Gorffenaf/July 2024

added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "</s>": 44
+}

all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 3.8714672861014323,
+    "eval_loss": 0.38884833455085754,
+    "eval_runtime": 156.3404,
+    "eval_samples": 5656,
+    "eval_samples_per_second": 36.177,
+    "eval_steps_per_second": 4.522,
+    "eval_wer": 0.2884883888879973,
+    "total_flos": 2.249387574100498e+19,
+    "train_loss": 0.15996522521972656,
+    "train_runtime": 19346.8732,
+    "train_samples": 41326,
+    "train_samples_per_second": 8.27,
+    "train_steps_per_second": 1.034
+}

config.json

@@ -1,6 +1,7 @@
 {
   "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
-  "activation_dropout": 0.055,
+  "activation_dropout": 0.0,
+  "adapter_attn_dim": null,
   "adapter_kernel_size": 3,
   "adapter_stride": 2,
   "add_adapter": false,
@@ -8,7 +9,7 @@
   "architectures": [
     "Wav2Vec2ForCTC"
   ],
-  "attention_dropout": 0.055,
+  "attention_dropout": 0.0,
   "bos_token_id": 1,
   "classifier_proj_size": 256,
   "codevector_dim": 768,
@@ -49,16 +50,16 @@
   "feat_extract_activation": "gelu",
   "feat_extract_dropout": 0.0,
   "feat_extract_norm": "layer",
-  "feat_proj_dropout": 0.04,
+  "feat_proj_dropout": 0.0,
   "feat_quantizer_dropout": 0.0,
   "final_dropout": 0.0,
   "hidden_act": "gelu",
-  "hidden_dropout": 0.047,
+  "hidden_dropout": 0.0,
   "hidden_size": 1024,
   "initializer_range": 0.02,
   "intermediate_size": 4096,
   "layer_norm_eps": 1e-05,
-  "layerdrop": 0.041,
+  "layerdrop": 0.0,
   "mask_channel_length": 10,
   "mask_channel_min_space": 1,
   "mask_channel_other": 0.0,
@@ -71,7 +72,7 @@
   "mask_time_min_masks": 2,
   "mask_time_min_space": 1,
   "mask_time_other": 0.0,
-  "mask_time_prob": 0.082,
+  "mask_time_prob": 0.05,
   "mask_time_selection": "static",
   "model_type": "wav2vec2",
   "num_adapter_layers": 3,
@@ -84,7 +85,7 @@
   "num_hidden_layers": 24,
   "num_negatives": 100,
   "output_hidden_size": 1024,
-  "pad_token_id": 46,
+  "pad_token_id": 42,
   "proj_codevector_dim": 768,
   "tdnn_dilation": [
     1,
@@ -108,8 +109,8 @@
     1
   ],
   "torch_dtype": "float32",
-  "transformers_version": "4.19.2",
+  "transformers_version": "4.40.2",
   "use_weighted_layer_sum": false,
-  "vocab_size": 47,
+  "vocab_size": 46,
   "xvector_output_dim": 512
 }

eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.8714672861014323,
+    "eval_loss": 0.38884833455085754,
+    "eval_runtime": 156.3404,
+    "eval_samples": 5656,
+    "eval_samples_per_second": 36.177,
+    "eval_steps_per_second": 4.522,
+    "eval_wer": 0.2884883888879973
+}

kenlm/en/kenlm.tar.gz

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cf2d457a6e31282c1e3f0973ae03d0e46aad7b0f9b35ea8fb927692299036189
-size 909740909

kenlm/cy/kenlm.tar.gz → model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09220ddcb7b283133757379f6a6339b2b8fc542887e49273815f97e56fcd7a8e
-size 254221337
+oid sha256:ccf348c9cd5b591752a4066f0fac73c1bca7151275cd9c3703a0b83922ba2564
+size 1261996080

preprocessor_config.json

@@ -3,7 +3,7 @@
   "feature_extractor_type": "Wav2Vec2FeatureExtractor",
   "feature_size": 1,
   "padding_side": "right",
-  "padding_value": 0.0,
+  "padding_value": 0,
   "processor_class": "Wav2Vec2Processor",
   "return_attention_mask": true,
   "sampling_rate": 16000

pytorch_model.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f35898183d31e3c5e0a3648ef3900ecf8672ab8e45737b0526585f19353e3af2
-size 1262091377

special_tokens_map.json

@@ -1 +1,30 @@
-{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]"}
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": true,
+    "normalized": false,
+    "rstrip": true,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": true,
+    "normalized": false,
+    "rstrip": true,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -1 +1,48 @@
-{"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "replace_word_delimiter_char": " ", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2Processor"}
+{
+  "added_tokens_decoder": {
+    "41": {
+      "content": "[UNK]",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "42": {
+      "content": "[PAD]",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "43": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "44": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "target_lang": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "unk_token": "[UNK]",
+  "word_delimiter_token": "|"
+}

train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.8714672861014323,
+    "total_flos": 2.249387574100498e+19,
+    "train_loss": 0.15996522521972656,
+    "train_runtime": 19346.8732,
+    "train_samples": 41326,
+    "train_samples_per_second": 8.27,
+    "train_steps_per_second": 1.034
+}

