add language model

alphabet.json

@@ -0,0 +1 @@
+{"labels": [" ", "'", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "\u00e0", "\u00e2", "\u00e4", "\u00e7", "\u00e8", "\u00e9", "\u00ea", "\u00eb", "\u00ee", "\u00ef", "\u00f4", "\u00f6", "\u00f9", "\u00fb", "\u00fc", "\u00ff", "\u2047", ""], "is_bpe": false}

create_lm_model.ipynb

@@ -98,18 +98,69 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 29,
-   "id": "b605f94f",
+   "execution_count": 20,
+   "id": "6d82daed",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with open(\"5gram.arpa\", \"r\") as read_file, open(\"5gram_correct.arpa\", \"w\") as write_file:\n",
+    "    has_added_eos = False\n",
+    "    for line in read_file:\n",
+    "        if not has_added_eos and \"ngram 1=\" in line:\n",
+    "            count=line.strip().split(\"=\")[-1]\n",
+    "            write_file.write(line.replace(f\"{count}\", f\"{int(count)+1}\"))\n",
+    "        elif not has_added_eos and \"<s>\" in line:\n",
+    "            write_file.write(line)\n",
+    "            write_file.write(line.replace(\"<s>\", \"</s>\"))\n",
+    "            has_added_eos = True\n",
+    "        else:\n",
+    "            write_file.write(line)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "07ff4067",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoProcessor"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "e75ab227",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "processor = AutoProcessor.from_pretrained(\"./\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "604776b7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "vocab_dict = processor.tokenizer.get_vocab()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "ef4dd957",
    "metadata": {},
    "outputs": [],
    "source": [
-    "sys.path.append(\"test.arpa\")"
+    "sorted_vocab_dict = {k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])}"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 33,
-   "id": "9262368d",
+   "execution_count": 6,
+   "id": "9a14839d",
    "metadata": {},
    "outputs": [
     {
@@ -117,180 +168,51 @@
      "output_type": "stream",
      "text": [
       "Loading the LM will be faster if you build a binary file.\n",
-      "Reading /workspace/xls-r-300m-fr/language_model/5gram.arpa\n",
-      "----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100\n"
-     ]
-    },
-    {
-     "ename": "OSError",
-     "evalue": "Cannot read model 'language_model/5gram.arpa' (End of file Byte: 0)",
-     "output_type": "error",
-     "traceback": [
-      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
-      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
-      "File \u001b[0;32mkenlm.pyx:139\u001b[0m, in \u001b[0;36mkenlm.Model.__init__\u001b[0;34m()\u001b[0m\n",
-      "\u001b[0;31mRuntimeError\u001b[0m: End of file Byte: 0",
-      "\nThe above exception was the direct cause of the following exception:\n",
-      "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
-      "Input \u001b[0;32mIn [33]\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01msys\u001b[39;00m\n\u001b[1;32m      5\u001b[0m LM \u001b[38;5;241m=\u001b[39m os\u001b[38;5;241m.\u001b[39mpath\u001b[38;5;241m.\u001b[39mjoin(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mlanguage_model/\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m'\u001b[39m\u001b[38;5;124m5gram.arpa\u001b[39m\u001b[38;5;124m'\u001b[39m)\n\u001b[0;32m----> 6\u001b[0m model \u001b[38;5;241m=\u001b[39m \u001b[43mkenlm\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mLanguageModel\u001b[49m\u001b[43m(\u001b[49m\u001b[43mLM\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m      7\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;124m'\u001b[39m\u001b[38;5;132;01m{0}\u001b[39;00m\u001b[38;5;124m-gram model\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;241m.\u001b[39mformat(model\u001b[38;5;241m.\u001b[39morder))\n",
-      "File \u001b[0;32mkenlm.pyx:142\u001b[0m, in \u001b[0;36mkenlm.Model.__init__\u001b[0;34m()\u001b[0m\n",
-      "\u001b[0;31mOSError\u001b[0m: Cannot read model 'language_model/5gram.arpa' (End of file Byte: 0)"
+      "Reading /home/pascal/kenlm/build/bin/xls-r-300m-lm-fr/language_model/5gram_correct.arpa\n",
+      "----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100\n",
+      "****************************************************************************************************\n"
      ]
     }
    ],
    "source": [
-    "import os\n",
-    "import kenlm\n",
-    "import sys\n",
+    "from pyctcdecode import build_ctcdecoder\n",
     "\n",
-    "LM = os.path.join(\"language_model/\", '5gram.arpa')\n",
-    "model = kenlm.LanguageModel(LM)\n",
-    "print('{0}-gram model'.format(model.order))"
+    "decoder = build_ctcdecoder(\n",
+    "    labels=list(sorted_vocab_dict.keys()),\n",
+    "    kenlm_model_path=\"./language_model/5gram_correct.arpa\",\n",
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 38,
-   "id": "130f7f47",
+   "execution_count": 7,
+   "id": "656979ca",
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "14200cf29dc74389aeceb56701ee9a5f",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/255 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "9da79daa4d6b4a6694b96d0608d5b6ec",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/221 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "862bc04ec9e241608eab1aa2e21ef41f",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/730 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "ada7a9b9364644658c3e8b58c1a34d2c",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/85.0 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "e7cbe24e2bec4efa997a0e7d3d87d082",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/683 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "6b903049438d4067a0ec13d55483e9e4",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/227 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "9d59cebc3b7b46228c365f6b73cee4ee",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/78.0 [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "7a2709adf3764eb29750700b6f58ec40",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/1.50G [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "7276b3af03ce408ebcb485a0ff5641a7",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Downloading:   0%|          | 0.00/11.1M [00:00<?, ?B/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    }
-   ],
+   "outputs": [],
    "source": [
     "from transformers import Wav2Vec2ProcessorWithLM\n",
     "\n",
-    "processor = Wav2Vec2ProcessorWithLM.from_pretrained(\"Harveenchadha/hindi_model_with_lm_vakyansh\")"
+    "processor_with_lm = Wav2Vec2ProcessorWithLM(\n",
+    "    feature_extractor=processor.feature_extractor,\n",
+    "    tokenizer=processor.tokenizer,\n",
+    "    decoder=decoder\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "d2dd8891",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "processor_with_lm.save_pretrained(\"xls-r-300m-lm-fr\")"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "e75ab227",
+   "id": "85908c6d",
    "metadata": {},
    "outputs": [],
    "source": []
@@ -298,7 +220,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -312,7 +234,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.8"
+   "version": "3.7.9"
   }
  },
  "nbformat": 4,

