu": 0.38677151343342,
    "rougeL": 13.705224328365462,
    "val_loss": 5.063231503331493,
    "params": 46769268,
    "train_peak_gpu_gb": 5.809763328,
    "decode_peak_gpu_gb": 1.06079488,
    "epoch_seconds": 210.26583820581436,
    "train_seconds_total": 4205.316764116287,
    "decode_seconds": 702.4448430538177,
    "decode_ms_per_seq": 449.7086063084621,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C1_s2": {
    "bit_accuracy": 63.93556128905492,
    "sequence_accuracy": 0.0,
    "levenshtein": 205.48271446862995,
    "bleu": 0.35829749412404477,
    "rougeL": 11.92432531264349,
    "val_loss": 5.076420719496937,
    "params": 46769268,
    "train_peak_gpu_gb": 5.80918784,
    "decode_peak_gpu_gb": 1.060191232,
    "epoch_seconds": 208.71450798511506,
    "train_seconds_total": 4174.290159702301,
    "decode_seconds": 704.412652015686,
    "decode_ms_per_seq": 450.9684071803368,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C2_s0": {
    "bit_accuracy": 65.3580897763381,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 160.2797695262484,
    "bleu": 0.6152015254707937,
    "rougeL": 14.964772489523774,
    "val_loss": 4.91474067884296,
    "params": 46769268,
    "train_peak_gpu_gb": 5.790985216,
    "decode_peak_gpu_gb": 1.036978176,
    "epoch_seconds": 205.03780821561813,
    "train_seconds_total": 4100.756164312363,
    "decode_seconds": 337.93311643600464,
    "decode_ms_per_seq": 216.34642537516302,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C2_s1": {
    "bit_accuracy": 65.494237320184,
    "sequence_accuracy": 0.0,
    "levenshtein": 158.2368758002561,
    "bleu": 0.6288094434402884,
    "rougeL": 16.887806987359713,
    "val_loss": 4.860477296296365,
    "params": 46769268,
    "train_peak_gpu_gb": 5.79356928,
    "decode_peak_gpu_gb": 1.039369216,
    "epoch_seconds": 214.60391768217087,
    "train_seconds_total": 4292.078353643417,
    "decode_seconds": 303.5473804473877,
    "decode_ms_per_seq": 194.33250988949277,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C2_s2": {
    "bit_accuracy": 65.71847677716195,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 161.64532650448143,
    "bleu": 0.5765812040980112,
    "rougeL": 14.726260061032582,
    "val_loss": 4.916989824601374,
    "params": 46769268,
    "train_peak_gpu_gb": 5.791143424,
    "decode_peak_gpu_gb": 0.988094464,
    "epoch_seconds": 215.044200026989,
    "train_seconds_total": 4300.88400053978,
    "decode_seconds": 291.3824396133423,
    "decode_ms_per_seq": 186.54445557832412,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C3_s0": {
    "bit_accuracy": 64.06945891392108,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 227.601152368758,
    "bleu": 0.2588824992999517,
    "rougeL": 12.123752783404163,
    "val_loss": 5.062767298755001,
    "params": 39677556,
    "train_peak_gpu_gb": 5.69828608,
    "decode_peak_gpu_gb": 0.950312448,
    "epoch_seconds": 192.7562213897705,
    "train_seconds_total": 3855.12442779541,
    "decode_seconds": 645.5705280303955,
    "decode_ms_per_seq": 413.2973931052468,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C3_s1": {
    "bit_accuracy": 63.95422436916653,
    "sequence_accuracy": 0.0,
    "levenshtein": 195.8245838668374,
    "bleu": 0.4609379598330146,
    "rougeL": 13.024580987738071,
    "val_loss": 5.067461036860475,
    "params": 39677556,
    "train_peak_gpu_gb": 5.699069952,
    "decode_peak_gpu_gb": 0.94782976,
    "epoch_seconds": 204.82737295627595,
    "train_seconds_total": 4096.547459125519,
    "decode_seconds": 496.8684492111206,
    "decode_ms_per_seq": 318.09759872670975,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C3_s2": {
    "bit_accuracy": 63.96199237166387,
    "sequence_accuracy": 0.0,
    "levenshtein": 187.6235595390525,
    "bleu": 0.347600695264389,