trainer_state.json

@@ -0,0 +1,2110 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.8714672861014323,
+  "eval_steps": 100,
+  "global_step": 20000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.019357336430507164,
+      "eval_loss": 3.5565404891967773,
+      "eval_runtime": 151.5266,
+      "eval_samples_per_second": 37.327,
+      "eval_steps_per_second": 4.666,
+      "eval_wer": 1.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.03871467286101433,
+      "eval_loss": 3.0301756858825684,
+      "eval_runtime": 150.582,
+      "eval_samples_per_second": 37.561,
+      "eval_steps_per_second": 4.695,
+      "eval_wer": 1.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.05807200929152149,
+      "eval_loss": 2.9460911750793457,
+      "eval_runtime": 148.9065,
+      "eval_samples_per_second": 37.984,
+      "eval_steps_per_second": 4.748,
+      "eval_wer": 1.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.07742934572202866,
+      "eval_loss": 1.8142520189285278,
+      "eval_runtime": 149.8655,
+      "eval_samples_per_second": 37.741,
+      "eval_steps_per_second": 4.718,
+      "eval_wer": 0.940732775914365,
+      "step": 400
+    },
+    {
+      "epoch": 0.09678668215253582,
+      "grad_norm": 3.132490396499634,
+      "learning_rate": 0.00029759999999999997,
+      "loss": 3.9521,
+      "step": 500
+    },
+    {
+      "epoch": 0.09678668215253582,
+      "eval_loss": 1.4195518493652344,
+      "eval_runtime": 150.5171,
+      "eval_samples_per_second": 37.577,
+      "eval_steps_per_second": 4.697,
+      "eval_wer": 0.8693007655149171,
+      "step": 500
+    },
+    {
+      "epoch": 0.11614401858304298,
+      "eval_loss": 1.16689133644104,
+      "eval_runtime": 150.5387,
+      "eval_samples_per_second": 37.572,
+      "eval_steps_per_second": 4.696,
+      "eval_wer": 0.8055239042865626,
+      "step": 600
+    },
+    {
+      "epoch": 0.13550135501355012,
+      "eval_loss": 1.0756505727767944,
+      "eval_runtime": 151.2385,
+      "eval_samples_per_second": 37.398,
+      "eval_steps_per_second": 4.675,
+      "eval_wer": 0.7596251063215163,
+      "step": 700
+    },
+    {
+      "epoch": 0.1548586914440573,
+      "eval_loss": 0.9944618344306946,
+      "eval_runtime": 151.1646,
+      "eval_samples_per_second": 37.416,
+      "eval_steps_per_second": 4.677,
+      "eval_wer": 0.7223925149652549,
+      "step": 800
+    },
+    {
+      "epoch": 0.17421602787456447,
+      "eval_loss": 0.9381263256072998,
+      "eval_runtime": 151.6289,
+      "eval_samples_per_second": 37.302,
+      "eval_steps_per_second": 4.663,
+      "eval_wer": 0.6870857472998347,
+      "step": 900
+    },
+    {
+      "epoch": 0.19357336430507163,
+      "grad_norm": 7.335289001464844,
+      "learning_rate": 0.0002844,
+      "loss": 1.0266,
+      "step": 1000
+    },
+    {
+      "epoch": 0.19357336430507163,
+      "eval_loss": 0.8977694511413574,
+      "eval_runtime": 156.0202,
+      "eval_samples_per_second": 36.252,
+      "eval_steps_per_second": 4.531,
+      "eval_wer": 0.661472292211648,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2129307007355788,
+      "eval_loss": 0.8770694136619568,
+      "eval_runtime": 151.6589,
+      "eval_samples_per_second": 37.294,
+      "eval_steps_per_second": 4.662,
+      "eval_wer": 0.6450385967164706,
+      "step": 1100
+    },
+    {
+      "epoch": 0.23228803716608595,
+      "eval_loss": 0.851553201675415,
+      "eval_runtime": 151.5945,
+      "eval_samples_per_second": 37.31,
+      "eval_steps_per_second": 4.664,
+      "eval_wer": 0.640432668389209,
+      "step": 1200
+    },
+    {
+      "epoch": 0.2516453735965931,
+      "eval_loss": 0.8273979425430298,
+      "eval_runtime": 151.4524,
+      "eval_samples_per_second": 37.345,
+      "eval_steps_per_second": 4.668,
+      "eval_wer": 0.6138081558633307,
+      "step": 1300
+    },
+    {
+      "epoch": 0.27100271002710025,
+      "eval_loss": 0.7992698550224304,
+      "eval_runtime": 152.8076,
+      "eval_samples_per_second": 37.014,
+      "eval_steps_per_second": 4.627,
+      "eval_wer": 0.596973247099228,
+      "step": 1400
+    },
+    {
+      "epoch": 0.29036004645760743,
+      "grad_norm": 4.0737223625183105,
+      "learning_rate": 0.00026861052631578947,
+      "loss": 0.8454,
+      "step": 1500
+    },
+    {
+      "epoch": 0.29036004645760743,
+      "eval_loss": 0.7768516540527344,
+      "eval_runtime": 152.3743,
+      "eval_samples_per_second": 37.119,
+      "eval_steps_per_second": 4.64,
+      "eval_wer": 0.5887563993516394,
+      "step": 1500
+    },
+    {
+      "epoch": 0.3097173828881146,
+      "eval_loss": 0.7664207220077515,
+      "eval_runtime": 154.3668,
+      "eval_samples_per_second": 36.64,
+      "eval_steps_per_second": 4.58,
+      "eval_wer": 0.5997977885124617,
+      "step": 1600
+    },
+    {
+      "epoch": 0.32907471931862176,
+      "eval_loss": 0.7400562763214111,
+      "eval_runtime": 153.7228,
+      "eval_samples_per_second": 36.793,
+      "eval_steps_per_second": 4.599,
+      "eval_wer": 0.5592110542279854,
+      "step": 1700
+    },
+    {
+      "epoch": 0.34843205574912894,
+      "eval_loss": 0.746478796005249,
+      "eval_runtime": 151.7535,
+      "eval_samples_per_second": 37.271,
+      "eval_steps_per_second": 4.659,
+      "eval_wer": 0.5650206223620228,
+      "step": 1800
+    },
+    {
+      "epoch": 0.3677893921796361,
+      "eval_loss": 0.7252949476242065,
+      "eval_runtime": 151.7548,
+      "eval_samples_per_second": 37.271,
+      "eval_steps_per_second": 4.659,
+      "eval_wer": 0.5791272808974338,
+      "step": 1900
+    },
+    {
+      "epoch": 0.38714672861014326,
+      "grad_norm": 2.4802448749542236,
+      "learning_rate": 0.0002528210526315789,
+      "loss": 0.7537,
+      "step": 2000
+    },
+    {
+      "epoch": 0.38714672861014326,
+      "eval_loss": 0.7039346098899841,
+      "eval_runtime": 152.7969,
+      "eval_samples_per_second": 37.016,
+      "eval_steps_per_second": 4.627,
+      "eval_wer": 0.5343518800853782,
+      "step": 2000
+    },
+    {
+      "epoch": 0.4065040650406504,
+      "eval_loss": 0.6932350397109985,
+      "eval_runtime": 152.4406,
+      "eval_samples_per_second": 37.103,
+      "eval_steps_per_second": 4.638,
+      "eval_wer": 0.5168429330294811,
+      "step": 2100
+    },
+    {
+      "epoch": 0.4258614014711576,
+      "eval_loss": 0.696869432926178,
+      "eval_runtime": 153.0527,
+      "eval_samples_per_second": 36.955,
+      "eval_steps_per_second": 4.619,
+      "eval_wer": 0.5364381890837894,
+      "step": 2200
+    },
+    {
+      "epoch": 0.4452187379016647,
+      "eval_loss": 0.6781283617019653,
+      "eval_runtime": 152.1378,
+      "eval_samples_per_second": 37.177,
+      "eval_steps_per_second": 4.647,
+      "eval_wer": 0.5173725345444624,
+      "step": 2300
+    },
+    {
+      "epoch": 0.4645760743321719,
+      "eval_loss": 0.6760829091072083,
+      "eval_runtime": 151.9712,
+      "eval_samples_per_second": 37.218,
+      "eval_steps_per_second": 4.652,
+      "eval_wer": 0.5050312143923223,
+      "step": 2400
+    },
+    {
+      "epoch": 0.48393341076267904,
+      "grad_norm": 3.791292667388916,
+      "learning_rate": 0.0002370315789473684,
+      "loss": 0.681,
+      "step": 2500
+    },
+    {
+      "epoch": 0.48393341076267904,
+      "eval_loss": 0.6720712780952454,
+      "eval_runtime": 152.2414,
+      "eval_samples_per_second": 37.152,
+      "eval_steps_per_second": 4.644,
+      "eval_wer": 0.528718845789668,
+      "step": 2500
+    },
+    {
+      "epoch": 0.5032907471931862,
+      "eval_loss": 0.6598270535469055,
+      "eval_runtime": 151.7192,
+      "eval_samples_per_second": 37.279,
+      "eval_steps_per_second": 4.66,
+      "eval_wer": 0.5195069891351447,
+      "step": 2600
+    },
+    {
+      "epoch": 0.5226480836236934,
+      "eval_loss": 0.6555168628692627,
+      "eval_runtime": 152.5678,
+      "eval_samples_per_second": 37.072,
+      "eval_steps_per_second": 4.634,
+      "eval_wer": 0.4975846961210701,
+      "step": 2700
+    },
+    {
+      "epoch": 0.5420054200542005,
+      "eval_loss": 0.6535276770591736,
+      "eval_runtime": 152.5246,
+      "eval_samples_per_second": 37.083,
+      "eval_steps_per_second": 4.635,
+      "eval_wer": 0.49936608303509816,
+      "step": 2800
+    },
+    {
+      "epoch": 0.5613627564847077,
+      "eval_loss": 0.6258506178855896,
+      "eval_runtime": 151.843,
+      "eval_samples_per_second": 37.249,
+      "eval_steps_per_second": 4.656,
+      "eval_wer": 0.48192133010222915,
+      "step": 2900
+    },
+    {
+      "epoch": 0.5807200929152149,
+      "grad_norm": 9.4619779586792,
+      "learning_rate": 0.00022124210526315786,
+      "loss": 0.6737,
+      "step": 3000
+    },
+    {
+      "epoch": 0.5807200929152149,
+      "eval_loss": 0.629943311214447,
+      "eval_runtime": 151.8389,
+      "eval_samples_per_second": 37.25,
+      "eval_steps_per_second": 4.656,
+      "eval_wer": 0.48022018584198617,
+      "step": 3000
+    },
+    {
+      "epoch": 0.6000774293457221,
+      "eval_loss": 0.6378594636917114,
+      "eval_runtime": 151.6255,
+      "eval_samples_per_second": 37.302,
+      "eval_steps_per_second": 4.663,
+      "eval_wer": 0.4893197027812104,
+      "step": 3100
+    },
+    {
+      "epoch": 0.6194347657762292,
+      "eval_loss": 0.6225672364234924,
+      "eval_runtime": 153.0144,
+      "eval_samples_per_second": 36.964,
+      "eval_steps_per_second": 4.62,
+      "eval_wer": 0.4806053505801544,
+      "step": 3200
+    },
+    {
+      "epoch": 0.6387921022067363,
+      "eval_loss": 0.6088670492172241,
+      "eval_runtime": 152.2222,
+      "eval_samples_per_second": 37.156,
+      "eval_steps_per_second": 4.645,
+      "eval_wer": 0.4627112387860891,
+      "step": 3300
+    },
+    {
+      "epoch": 0.6581494386372435,
+      "eval_loss": 0.6028585433959961,
+      "eval_runtime": 153.0615,
+      "eval_samples_per_second": 36.952,
+      "eval_steps_per_second": 4.619,
+      "eval_wer": 0.47354399704707034,