language_model/{5gram.arpa → 5gram.bin}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adf4dc592314c578e7a8c2c0b4618fdef0a6a6ac06c4dd75acb30108c4d8d133
-size 507942255
+oid sha256:22eb4d68f40ad795dafa04ead8f220eb438bed4518f2b2af5a37ae64e4ffa9c7
+size 251967234

language_model/5gram_correct.arpa

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1909f82cc26181421c79ba6b774a59e68cbb80f6c61e53e4d6f68d09d6d987de
-size 507942072

language_model/attrs.json

@@ -0,0 +1 @@
+{"alpha": 0.5, "beta": 1.5, "unk_score_offset": -10.0, "score_boundary": true}

preprocessor_config.json

@@ -4,6 +4,7 @@
   "feature_size": 1,
   "padding_side": "right",
   "padding_value": 0,
+  "processor_class": "Wav2Vec2ProcessorWithLM",
   "return_attention_mask": true,
   "sampling_rate": 16000
 }

test_results/mozilla-foundation_common_voice_7_0_fr_test_eval_results.txt

@@ -1,2 +0,0 @@
-WER: 0.24561764914155493
-CER: 0.07285207821118034

tokenizer_config.json

@@ -1 +1 @@
-{"unk_token": "[UNK]", "bos_token": null, "eos_token": null, "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "special_tokens_map_file": null, "tokenizer_file": null, "name_or_path": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer"}
+{"unk_token": "[UNK]", "bos_token": null, "eos_token": null, "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "special_tokens_map_file": null, "tokenizer_file": null, "name_or_path": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2ProcessorWithLM"}