    "rougeL": 13.382627690824124,
    "val_loss": 5.0575279543446365,
    "params": 39677556,
    "train_peak_gpu_gb": 5.7004544,
    "decode_peak_gpu_gb": 0.947551232,
    "epoch_seconds": 199.84601358175277,
    "train_seconds_total": 3996.9202716350555,
    "decode_seconds": 590.325471162796,
    "decode_ms_per_seq": 377.92923890063764,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C4_s0": {
    "bit_accuracy": 64.0224222051626,
    "sequence_accuracy": 0.12804097311139565,
    "levenshtein": 187.0147247119078,
    "bleu": 0.4022859374705437,
    "rougeL": 13.788644159270168,
    "val_loss": 5.046384166488765,
    "params": 46752884,
    "train_peak_gpu_gb": 5.324237312,
    "decode_peak_gpu_gb": 1.060670976,
    "epoch_seconds": 198.31388070583344,
    "train_seconds_total": 3966.2776141166687,
    "decode_seconds": 639.8467614650726,
    "decode_ms_per_seq": 409.6330099008147,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C4_s1": {
    "bit_accuracy": 63.74831967254383,
    "sequence_accuracy": 0.0,
    "levenshtein": 191.32970550576184,
    "bleu": 0.298144565800722,
    "rougeL": 12.915040428110235,
    "val_loss": 5.063979824953497,
    "params": 46752884,
    "train_peak_gpu_gb": 5.327401984,
    "decode_peak_gpu_gb": 1.061006336,
    "epoch_seconds": 208.28824691772462,
    "train_seconds_total": 4165.764938354492,
    "decode_seconds": 687.9825496673584,
    "decode_ms_per_seq": 440.44977571533826,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C4_s2": {
    "bit_accuracy": 64.39403232521485,
    "sequence_accuracy": 0.0,
    "levenshtein": 192.56274007682458,
    "bleu": 0.4572279342683988,
    "rougeL": 12.588028044639158,
    "val_loss": 5.073893132307736,
    "params": 46752884,
    "train_peak_gpu_gb": 5.326587904,
    "decode_peak_gpu_gb": 1.061219328,
    "epoch_seconds": 205.61277594566346,
    "train_seconds_total": 4112.255518913269,
    "decode_seconds": 663.1803505420685,
    "decode_ms_per_seq": 424.57128715881464,
    "src_len_mean": 114.26600985221675,
    "src_len_max": 155,
    "tgt_len_mean": 62.00985221674877,
    "tgt_len_max": 97
  },
  "C5_s0": {
    "bit_accuracy": 66.99002868240039,
    "sequence_accuracy": 0.0,
    "levenshtein": 164.4122919334187,
    "val_loss": 1.6322713189138296,
    "params": 57151747,
    "train_peak_gpu_gb": 7.808239616,
    "decode_peak_gpu_gb": 1.33983744,
    "epoch_seconds": 114.09861674308777,
    "train_seconds_total": 2281.9723348617554,
    "decode_seconds": 1039.3023602962494,
    "decode_ms_per_seq": 665.3664278465105,
    "src_len_mean": 204.0935960591133,
    "src_len_max": 256,
    "tgt_len_mean": 205.0935960591133,
    "tgt_len_max": 257
  },
  "C5_s1": {
    "bit_accuracy": 67.33456762374041,
    "sequence_accuracy": 0.0,
    "levenshtein": 164.83802816901408,
    "val_loss": 1.6389507850523148,
    "params": 57151747,
    "train_peak_gpu_gb": 7.807245824,
    "decode_peak_gpu_gb": 1.339379712,
    "epoch_seconds": 113.65547159910201,
    "train_seconds_total": 2273.1094319820404,
    "decode_seconds": 1048.301531791687,
    "decode_ms_per_seq": 671.1277412238713,
    "src_len_mean": 204.0935960591133,
    "src_len_max": 256,
    "tgt_len_mean": 205.0935960591133,
    "tgt_len_max": 257
  },
  "C5_s2": {
    "bit_accuracy": 67.11330848286109,
    "sequence_accuracy": 0.0,
    "levenshtein": 165.42061459667093,
    "val_loss": 1.6107058093837328,
    "params": 57151747,
    "train_peak_gpu_gb": 7.806793216,
    "decode_peak_gpu_gb": 1.33929216,
    "epoch_seconds": 113.89586465358734,
    "train_seconds_total": 2277.917293071747,
    "decode_seconds": 1049.2813975811005,
    "decode_ms_per_seq": 671.7550560698467,
    "src_len_mean": 204.0935960591133,
    "src_len_max": 256,
    "tgt_len_mean": 205.0935960591133,
    "tgt_len_max": 257
  }
}

License: mit
Created: Sep 2, 2026
Updated: Sep 2, 2026