+      "step": 3400
+    },
+    {
+      "epoch": 0.6775067750677507,
+      "grad_norm": 3.4705822467803955,
+      "learning_rate": 0.00020545263157894736,
+      "loss": 0.6419,
+      "step": 3500
+    },
+    {
+      "epoch": 0.6775067750677507,
+      "eval_loss": 0.5871421694755554,
+      "eval_runtime": 152.5739,
+      "eval_samples_per_second": 37.071,
+      "eval_steps_per_second": 4.634,
+      "eval_wer": 0.4592126590810611,
+      "step": 3500
+    },
+    {
+      "epoch": 0.6968641114982579,
+      "eval_loss": 0.6001027226448059,
+      "eval_runtime": 152.1697,
+      "eval_samples_per_second": 37.169,
+      "eval_steps_per_second": 4.646,
+      "eval_wer": 0.4610742886488742,
+      "step": 3600
+    },
+    {
+      "epoch": 0.716221447928765,
+      "eval_loss": 0.5848923921585083,
+      "eval_runtime": 152.6563,
+      "eval_samples_per_second": 37.051,
+      "eval_steps_per_second": 4.631,
+      "eval_wer": 0.4472565036670893,
+      "step": 3700
+    },
+    {
+      "epoch": 0.7355787843592722,
+      "eval_loss": 0.5923960208892822,
+      "eval_runtime": 152.6559,
+      "eval_samples_per_second": 37.051,
+      "eval_steps_per_second": 4.631,
+      "eval_wer": 0.46377044181605176,
+      "step": 3800
+    },
+    {
+      "epoch": 0.7549361207897793,
+      "eval_loss": 0.5767965316772461,
+      "eval_runtime": 152.1652,
+      "eval_samples_per_second": 37.17,
+      "eval_steps_per_second": 4.646,
+      "eval_wer": 0.4584904751969957,
+      "step": 3900
+    },
+    {
+      "epoch": 0.7742934572202865,
+      "grad_norm": 3.628082275390625,
+      "learning_rate": 0.00018966315789473683,
+      "loss": 0.6183,
+      "step": 4000
+    },
+    {
+      "epoch": 0.7742934572202865,
+      "eval_loss": 0.5672534704208374,
+      "eval_runtime": 152.4329,
+      "eval_samples_per_second": 37.105,
+      "eval_steps_per_second": 4.638,
+      "eval_wer": 0.44531463144549116,
+      "step": 4000
+    },
+    {
+      "epoch": 0.7936507936507936,
+      "eval_loss": 0.5575382113456726,
+      "eval_runtime": 152.2388,
+      "eval_samples_per_second": 37.152,
+      "eval_steps_per_second": 4.644,
+      "eval_wer": 0.4451862431994351,
+      "step": 4100
+    },
+    {
+      "epoch": 0.8130081300813008,
+      "eval_loss": 0.5631808042526245,
+      "eval_runtime": 152.7545,
+      "eval_samples_per_second": 37.027,
+      "eval_steps_per_second": 4.628,
+      "eval_wer": 0.4474972316284444,
+      "step": 4200
+    },
+    {
+      "epoch": 0.832365466511808,
+      "eval_loss": 0.5498641729354858,
+      "eval_runtime": 153.7788,
+      "eval_samples_per_second": 36.78,
+      "eval_steps_per_second": 4.598,
+      "eval_wer": 0.44008281041870617,
+      "step": 4300
+    },
+    {
+      "epoch": 0.8517228029423152,
+      "eval_loss": 0.5662574172019958,
+      "eval_runtime": 152.5034,
+      "eval_samples_per_second": 37.088,
+      "eval_steps_per_second": 4.636,
+      "eval_wer": 0.43101539054099597,
+      "step": 4400
+    },
+    {
+      "epoch": 0.8710801393728222,
+      "grad_norm": 2.376349925994873,
+      "learning_rate": 0.0001738736842105263,
+      "loss": 0.5877,
+      "step": 4500
+    },
+    {
+      "epoch": 0.8710801393728222,
+      "eval_loss": 0.5584732294082642,
+      "eval_runtime": 152.1714,
+      "eval_samples_per_second": 37.169,
+      "eval_steps_per_second": 4.646,
+      "eval_wer": 0.4317215258943044,
+      "step": 4500
+    },
+    {
+      "epoch": 0.8904374758033294,
+      "eval_loss": 0.5463821291923523,
+      "eval_runtime": 152.4923,
+      "eval_samples_per_second": 37.09,
+      "eval_steps_per_second": 4.636,
+      "eval_wer": 0.41997400138017366,
+      "step": 4600
+    },
+    {
+      "epoch": 0.9097948122338366,
+      "eval_loss": 0.5381494164466858,
+      "eval_runtime": 153.2139,
+      "eval_samples_per_second": 36.916,
+      "eval_steps_per_second": 4.614,
+      "eval_wer": 0.4192197204345942,
+      "step": 4700
+    },
+    {
+      "epoch": 0.9291521486643438,
+      "eval_loss": 0.5453722476959229,
+      "eval_runtime": 151.9737,
+      "eval_samples_per_second": 37.217,
+      "eval_steps_per_second": 4.652,
+      "eval_wer": 0.4201986808107718,
+      "step": 4800
+    },
+    {
+      "epoch": 0.948509485094851,
+      "eval_loss": 0.5237515568733215,
+      "eval_runtime": 151.8558,
+      "eval_samples_per_second": 37.246,
+      "eval_steps_per_second": 4.656,
+      "eval_wer": 0.41241514339362234,
+      "step": 4900
+    },
+    {
+      "epoch": 0.9678668215253581,
+      "grad_norm": 2.5489518642425537,
+      "learning_rate": 0.0001581157894736842,
+      "loss": 0.5621,
+      "step": 5000
+    },
+    {
+      "epoch": 0.9678668215253581,
+      "eval_loss": 0.5303541421890259,
+      "eval_runtime": 152.515,
+      "eval_samples_per_second": 37.085,
+      "eval_steps_per_second": 4.636,
+      "eval_wer": 0.41353854054661293,
+      "step": 5000
+    },
+    {
+      "epoch": 0.9872241579558653,
+      "eval_loss": 0.5163344740867615,
+      "eval_runtime": 156.7945,
+      "eval_samples_per_second": 36.073,
+      "eval_steps_per_second": 4.509,
+      "eval_wer": 0.4061080708061177,
+      "step": 5100
+    },
+    {
+      "epoch": 1.0065814943863725,
+      "eval_loss": 0.51596599817276,
+      "eval_runtime": 153.2891,
+      "eval_samples_per_second": 36.898,
+      "eval_steps_per_second": 4.612,
+      "eval_wer": 0.39927139670363176,
+      "step": 5200
+    },
+    {
+      "epoch": 1.0259388308168795,
+      "eval_loss": 0.5088583827018738,
+      "eval_runtime": 152.7112,
+      "eval_samples_per_second": 37.037,
+      "eval_steps_per_second": 4.63,
+      "eval_wer": 0.3898509091492674,
+      "step": 5300
+    },
+    {
+      "epoch": 1.0452961672473868,
+      "eval_loss": 0.5110610723495483,
+      "eval_runtime": 152.5555,
+      "eval_samples_per_second": 37.075,
+      "eval_steps_per_second": 4.634,
+      "eval_wer": 0.3985652613503234,
+      "step": 5400
+    },
+    {
+      "epoch": 1.064653503677894,
+      "grad_norm": 1.1362248659133911,
+      "learning_rate": 0.0001423578947368421,
+      "loss": 0.4882,
+      "step": 5500
+    },
+    {
+      "epoch": 1.064653503677894,
+      "eval_loss": 0.5010027885437012,
+      "eval_runtime": 152.1249,
+      "eval_samples_per_second": 37.18,
+      "eval_steps_per_second": 4.647,
+      "eval_wer": 0.38574248527547306,
+      "step": 5500
+    },
+    {
+      "epoch": 1.084010840108401,
+      "eval_loss": 0.49406561255455017,
+      "eval_runtime": 151.5623,
+      "eval_samples_per_second": 37.318,
+      "eval_steps_per_second": 4.665,
+      "eval_wer": 0.3858548249907721,
+      "step": 5600
+    },
+    {
+      "epoch": 1.1033681765389083,
+      "eval_loss": 0.49403733015060425,
+      "eval_runtime": 152.7631,
+      "eval_samples_per_second": 37.025,
+      "eval_steps_per_second": 4.628,
+      "eval_wer": 0.3813451878480525,
+      "step": 5700
+    },
+    {
+      "epoch": 1.1227255129694154,
+      "eval_loss": 0.4913772642612457,
+      "eval_runtime": 152.1406,
+      "eval_samples_per_second": 37.176,
+      "eval_steps_per_second": 4.647,
+      "eval_wer": 0.37815153022740766,
+      "step": 5800
+    },
+    {
+      "epoch": 1.1420828493999227,
+      "eval_loss": 0.48747047781944275,
+      "eval_runtime": 151.3195,
+      "eval_samples_per_second": 37.378,
+      "eval_steps_per_second": 4.672,
+      "eval_wer": 0.3745406108070806,
+      "step": 5900
+    },
+    {
+      "epoch": 1.1614401858304297,
+      "grad_norm": 1.0150744915008545,
+      "learning_rate": 0.00012656842105263156,
+      "loss": 0.4569,
+      "step": 6000
+    },
+    {
+      "epoch": 1.1614401858304297,
+      "eval_loss": 0.4841971695423126,
+      "eval_runtime": 151.8567,
+      "eval_samples_per_second": 37.246,
+      "eval_steps_per_second": 4.656,
+      "eval_wer": 0.38071929514852915,
+      "step": 6000
+    },
+    {
+      "epoch": 1.1807975222609368,
+      "eval_loss": 0.48611822724342346,
+      "eval_runtime": 150.971,
+      "eval_samples_per_second": 37.464,
+      "eval_steps_per_second": 4.683,
+      "eval_wer": 0.37370608720771614,
+      "step": 6100
+    },
+    {
+      "epoch": 1.2001548586914441,
+      "eval_loss": 0.48144644498825073,
+      "eval_runtime": 151.4548,
+      "eval_samples_per_second": 37.344,
+      "eval_steps_per_second": 4.668,
+      "eval_wer": 0.3760973182905105,
+      "step": 6200
+    },
+    {
+      "epoch": 1.2195121951219512,
+      "eval_loss": 0.47813892364501953,
+      "eval_runtime": 151.1935,
+      "eval_samples_per_second": 37.409,
+      "eval_steps_per_second": 4.676,
+      "eval_wer": 0.37409125194588433,
+      "step": 6300
+    },
+    {
+      "epoch": 1.2388695315524583,
+      "eval_loss": 0.4771001935005188,
+      "eval_runtime": 151.1732,
+      "eval_samples_per_second": 37.414,
+      "eval_steps_per_second": 4.677,
+      "eval_wer": 0.36815329556579096,
+      "step": 6400
+    },
+    {
+      "epoch": 1.2582268679829656,
+      "grad_norm": 1.3292571306228638,
+      "learning_rate": 0.00011077894736842105,
+      "loss": 0.4416,
+      "step": 6500
+    },
+    {
+      "epoch": 1.2582268679829656,
+      "eval_loss": 0.47095027565956116,
+      "eval_runtime": 151.5037,
+      "eval_samples_per_second": 37.332,
+      "eval_steps_per_second": 4.667,
+      "eval_wer": 0.37338511659257595,
+      "step": 6500
+    },
+    {
+      "epoch": 1.2775842044134726,
+      "eval_loss": 0.47211408615112305,
+      "eval_runtime": 150.9455,
+      "eval_samples_per_second": 37.47,
+      "eval_steps_per_second": 4.684,
+      "eval_wer": 0.3659706953828377,
+      "step": 6600
+    },
+    {
+      "epoch": 1.29694154084398,
+      "eval_loss": 0.4679400622844696,
+      "eval_runtime": 151.4191,
+      "eval_samples_per_second": 37.353,
+      "eval_steps_per_second": 4.669,
+      "eval_wer": 0.3638843863844265,
+      "step": 6700
+    },
+    {
+      "epoch": 1.316298877274487,
+      "eval_loss": 0.46228036284446716,
+      "eval_runtime": 151.3839,
+      "eval_samples_per_second": 37.362,
+      "eval_steps_per_second": 4.67,
+      "eval_wer": 0.366532393959333,
+      "step": 6800
+    },
+    {
+      "epoch": 1.3356562137049943,
+      "eval_loss": 0.46108925342559814,
+      "eval_runtime": 151.8163,
+      "eval_samples_per_second": 37.256,
+      "eval_steps_per_second": 4.657,
+      "eval_wer": 0.3601771757795574,
+      "step": 6900
+    },
+    {
+      "epoch": 1.3550135501355014,
+      "grad_norm": 0.8062695860862732,
+      "learning_rate": 9.498947368421052e-05,
+      "loss": 0.4324,
+      "step": 7000
+    },
+    {
+      "epoch": 1.3550135501355014,
+      "eval_loss": 0.46888086199760437,
+      "eval_runtime": 152.4379,
+      "eval_samples_per_second": 37.104,
+      "eval_steps_per_second": 4.638,
+      "eval_wer": 0.3609314567251368,
+      "step": 7000
+    },
+    {
+      "epoch": 1.3743708865660085,
+      "eval_loss": 0.4573034346103668,
+      "eval_runtime": 151.3077,
+      "eval_samples_per_second": 37.381,
+      "eval_steps_per_second": 4.673,
+      "eval_wer": 0.3602574184333424,
+      "step": 7100
+    },
+    {
+      "epoch": 1.3937282229965158,
+      "eval_loss": 0.45749789476394653,
+      "eval_runtime": 151.5824,
+      "eval_samples_per_second": 37.313,
+      "eval_steps_per_second": 4.664,
+      "eval_wer": 0.3546083356068752,
+      "step": 7200
+    },
+    {
+      "epoch": 1.4130855594270229,
+      "eval_loss": 0.4555954933166504,
+      "eval_runtime": 151.6035,
+      "eval_samples_per_second": 37.308,
+      "eval_steps_per_second": 4.663,
+      "eval_wer": 0.35836369180401534,
+      "step": 7300
+    },
+    {
+      "epoch": 1.43244289585753,
+      "eval_loss": 0.4495578408241272,
+      "eval_runtime": 152.5621,
+      "eval_samples_per_second": 37.073,
+      "eval_steps_per_second": 4.634,
+      "eval_wer": 0.350724591163679,
+      "step": 7400
+    },
+    {
+      "epoch": 1.4518002322880372,
+      "grad_norm": 0.7916799187660217,
+      "learning_rate": 7.92e-05,
+      "loss": 0.4255,
+      "step": 7500
+    },
+    {
+      "epoch": 1.4518002322880372,
+      "eval_loss": 0.44609567523002625,
+      "eval_runtime": 151.8498,
+      "eval_samples_per_second": 37.247,
+      "eval_steps_per_second": 4.656,
+      "eval_wer": 0.34671245847442667,
+      "step": 7500
+    },
+    {
+      "epoch": 1.4711575687185443,
+      "eval_loss": 0.44341230392456055,
+      "eval_runtime": 152.528,
+      "eval_samples_per_second": 37.082,
+      "eval_steps_per_second": 4.635,
+      "eval_wer": 0.3462470510824734,
+      "step": 7600
+    },
+    {
+      "epoch": 1.4905149051490514,
+      "eval_loss": 0.44362780451774597,
+      "eval_runtime": 152.5253,
+      "eval_samples_per_second": 37.082,
+      "eval_steps_per_second": 4.635,
+      "eval_wer": 0.3516393574168285,
+      "step": 7700
+    },
+    {
+      "epoch": 1.5098722415795587,
+      "eval_loss": 0.4406072199344635,
+      "eval_runtime": 152.4039,
+      "eval_samples_per_second": 37.112,
+      "eval_steps_per_second": 4.639,
+      "eval_wer": 0.34579769222127715,
+      "step": 7800
+    },
+    {
+      "epoch": 1.5292295780100658,
+      "eval_loss": 0.43874725699424744,
+      "eval_runtime": 152.6604,
+      "eval_samples_per_second": 37.05,
+      "eval_steps_per_second": 4.631,
+      "eval_wer": 0.3439360626534641,
+      "step": 7900
+    },
+    {
+      "epoch": 1.5485869144405728,
+      "grad_norm": 0.7491864562034607,
+      "learning_rate": 6.344210526315788e-05,
+      "loss": 0.4094,
+      "step": 8000
+    },
+    {
+      "epoch": 1.5485869144405728,
+      "eval_loss": 0.43253499269485474,
+      "eval_runtime": 153.8006,
+      "eval_samples_per_second": 36.775,
+      "eval_steps_per_second": 4.597,
+      "eval_wer": 0.3409831329941744,
+      "step": 8000
+    },
+    {
+      "epoch": 1.5679442508710801,
+      "eval_loss": 0.4359830617904663,
+      "eval_runtime": 153.3674,
+      "eval_samples_per_second": 36.879,
+      "eval_steps_per_second": 4.61,
+      "eval_wer": 0.3419299963088379,
+      "step": 8100
+    },
+    {
+      "epoch": 1.5873015873015874,
+      "eval_loss": 0.4285949170589447,
+      "eval_runtime": 153.3711,
+      "eval_samples_per_second": 36.878,
+      "eval_steps_per_second": 4.61,
+      "eval_wer": 0.3377252812505015,
+      "step": 8200
+    },
+    {
+      "epoch": 1.6066589237320945,
+      "eval_loss": 0.43007034063339233,
+      "eval_runtime": 152.2201,
+      "eval_samples_per_second": 37.157,
+      "eval_steps_per_second": 4.645,
+      "eval_wer": 0.3335526632536791,
+      "step": 8300
+    },
+    {
+      "epoch": 1.6260162601626016,
+      "eval_loss": 0.42966797947883606,
+      "eval_runtime": 152.0163,
+      "eval_samples_per_second": 37.207,
+      "eval_steps_per_second": 4.651,
+      "eval_wer": 0.3322848293238754,
+      "step": 8400
+    },
+    {
+      "epoch": 1.645373596593109,
+      "grad_norm": 1.047472596168518,
+      "learning_rate": 4.765263157894736e-05,
+      "loss": 0.4018,
+      "step": 8500
+    },
+    {
+      "epoch": 1.645373596593109,
+      "eval_loss": 0.4270441234111786,
+      "eval_runtime": 152.8058,
+      "eval_samples_per_second": 37.014,
+      "eval_steps_per_second": 4.627,
+      "eval_wer": 0.3338575853380623,
+      "step": 8500
+    },
+    {
+      "epoch": 1.664730933023616,
+      "eval_loss": 0.4267289638519287,
+      "eval_runtime": 152.5032,
+      "eval_samples_per_second": 37.088,
+      "eval_steps_per_second": 4.636,
+      "eval_wer": 0.3319959557702492,
+      "step": 8600
+    },
+    {
+      "epoch": 1.684088269454123,
+      "eval_loss": 0.4224300980567932,
+      "eval_runtime": 152.5862,
+      "eval_samples_per_second": 37.068,
+      "eval_steps_per_second": 4.633,
+      "eval_wer": 0.33275023671582865,
+      "step": 8700
+    },
+    {
+      "epoch": 1.7034456058846303,
+      "eval_loss": 0.4207303822040558,
+      "eval_runtime": 154.5205,
+      "eval_samples_per_second": 36.604,
+      "eval_steps_per_second": 4.575,
+      "eval_wer": 0.32984545264881,
+      "step": 8800
+    },
+    {
+      "epoch": 1.7228029423151374,
+      "eval_loss": 0.4197385013103485,
+      "eval_runtime": 152.0624,
+      "eval_samples_per_second": 37.195,
+      "eval_steps_per_second": 4.649,
+      "eval_wer": 0.32978125852578194,
+      "step": 8900
+    },
+    {
+      "epoch": 1.7421602787456445,
+      "grad_norm": 1.4507739543914795,
+      "learning_rate": 3.189473684210526e-05,
+      "loss": 0.3899,
+      "step": 9000
+    },
+    {
+      "epoch": 1.7421602787456445,
+      "eval_loss": 0.4183507561683655,
+      "eval_runtime": 157.4278,
+      "eval_samples_per_second": 35.928,
+      "eval_steps_per_second": 4.491,
+      "eval_wer": 0.3258493684903147,
+      "step": 9000
+    },
+    {
+      "epoch": 1.7615176151761518,
+      "eval_loss": 0.4164830148220062,
+      "eval_runtime": 153.0475,
+      "eval_samples_per_second": 36.956,
+      "eval_steps_per_second": 4.619,
+      "eval_wer": 0.3262024361669689,
+      "step": 9100
+    },
+    {
+      "epoch": 1.7808749516066589,
+      "eval_loss": 0.41182050108909607,
+      "eval_runtime": 152.4839,
+      "eval_samples_per_second": 37.092,
+      "eval_steps_per_second": 4.637,
+      "eval_wer": 0.322864341769511,
+      "step": 9200
+    },
+    {
+      "epoch": 1.800232288037166,
+      "eval_loss": 0.4134317636489868,
+      "eval_runtime": 152.6353,
+      "eval_samples_per_second": 37.056,
+      "eval_steps_per_second": 4.632,
+      "eval_wer": 0.3232334579769222,
+      "step": 9300
+    },
+    {
+      "epoch": 1.8195896244676733,
+      "eval_loss": 0.4126824736595154,
+      "eval_runtime": 152.5246,
+      "eval_samples_per_second": 37.083,
+      "eval_steps_per_second": 4.635,
+      "eval_wer": 0.3209064210171559,
+      "step": 9400
+    },
+    {
+      "epoch": 1.8389469608981805,
+      "grad_norm": 1.0012460947036743,
+      "learning_rate": 1.6105263157894736e-05,
+      "loss": 0.3665,
+      "step": 9500
+    },
+    {
+      "epoch": 1.8389469608981805,
+      "eval_loss": 0.41083237528800964,
+      "eval_runtime": 152.9993,
+      "eval_samples_per_second": 36.967,
+      "eval_steps_per_second": 4.621,
+      "eval_wer": 0.32109900338624,
+      "step": 9500
+    },
+    {
+      "epoch": 1.8583042973286876,
+      "eval_loss": 0.4090138077735901,
+      "eval_runtime": 152.5291,
+      "eval_samples_per_second": 37.081,
+      "eval_steps_per_second": 4.635,
+      "eval_wer": 0.3199114121102213,
+      "step": 9600
+    },
+    {
+      "epoch": 1.8776616337591947,
+      "eval_loss": 0.407578706741333,
+      "eval_runtime": 153.0711,
+      "eval_samples_per_second": 36.95,
+      "eval_steps_per_second": 4.619,
+      "eval_wer": 0.32087432395564186,
+      "step": 9700
+    },
+    {
+      "epoch": 1.897018970189702,
+      "eval_loss": 0.40649694204330444,
+      "eval_runtime": 154.4136,
+      "eval_samples_per_second": 36.629,
+      "eval_steps_per_second": 4.579,
+      "eval_wer": 0.31981512092567926,
+      "step": 9800
+    },
+    {
+      "epoch": 1.916376306620209,
+      "eval_loss": 0.40620651841163635,
+      "eval_runtime": 153.7508,
+      "eval_samples_per_second": 36.787,
+      "eval_steps_per_second": 4.598,
+      "eval_wer": 0.31923737381842693,
+      "step": 9900
+    },
+    {
+      "epoch": 1.9357336430507162,
+      "grad_norm": 0.7244949340820312,
+      "learning_rate": 3.157894736842105e-07,
+      "loss": 0.3698,
+      "step": 10000
+    },
+    {
+      "epoch": 1.9357336430507162,
+      "eval_loss": 0.4060620963573456,
+      "eval_runtime": 153.976,
+      "eval_samples_per_second": 36.733,
+      "eval_steps_per_second": 4.592,
+      "eval_wer": 0.31928551941069794,
+      "step": 10000
+    },
+    {
+      "epoch": 1.9550909794812235,
+      "eval_loss": 0.45229342579841614,
+      "eval_runtime": 154.2948,
+      "eval_samples_per_second": 36.657,
+      "eval_steps_per_second": 4.582,
+      "eval_wer": 0.3406140167867632,
+      "step": 10100
+    },
+    {
+      "epoch": 1.9744483159117305,
+      "eval_loss": 0.4579542577266693,
+      "eval_runtime": 151.5074,
+      "eval_samples_per_second": 37.331,
+      "eval_steps_per_second": 4.666,
+      "eval_wer": 0.3517837941936416,
+      "step": 10200
+    },
+    {
+      "epoch": 1.9938056523422376,
+      "eval_loss": 0.46043792366981506,
+      "eval_runtime": 151.4438,
+      "eval_samples_per_second": 37.347,
+      "eval_steps_per_second": 4.668,
+      "eval_wer": 0.35115790149411824,
+      "step": 10300
+    },
+    {
+      "epoch": 2.013162988772745,
+      "eval_loss": 0.46549099683761597,
+      "eval_runtime": 151.5994,
+      "eval_samples_per_second": 37.309,
+      "eval_steps_per_second": 4.664,
+      "eval_wer": 0.3552181797756415,
+      "step": 10400
+    },
+    {
+      "epoch": 2.032520325203252,
+      "grad_norm": 0.703632652759552,
+      "learning_rate": 0.0001463076923076923,
+      "loss": 0.3624,
+      "step": 10500
+    },
+    {
+      "epoch": 2.032520325203252,
+      "eval_loss": 0.4670031666755676,
+      "eval_runtime": 151.5063,
+      "eval_samples_per_second": 37.332,
+      "eval_steps_per_second": 4.666,
+      "eval_wer": 0.35144677504774435,
+      "step": 10500
+    },
+    {
+      "epoch": 2.051877661633759,
+      "eval_loss": 0.459250271320343,
+      "eval_runtime": 153.0971,
+      "eval_samples_per_second": 36.944,
+      "eval_steps_per_second": 4.618,
+      "eval_wer": 0.3628251833544639,
+      "step": 10600
+    },
+    {
+      "epoch": 2.0712349980642664,
+      "eval_loss": 0.46061432361602783,
+      "eval_runtime": 152.0732,
+      "eval_samples_per_second": 37.193,
+      "eval_steps_per_second": 4.649,
+      "eval_wer": 0.3545922870761182,
+      "step": 10700
+    },
+    {
+      "epoch": 2.0905923344947737,
+      "eval_loss": 0.46500489115715027,
+      "eval_runtime": 151.985,
+      "eval_samples_per_second": 37.214,
+      "eval_steps_per_second": 4.652,
+      "eval_wer": 0.35905377862656673,
+      "step": 10800
+    },
+    {
+      "epoch": 2.1099496709252805,
+      "eval_loss": 0.46085453033447266,
+      "eval_runtime": 152.4835,
+      "eval_samples_per_second": 37.093,
+      "eval_steps_per_second": 4.637,
+      "eval_wer": 0.35483301503747333,
+      "step": 10900
+    },
+    {
+      "epoch": 2.129307007355788,
+      "grad_norm": 0.5008242726325989,
+      "learning_rate": 0.00013863076923076922,
+      "loss": 0.3755,
+      "step": 11000
+    },
+    {
+      "epoch": 2.129307007355788,
+      "eval_loss": 0.4708138406276703,
+      "eval_runtime": 152.3457,
+      "eval_samples_per_second": 37.126,
+      "eval_steps_per_second": 4.641,
+      "eval_wer": 0.35573173275986586,
+      "step": 11000
+    },
+    {
+      "epoch": 2.148664343786295,
+      "eval_loss": 0.4649392366409302,
+      "eval_runtime": 152.7087,
+      "eval_samples_per_second": 37.038,
+      "eval_steps_per_second": 4.63,
+      "eval_wer": 0.3548009179759593,
+      "step": 11100
+    },
+    {
+      "epoch": 2.168021680216802,
+      "eval_loss": 0.4624271094799042,
+      "eval_runtime": 153.153,
+      "eval_samples_per_second": 36.93,
+      "eval_steps_per_second": 4.616,
+      "eval_wer": 0.355956412190464,
+      "step": 11200
+    },
+    {
+      "epoch": 2.1873790166473093,
+      "eval_loss": 0.45822229981422424,
+      "eval_runtime": 156.1964,
+      "eval_samples_per_second": 36.211,
+      "eval_steps_per_second": 4.526,
+      "eval_wer": 0.35229734717786587,
+      "step": 11300
+    },
+    {
+      "epoch": 2.2067363530778166,
+      "eval_loss": 0.466250479221344,
+      "eval_runtime": 152.6707,
+      "eval_samples_per_second": 37.047,
+      "eval_steps_per_second": 4.631,
+      "eval_wer": 0.3586044197653705,
+      "step": 11400
+    },
+    {
+      "epoch": 2.226093689508324,
+      "grad_norm": 0.9631055593490601,
+      "learning_rate": 0.00013093846153846151,
+      "loss": 0.3891,
+      "step": 11500
+    },
+    {
+      "epoch": 2.226093689508324,
+      "eval_loss": 0.46153655648231506,
+      "eval_runtime": 153.1909,
+      "eval_samples_per_second": 36.921,
+      "eval_steps_per_second": 4.615,
+      "eval_wer": 0.3552181797756415,
+      "step": 11500
+    },
+    {
+      "epoch": 2.2454510259388307,
+      "eval_loss": 0.4631531238555908,
+      "eval_runtime": 152.9395,
+      "eval_samples_per_second": 36.982,
+      "eval_steps_per_second": 4.623,
+      "eval_wer": 0.35886119625748264,
+      "step": 11600
+    },
+    {
+      "epoch": 2.264808362369338,
+      "eval_loss": 0.4495234191417694,
+      "eval_runtime": 153.0237,
+      "eval_samples_per_second": 36.962,
+      "eval_steps_per_second": 4.62,
+      "eval_wer": 0.3425398404776043,
+      "step": 11700
+    },
+    {
+      "epoch": 2.2841656987998453,
+      "eval_loss": 0.462666779756546,
+      "eval_runtime": 152.4714,
+      "eval_samples_per_second": 37.095,
+      "eval_steps_per_second": 4.637,
+      "eval_wer": 0.34942466017236123,
+      "step": 11800
+    },
+    {
+      "epoch": 2.303523035230352,
+      "eval_loss": 0.4550352096557617,
+      "eval_runtime": 152.8072,
+      "eval_samples_per_second": 37.014,
+      "eval_steps_per_second": 4.627,
+      "eval_wer": 0.3451717995217538,
+      "step": 11900
+    },
+    {
+      "epoch": 2.3228803716608595,
+      "grad_norm": 0.7961182594299316,
+      "learning_rate": 0.00012324615384615384,
+      "loss": 0.3946,
+      "step": 12000
+    },
+    {
+      "epoch": 2.3228803716608595,
+      "eval_loss": 0.44988927245140076,
+      "eval_runtime": 152.9644,
+      "eval_samples_per_second": 36.976,
+      "eval_steps_per_second": 4.622,
+      "eval_wer": 0.3462310025517164,
+      "step": 12000
+    },
+    {
+      "epoch": 2.3422377080913668,
+      "eval_loss": 0.4501667320728302,
+      "eval_runtime": 153.061,
+      "eval_samples_per_second": 36.953,
+      "eval_steps_per_second": 4.619,
+      "eval_wer": 0.341978141901109,
+      "step": 12100
+    },
+    {
+      "epoch": 2.3615950445218736,
+      "eval_loss": 0.4580215513706207,
+      "eval_runtime": 153.2108,
+      "eval_samples_per_second": 36.916,
+      "eval_steps_per_second": 4.615,
+      "eval_wer": 0.3412399094862865,
+      "step": 12200
+    },
+    {
+      "epoch": 2.380952380952381,
+      "eval_loss": 0.4506891667842865,
+      "eval_runtime": 153.6611,
+      "eval_samples_per_second": 36.808,
+      "eval_steps_per_second": 4.601,
+      "eval_wer": 0.34339041260772574,
+      "step": 12300
+    },
+    {
+      "epoch": 2.4003097173828882,
+      "eval_loss": 0.44618555903434753,
+      "eval_runtime": 153.273,
+      "eval_samples_per_second": 36.901,
+      "eval_steps_per_second": 4.613,
+      "eval_wer": 0.34475453772207154,
+      "step": 12400
+    },
+    {
+      "epoch": 2.419667053813395,
+      "grad_norm": 0.828158974647522,
+      "learning_rate": 0.00011556923076923076,
+      "loss": 0.3824,
+      "step": 12500
+    },
+    {
+      "epoch": 2.419667053813395,
+      "eval_loss": 0.44126543402671814,
+      "eval_runtime": 153.3979,
+      "eval_samples_per_second": 36.871,
+      "eval_steps_per_second": 4.609,
+      "eval_wer": 0.34127200654780054,
+      "step": 12500
+    },
+    {
+      "epoch": 2.4390243902439024,
+      "eval_loss": 0.44880929589271545,
+      "eval_runtime": 153.7143,
+      "eval_samples_per_second": 36.796,
+      "eval_steps_per_second": 4.599,
+      "eval_wer": 0.3443212273916323,
+      "step": 12600
+    },
+    {
+      "epoch": 2.4583817266744097,
+      "eval_loss": 0.44148463010787964,
+      "eval_runtime": 153.647,
+      "eval_samples_per_second": 36.812,
+      "eval_steps_per_second": 4.601,
+      "eval_wer": 0.3431657331771276,
+      "step": 12700
+    },
+    {
+      "epoch": 2.4777390631049165,
+      "eval_loss": 0.44202256202697754,
+      "eval_runtime": 153.5743,
+      "eval_samples_per_second": 36.829,
+      "eval_steps_per_second": 4.604,
+      "eval_wer": 0.34093498740190337,
+      "step": 12800
+    },
+    {
+      "epoch": 2.497096399535424,
+      "eval_loss": 0.4379221200942993,
+      "eval_runtime": 153.5736,
+      "eval_samples_per_second": 36.829,
+      "eval_steps_per_second": 4.604,
+      "eval_wer": 0.3361204281748006,
+      "step": 12900
+    },
+    {
+      "epoch": 2.516453735965931,
+      "grad_norm": 1.2163615226745605,
+      "learning_rate": 0.00010787692307692307,
+      "loss": 0.372,
+      "step": 13000
+    },
+    {
+      "epoch": 2.516453735965931,
+      "eval_loss": 0.43855318427085876,
+      "eval_runtime": 153.5476,
+      "eval_samples_per_second": 36.835,
+      "eval_steps_per_second": 4.604,
+      "eval_wer": 0.3334884691306511,
+      "step": 13000
+    },
+    {
+      "epoch": 2.535811072396438,
+      "eval_loss": 0.44449883699417114,
+      "eval_runtime": 153.7016,
+      "eval_samples_per_second": 36.799,
+      "eval_steps_per_second": 4.6,
+      "eval_wer": 0.3397794931873987,
+      "step": 13100
+    },
+    {
+      "epoch": 2.5551684088269453,
+      "eval_loss": 0.4401286542415619,
+      "eval_runtime": 154.2488,
+      "eval_samples_per_second": 36.668,
+      "eval_steps_per_second": 4.584,
+      "eval_wer": 0.3392819887339314,
+      "step": 13200
+    },
+    {
+      "epoch": 2.5745257452574526,
+      "eval_loss": 0.437770813703537,
+      "eval_runtime": 153.8927,
+      "eval_samples_per_second": 36.753,
+      "eval_steps_per_second": 4.594,
+      "eval_wer": 0.335077273675595,
+      "step": 13300
+    },
+    {
+      "epoch": 2.59388308168796,
+      "eval_loss": 0.4315861463546753,
+      "eval_runtime": 153.7886,
+      "eval_samples_per_second": 36.778,
+      "eval_steps_per_second": 4.597,
+      "eval_wer": 0.33517356486013705,
+      "step": 13400
+    },
+    {
+      "epoch": 2.6132404181184667,
+      "grad_norm": 1.084632158279419,
+      "learning_rate": 0.0001002,
+      "loss": 0.3521,
+      "step": 13500
+    },
+    {
+      "epoch": 2.6132404181184667,
+      "eval_loss": 0.43864014744758606,
+      "eval_runtime": 153.9711,
+      "eval_samples_per_second": 36.734,
+      "eval_steps_per_second": 4.592,
+      "eval_wer": 0.33398597358411836,
+      "step": 13500
+    },
+    {
+      "epoch": 2.632597754548974,
+      "eval_loss": 0.43551018834114075,
+      "eval_runtime": 154.3017,
+      "eval_samples_per_second": 36.655,
+      "eval_steps_per_second": 4.582,
+      "eval_wer": 0.33096884980180064,
+      "step": 13600
+    },
+    {
+      "epoch": 2.6519550909794813,
+      "eval_loss": 0.4325660765171051,
+      "eval_runtime": 154.4812,
+      "eval_samples_per_second": 36.613,
+      "eval_steps_per_second": 4.577,
+      "eval_wer": 0.3343871868530436,
+      "step": 13700
+    },
+    {
+      "epoch": 2.6713124274099886,
+      "eval_loss": 0.4263465404510498,
+      "eval_runtime": 154.0733,
+      "eval_samples_per_second": 36.71,
+      "eval_steps_per_second": 4.589,
+      "eval_wer": 0.32629872735151094,
+      "step": 13800
+    },
+    {
+      "epoch": 2.6906697638404955,
+      "eval_loss": 0.42636117339134216,
+      "eval_runtime": 154.1615,
+      "eval_samples_per_second": 36.689,
+      "eval_steps_per_second": 4.586,
+      "eval_wer": 0.32353838006130536,
+      "step": 13900
+    },
+    {
+      "epoch": 2.710027100271003,
+      "grad_norm": 1.1979655027389526,
+      "learning_rate": 9.25076923076923e-05,
+      "loss": 0.3592,
+      "step": 14000
+    },
+    {
+      "epoch": 2.710027100271003,
+      "eval_loss": 0.4322036802768707,
+      "eval_runtime": 154.5242,
+      "eval_samples_per_second": 36.603,
+      "eval_steps_per_second": 4.575,
+      "eval_wer": 0.3299738408948661,
+      "step": 14000
+    },
+    {
+      "epoch": 2.72938443670151,
+      "eval_loss": 0.4294193983078003,
+      "eval_runtime": 154.4329,
+      "eval_samples_per_second": 36.624,
+      "eval_steps_per_second": 4.578,
+      "eval_wer": 0.3261542905746979,
+      "step": 14100
+    },
+    {
+      "epoch": 2.748741773132017,
+      "eval_loss": 0.43099814653396606,
+      "eval_runtime": 154.4209,
+      "eval_samples_per_second": 36.627,
+      "eval_steps_per_second": 4.578,
+      "eval_wer": 0.32329765209995026,
+      "step": 14200
+    },
+    {
+      "epoch": 2.7680991095625243,
+      "eval_loss": 0.42700281739234924,
+      "eval_runtime": 155.4008,
+      "eval_samples_per_second": 36.396,
+      "eval_steps_per_second": 4.55,
+      "eval_wer": 0.3268122803357353,
+      "step": 14300
+    },
+    {
+      "epoch": 2.7874564459930316,
+      "eval_loss": 0.4209098219871521,
+      "eval_runtime": 156.5271,
+      "eval_samples_per_second": 36.134,
+      "eval_steps_per_second": 4.517,
+      "eval_wer": 0.3254321066906325,
+      "step": 14400
+    },
+    {
+      "epoch": 2.8068137824235384,
+      "grad_norm": 0.6974443793296814,
+      "learning_rate": 8.48153846153846e-05,
+      "loss": 0.3459,
+      "step": 14500
+    },
+    {
+      "epoch": 2.8068137824235384,
+      "eval_loss": 0.42542555928230286,
+      "eval_runtime": 157.9392,
+      "eval_samples_per_second": 35.811,
+      "eval_steps_per_second": 4.476,
+      "eval_wer": 0.32729373625844554,
+      "step": 14500
+    },
+    {
+      "epoch": 2.8261711188540457,
+      "eval_loss": 0.42783817648887634,
+      "eval_runtime": 155.0217,
+      "eval_samples_per_second": 36.485,
+      "eval_steps_per_second": 4.561,
+      "eval_wer": 0.3231532153231372,
+      "step": 14600
+    },
+    {
+      "epoch": 2.845528455284553,
+      "eval_loss": 0.4212438464164734,
+      "eval_runtime": 154.853,
+      "eval_samples_per_second": 36.525,
+      "eval_steps_per_second": 4.566,
+      "eval_wer": 0.3215002166551652,
+      "step": 14700
+    },
+    {
+      "epoch": 2.86488579171506,
+      "eval_loss": 0.4169256389141083,
+      "eval_runtime": 154.5142,
+      "eval_samples_per_second": 36.605,
+      "eval_steps_per_second": 4.576,
+      "eval_wer": 0.31928551941069794,
+      "step": 14800
+    },
+    {
+      "epoch": 2.884243128145567,
+      "eval_loss": 0.42132049798965454,
+      "eval_runtime": 154.8091,
+      "eval_samples_per_second": 36.535,
+      "eval_steps_per_second": 4.567,
+      "eval_wer": 0.3195262473720531,
+      "step": 14900
+    },
+    {
+      "epoch": 2.9036004645760745,
+      "grad_norm": 1.099702000617981,
+      "learning_rate": 7.713846153846152e-05,
+      "loss": 0.3483,
+      "step": 15000
+    },
+    {
+      "epoch": 2.9036004645760745,
+      "eval_loss": 0.41696369647979736,
+      "eval_runtime": 155.3223,
+      "eval_samples_per_second": 36.415,
+      "eval_steps_per_second": 4.552,
+      "eval_wer": 0.31652517212049236,
+      "step": 15000
+    },
+    {
+      "epoch": 2.9229578010065813,
+      "eval_loss": 0.41230952739715576,
+      "eval_runtime": 154.9181,
+      "eval_samples_per_second": 36.51,
+      "eval_steps_per_second": 4.564,
+      "eval_wer": 0.31418208662996905,
+      "step": 15100
+    },
+    {
+      "epoch": 2.9423151374370886,
+      "eval_loss": 0.4116990566253662,
+      "eval_runtime": 154.97,
+      "eval_samples_per_second": 36.497,
+      "eval_steps_per_second": 4.562,
+      "eval_wer": 0.31337966009211854,
+      "step": 15200
+    },
+    {
+      "epoch": 2.961672473867596,
+      "eval_loss": 0.410386323928833,
+      "eval_runtime": 155.0232,
+      "eval_samples_per_second": 36.485,
+      "eval_steps_per_second": 4.561,
+      "eval_wer": 0.31158222464733354,
+      "step": 15300
+    },
+    {
+      "epoch": 2.9810298102981028,
+      "eval_loss": 0.41244322061538696,
+      "eval_runtime": 154.4682,
+      "eval_samples_per_second": 36.616,
+      "eval_steps_per_second": 4.577,
+      "eval_wer": 0.31419813516072603,
+      "step": 15400
+    },
+    {
+      "epoch": 3.00038714672861,
+      "grad_norm": 0.725528359413147,
+      "learning_rate": 6.946153846153845e-05,
+      "loss": 0.3501,
+      "step": 15500
+    },
+    {
+      "epoch": 3.00038714672861,
+      "eval_loss": 0.40684688091278076,
+      "eval_runtime": 154.5863,
+      "eval_samples_per_second": 36.588,
+      "eval_steps_per_second": 4.573,
+      "eval_wer": 0.31272167033108117,
+      "step": 15500
+    },
+    {
+      "epoch": 3.0197444831591174,
+      "eval_loss": 0.4200752079486847,
+      "eval_runtime": 154.5821,
+      "eval_samples_per_second": 36.589,
+      "eval_steps_per_second": 4.574,
+      "eval_wer": 0.3087416347033429,
+      "step": 15600
+    },
+    {
+      "epoch": 3.0391018195896247,
+      "eval_loss": 0.4186869263648987,
+      "eval_runtime": 154.7417,
+      "eval_samples_per_second": 36.551,
+      "eval_steps_per_second": 4.569,
+      "eval_wer": 0.3137808733610438,
+      "step": 15700
+    },
+    {
+      "epoch": 3.0584591560201315,
+      "eval_loss": 0.41133585572242737,
+      "eval_runtime": 155.21,
+      "eval_samples_per_second": 36.441,
+      "eval_steps_per_second": 4.555,
+      "eval_wer": 0.31092423488629617,
+      "step": 15800
+    },
+    {
+      "epoch": 3.077816492450639,
+      "eval_loss": 0.4191639721393585,
+      "eval_runtime": 155.276,
+      "eval_samples_per_second": 36.425,
+      "eval_steps_per_second": 4.553,
+      "eval_wer": 0.30851695527274475,
+      "step": 15900
+    },
+    {
+      "epoch": 3.097173828881146,
+      "grad_norm": 0.5114701390266418,
+      "learning_rate": 6.176923076923076e-05,
+      "loss": 0.2754,
+      "step": 16000
+    },
+    {
+      "epoch": 3.097173828881146,
+      "eval_loss": 0.4161028265953064,
+      "eval_runtime": 154.597,
+      "eval_samples_per_second": 36.585,
+      "eval_steps_per_second": 4.573,
+      "eval_wer": 0.30901445972621205,
+      "step": 16000
+    },
+    {
+      "epoch": 3.116531165311653,
+      "eval_loss": 0.4183988571166992,
+      "eval_runtime": 155.0124,
+      "eval_samples_per_second": 36.487,
+      "eval_steps_per_second": 4.561,
+      "eval_wer": 0.307152830158399,
+      "step": 16100
+    },
+    {
+      "epoch": 3.1358885017421603,
+      "eval_loss": 0.4186756908893585,
+      "eval_runtime": 154.8535,
+      "eval_samples_per_second": 36.525,
+      "eval_steps_per_second": 4.566,
+      "eval_wer": 0.3060936271284364,
+      "step": 16200
+    },
+    {
+      "epoch": 3.1552458381726676,
+      "eval_loss": 0.4193824827671051,
+      "eval_runtime": 154.3195,
+      "eval_samples_per_second": 36.651,
+      "eval_steps_per_second": 4.581,
+      "eval_wer": 0.3059652388823803,
+      "step": 16300
+    },
+    {
+      "epoch": 3.1746031746031744,
+      "eval_loss": 0.40788766741752625,
+      "eval_runtime": 154.9673,
+      "eval_samples_per_second": 36.498,
+      "eval_steps_per_second": 4.562,
+      "eval_wer": 0.3038949784147261,
+      "step": 16400
+    },
+    {
+      "epoch": 3.1939605110336817,
+      "grad_norm": 0.5594165325164795,
+      "learning_rate": 5.4076923076923074e-05,
+      "loss": 0.2802,
+      "step": 16500
+    },
+    {
+      "epoch": 3.1939605110336817,
+      "eval_loss": 0.41461309790611267,
+      "eval_runtime": 154.8662,
+      "eval_samples_per_second": 36.522,
+      "eval_steps_per_second": 4.565,
+      "eval_wer": 0.30424804609138034,
+      "step": 16500
+    },
+    {
+      "epoch": 3.213317847464189,
+      "eval_loss": 0.4168522357940674,
+      "eval_runtime": 155.0374,
+      "eval_samples_per_second": 36.482,
+      "eval_steps_per_second": 4.56,
+      "eval_wer": 0.30116672818603457,
+      "step": 16600
+    },
+    {
+      "epoch": 3.2326751838946963,
+      "eval_loss": 0.40926745533943176,
+      "eval_runtime": 154.7423,
+      "eval_samples_per_second": 36.551,
+      "eval_steps_per_second": 4.569,
+      "eval_wer": 0.3023864165235673,
+      "step": 16700
+    },
+    {
+      "epoch": 3.252032520325203,
+      "eval_loss": 0.4115259051322937,
+      "eval_runtime": 154.7933,
+      "eval_samples_per_second": 36.539,
+      "eval_steps_per_second": 4.567,
+      "eval_wer": 0.3005408354865112,
+      "step": 16800
+    },
+    {
+      "epoch": 3.2713898567557105,
+      "eval_loss": 0.40197211503982544,
+      "eval_runtime": 155.5964,
+      "eval_samples_per_second": 36.35,
+      "eval_steps_per_second": 4.544,
+      "eval_wer": 0.30410360931456726,
+      "step": 16900
+    },
+    {
+      "epoch": 3.290747193186218,
+      "grad_norm": 1.4730154275894165,
+      "learning_rate": 4.6384615384615385e-05,
+      "loss": 0.2723,
+      "step": 17000
+    },
+    {
+      "epoch": 3.290747193186218,
+      "eval_loss": 0.4058869779109955,
+      "eval_runtime": 155.0898,
+      "eval_samples_per_second": 36.469,
+      "eval_steps_per_second": 4.559,
+      "eval_wer": 0.30442457992970745,
+      "step": 17000
+    },
+    {
+      "epoch": 3.3101045296167246,
+      "eval_loss": 0.40676185488700867,
+      "eval_runtime": 155.0576,
+      "eval_samples_per_second": 36.477,
+      "eval_steps_per_second": 4.56,
+      "eval_wer": 0.3013753590858757,
+      "step": 17100
+    },
+    {
+      "epoch": 3.329461866047232,
+      "eval_loss": 0.40653425455093384,
+      "eval_runtime": 155.8377,
+      "eval_samples_per_second": 36.294,
+      "eval_steps_per_second": 4.537,
+      "eval_wer": 0.30878978029561394,
+      "step": 17200
+    },
+    {
+      "epoch": 3.3488192024777392,
+      "eval_loss": 0.4082197844982147,
+      "eval_runtime": 155.7924,
+      "eval_samples_per_second": 36.305,
+      "eval_steps_per_second": 4.538,
+      "eval_wer": 0.3010543884707355,
+      "step": 17300
+    },
+    {
+      "epoch": 3.368176538908246,
+      "eval_loss": 0.4083554446697235,
+      "eval_runtime": 155.6775,
+      "eval_samples_per_second": 36.332,
+      "eval_steps_per_second": 4.541,
+      "eval_wer": 0.3007494663863523,
+      "step": 17400
+    },
+    {
+      "epoch": 3.3875338753387534,
+      "grad_norm": 0.5211097598075867,
+      "learning_rate": 3.87076923076923e-05,
+      "loss": 0.2557,
+      "step": 17500
+    },
+    {
+      "epoch": 3.3875338753387534,
+      "eval_loss": 0.4009736180305481,
+      "eval_runtime": 155.105,
+      "eval_samples_per_second": 36.466,
+      "eval_steps_per_second": 4.558,
+      "eval_wer": 0.29924090449519347,
+      "step": 17500
+    },
+    {
+      "epoch": 3.4068912117692607,
+      "eval_loss": 0.4061805009841919,
+      "eval_runtime": 154.8792,
+      "eval_samples_per_second": 36.519,
+      "eval_steps_per_second": 4.565,
+      "eval_wer": 0.2999951854407729,
+      "step": 17600
+    },
+    {
+      "epoch": 3.4262485481997675,
+      "eval_loss": 0.40264037251472473,
+      "eval_runtime": 155.9957,
+      "eval_samples_per_second": 36.257,
+      "eval_steps_per_second": 4.532,
+      "eval_wer": 0.2980533132191748,
+      "step": 17700
+    },
+    {
+      "epoch": 3.445605884630275,
+      "eval_loss": 0.40035372972488403,
+      "eval_runtime": 155.0928,
+      "eval_samples_per_second": 36.468,
+      "eval_steps_per_second": 4.559,
+      "eval_wer": 0.29893598241081026,
+      "step": 17800
+    },
+    {
+      "epoch": 3.464963221060782,
+      "eval_loss": 0.40443336963653564,
+      "eval_runtime": 154.9305,
+      "eval_samples_per_second": 36.507,
+      "eval_steps_per_second": 4.563,
+      "eval_wer": 0.29906437065686636,
+      "step": 17900
+    },
+    {
+      "epoch": 3.484320557491289,
+      "grad_norm": 0.7458967566490173,
+      "learning_rate": 3.101538461538461e-05,
+      "loss": 0.2578,
+      "step": 18000
+    },
+    {
+      "epoch": 3.484320557491289,
+      "eval_loss": 0.4003549814224243,
+      "eval_runtime": 155.7394,
+      "eval_samples_per_second": 36.317,
+      "eval_steps_per_second": 4.54,
+      "eval_wer": 0.29660894545104394,
+      "step": 18000
+    },
+    {
+      "epoch": 3.5036778939217963,
+      "eval_loss": 0.40592488646507263,
+      "eval_runtime": 159.2644,
+      "eval_samples_per_second": 35.513,
+      "eval_steps_per_second": 4.439,
+      "eval_wer": 0.29449053939111874,
+      "step": 18100
+    },
+    {
+      "epoch": 3.5230352303523036,
+      "eval_loss": 0.4014962613582611,
+      "eval_runtime": 155.6654,
+      "eval_samples_per_second": 36.334,
+      "eval_steps_per_second": 4.542,
+      "eval_wer": 0.29632007189741777,
+      "step": 18200
+    },
+    {
+      "epoch": 3.5423925667828104,
+      "eval_loss": 0.396659791469574,
+      "eval_runtime": 156.1536,
+      "eval_samples_per_second": 36.221,
+      "eval_steps_per_second": 4.528,
+      "eval_wer": 0.29585466450546455,
+      "step": 18300
+    },
+    {
+      "epoch": 3.5617499032133177,
+      "eval_loss": 0.4001907706260681,
+      "eval_runtime": 155.7578,
+      "eval_samples_per_second": 36.313,
+      "eval_steps_per_second": 4.539,
+      "eval_wer": 0.29412142318370754,
+      "step": 18400
+    },
+    {
+      "epoch": 3.581107239643825,
+      "grad_norm": 0.6122294664382935,
+      "learning_rate": 2.3338461538461535e-05,
+      "loss": 0.2508,
+      "step": 18500
+    },
+    {
+      "epoch": 3.581107239643825,
+      "eval_loss": 0.39826107025146484,
+      "eval_runtime": 155.467,
+      "eval_samples_per_second": 36.381,
+      "eval_steps_per_second": 4.548,
+      "eval_wer": 0.2945547335141468,
+      "step": 18500
+    },
+    {
+      "epoch": 3.600464576074332,
+      "eval_loss": 0.3958674967288971,
+      "eval_runtime": 155.8242,
+      "eval_samples_per_second": 36.297,
+      "eval_steps_per_second": 4.537,
+      "eval_wer": 0.29365601579175427,
+      "step": 18600
+    },
+    {
+      "epoch": 3.619821912504839,
+      "eval_loss": 0.3970955014228821,
+      "eval_runtime": 155.4329,
+      "eval_samples_per_second": 36.389,
+      "eval_steps_per_second": 4.549,
+      "eval_wer": 0.2942016658374926,
+      "step": 18700
+    },
+    {
+      "epoch": 3.6391792489353465,
+      "eval_loss": 0.3906669616699219,
+      "eval_runtime": 155.4929,
+      "eval_samples_per_second": 36.375,
+      "eval_steps_per_second": 4.547,
+      "eval_wer": 0.2923239877389225,
+      "step": 18800
+    },
+    {
+      "epoch": 3.658536585365854,
+      "eval_loss": 0.39506247639656067,
+      "eval_runtime": 155.5246,
+      "eval_samples_per_second": 36.367,
+      "eval_steps_per_second": 4.546,
+      "eval_wer": 0.2903981640480814,
+      "step": 18900
+    },
+    {
+      "epoch": 3.6778939217963607,
+      "grad_norm": 0.33715635538101196,
+      "learning_rate": 1.5646153846153846e-05,
+      "loss": 0.2659,
+      "step": 19000
+    },
+    {
+      "epoch": 3.6778939217963607,
+      "eval_loss": 0.3892674744129181,
+      "eval_runtime": 155.5533,
+      "eval_samples_per_second": 36.361,
+      "eval_steps_per_second": 4.545,
+      "eval_wer": 0.29309431721525897,
+      "step": 19000
+    },
+    {
+      "epoch": 3.697251258226868,
+      "eval_loss": 0.39077267050743103,
+      "eval_runtime": 155.448,
+      "eval_samples_per_second": 36.385,
+      "eval_steps_per_second": 4.548,
+      "eval_wer": 0.2900771934329412,
+      "step": 19100
+    },
+    {
+      "epoch": 3.7166085946573753,
+      "eval_loss": 0.39407432079315186,
+      "eval_runtime": 155.4696,
+      "eval_samples_per_second": 36.38,
+      "eval_steps_per_second": 4.548,
+      "eval_wer": 0.2884241947649693,
+      "step": 19200
+    },
+    {
+      "epoch": 3.7359659310878826,
+      "eval_loss": 0.3924821615219116,
+      "eval_runtime": 155.4791,
+      "eval_samples_per_second": 36.378,
+      "eval_steps_per_second": 4.547,
+      "eval_wer": 0.2890019418722216,
+      "step": 19300
+    },
+    {
+      "epoch": 3.7553232675183894,
+      "eval_loss": 0.3916691243648529,
+      "eval_runtime": 155.9516,
+      "eval_samples_per_second": 36.268,
+      "eval_steps_per_second": 4.533,
+      "eval_wer": 0.2892908154258478,
+      "step": 19400
+    },
+    {
+      "epoch": 3.7746806039488967,
+      "grad_norm": 0.4647356867790222,
+      "learning_rate": 7.953846153846153e-06,
+      "loss": 0.2488,
+      "step": 19500
+    },
+    {
+      "epoch": 3.7746806039488967,
+      "eval_loss": 0.39043277502059937,
+      "eval_runtime": 155.3552,
+      "eval_samples_per_second": 36.407,
+      "eval_steps_per_second": 4.551,
+      "eval_wer": 0.2884562918264833,
+      "step": 19500
+    },
+    {
+      "epoch": 3.794037940379404,
+      "eval_loss": 0.39014604687690735,
+      "eval_runtime": 155.2137,
+      "eval_samples_per_second": 36.44,
+      "eval_steps_per_second": 4.555,
+      "eval_wer": 0.2887933109723805,
+      "step": 19600
+    },
+    {
+      "epoch": 3.813395276809911,
+      "eval_loss": 0.3883425295352936,
+      "eval_runtime": 155.5369,
+      "eval_samples_per_second": 36.364,
+      "eval_steps_per_second": 4.546,
+      "eval_wer": 0.28922662130281973,
+      "step": 19700
+    },
+    {
+      "epoch": 3.832752613240418,
+      "eval_loss": 0.38913780450820923,
+      "eval_runtime": 155.8958,
+      "eval_samples_per_second": 36.281,
+      "eval_steps_per_second": 4.535,
+      "eval_wer": 0.28903403893373564,
+      "step": 19800
+    },
+    {
+      "epoch": 3.8521099496709255,
+      "eval_loss": 0.3888201415538788,
+      "eval_runtime": 155.6372,
+      "eval_samples_per_second": 36.341,
+      "eval_steps_per_second": 4.543,
+      "eval_wer": 0.2888254080338945,
+      "step": 19900
+    },
+    {
+      "epoch": 3.8714672861014323,
+      "grad_norm": 0.3741956055164337,
+      "learning_rate": 2.615384615384615e-07,
+      "loss": 0.2602,
+      "step": 20000
+    },
+    {
+      "epoch": 3.8714672861014323,
+      "eval_loss": 0.38884833455085754,
+      "eval_runtime": 155.0772,
+      "eval_samples_per_second": 36.472,
+      "eval_steps_per_second": 4.559,
+      "eval_wer": 0.2884883888879973,
+      "step": 20000
+    },
+    {
+      "epoch": 3.8714672861014323,
+      "step": 20000,
+      "total_flos": 2.249387574100498e+19,
+      "train_loss": 0.15996522521972656,
+      "train_runtime": 19346.8732,
+      "train_samples_per_second": 8.27,
+      "train_steps_per_second": 1.034
+    }
+  ],
+  "logging_steps": 500,
+  "max_steps": 20000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 400,
+  "total_flos": 2.249387574100498e+19,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5642a4c90641028b1a9fd2ed5c13c87bb627ddc09bd278c1129e824a9f2dd851
-size 3247
+oid sha256:bd8f3d0ded2642d5d200c2ed12ad9bf454ae146a69f8cb7e882c46c9847874d8
+size 5048

vocab.json

@@ -1 +1,46 @@
-{"f": 0, "h": 2, "w": 3, "z": 4, "á": 5, "'": 6, "ŷ": 7, "i": 8, "-": 9, "é": 10, "ÿ": 11, "m": 12, "n": 13, "b": 14, "o": 15, "ï": 16, "k": 17, "ŵ": 18, "j": 19, "ô": 20, "l": 21, "ò": 22, "ä": 23, "r": 24, "s": 25, "u": 26, "ë": 27, "p": 28, "y": 29, "e": 30, "q": 31, "â": 32, "g": 33, "ö": 34, "a": 35, "à": 36, "ê": 37, "d": 38, "c": 39, "û": 40, "v": 41, "t": 42, "î": 43, "x": 44, "|": 1, "[UNK]": 45, "[PAD]": 46}
+{
+  "'": 1,
+  "-": 2,
+  "[PAD]": 43,
+  "[UNK]": 42,
+  "a": 3,
+  "b": 4,
+  "c": 5,
+  "d": 6,
+  "e": 7,
+  "f": 8,
+  "g": 9,
+  "h": 10,
+  "i": 11,
+  "j": 12,
+  "k": 13,
+  "l": 14,
+  "m": 15,
+  "n": 16,
+  "o": 17,
+  "p": 18,
+  "q": 19,
+  "r": 20,
+  "s": 21,
+  "t": 22,
+  "u": 23,
+  "v": 24,
+  "w": 25,
+  "x": 26,
+  "y": 27,
+  "z": 28,
+  "|": 0,
+  "à": 29,
+  "â": 30,
+  "ä": 31,
+  "ê": 32,
+  "ë": 33,
+  "î": 34,
+  "ï": 35,
+  "ò": 36,
+  "ô": 37,
+  "û": 38,
+  "ÿ": 39,
+  "ŵ": 40,
+  "ŷ": 41
+}