train_results/all_results.json

@@ -1,14 +0,0 @@
-{
-    "epoch": 2.0,
-    "eval_loss": 0.26187804341316223,
-    "eval_runtime": 722.142,
-    "eval_samples": 15941,
-    "eval_samples_per_second": 22.075,
-    "eval_steps_per_second": 1.381,
-    "eval_wer": 0.24574541380398318,
-    "train_loss": 1.788894365302016,
-    "train_runtime": 52105.5599,
-    "train_samples": 395042,
-    "train_samples_per_second": 15.163,
-    "train_steps_per_second": 0.118
-}

train_results/eval_results.json

@@ -1,9 +0,0 @@
-{
-    "epoch": 2.0,
-    "eval_loss": 0.26187804341316223,
-    "eval_runtime": 722.142,
-    "eval_samples": 15941,
-    "eval_samples_per_second": 22.075,
-    "eval_steps_per_second": 1.381,
-    "eval_wer": 0.24574541380398318
-}

train_results/train_results.json

@@ -1,8 +0,0 @@
-{
-    "epoch": 2.0,
-    "train_loss": 1.788894365302016,
-    "train_runtime": 52105.5599,
-    "train_samples": 395042,
-    "train_samples_per_second": 15.163,
-    "train_steps_per_second": 0.118
-}

train_results/trainer_state.json

@@ -1,499 +0,0 @@
-{
-  "best_metric": 0.26187804341316223,
-  "best_model_checkpoint": "./checkpoint-6000",
-  "epoch": 1.9998784982382245,
-  "global_step": 6172,
-  "is_hyper_param_search": false,
-  "is_local_process_zero": true,
-  "is_world_process_zero": true,
-  "log_history": [
-    {
-      "epoch": 0.03,
-      "learning_rate": 3.7499999999999997e-06,
-      "loss": 12.1043,
-      "step": 100
-    },
-    {
-      "epoch": 0.06,
-      "learning_rate": 7.499999999999999e-06,
-      "loss": 6.4771,
-      "step": 200
-    },
-    {
-      "epoch": 0.1,
-      "learning_rate": 1.1249999999999999e-05,
-      "loss": 4.4866,
-      "step": 300
-    },
-    {
-      "epoch": 0.13,
-      "learning_rate": 1.4999999999999999e-05,
-      "loss": 3.8842,
-      "step": 400
-    },
-    {
-      "epoch": 0.16,
-      "learning_rate": 1.8712499999999997e-05,
-      "loss": 3.495,
-      "step": 500
-    },
-    {
-      "epoch": 0.16,
-      "eval_loss": 3.3882696628570557,
-      "eval_runtime": 721.337,
-      "eval_samples_per_second": 22.099,
-      "eval_steps_per_second": 1.382,
-      "eval_wer": 1.0,
-      "step": 500
-    },
-    {
-      "epoch": 0.19,
-      "learning_rate": 2.2462499999999997e-05,
-      "loss": 3.171,
-      "step": 600
-    },
-    {
-      "epoch": 0.23,
-      "learning_rate": 2.6212499999999997e-05,
-      "loss": 3.0275,
-      "step": 700
-    },
-    {
-      "epoch": 0.26,
-      "learning_rate": 2.99625e-05,
-      "loss": 2.9681,
-      "step": 800
-    },
-    {
-      "epoch": 0.29,
-      "learning_rate": 3.37125e-05,
-      "loss": 2.9347,
-      "step": 900
-    },
-    {
-      "epoch": 0.32,
-      "learning_rate": 3.7462499999999996e-05,
-      "loss": 2.9095,
-      "step": 1000
-    },
-    {
-      "epoch": 0.32,
-      "eval_loss": 2.9152133464813232,
-      "eval_runtime": 718.1623,
-      "eval_samples_per_second": 22.197,
-      "eval_steps_per_second": 1.388,
-      "eval_wer": 0.9999871219487068,
-      "step": 1000
-    },
-    {
-      "epoch": 0.36,
-      "learning_rate": 4.12125e-05,
-      "loss": 2.8888,
-      "step": 1100
-    },
-    {
-      "epoch": 0.39,
-      "learning_rate": 4.4962499999999995e-05,
-      "loss": 2.8347,
-      "step": 1200
-    },
-    {
-      "epoch": 0.42,
-      "learning_rate": 4.871249999999999e-05,
-      "loss": 2.5318,
-      "step": 1300
-    },
-    {
-      "epoch": 0.45,
-      "learning_rate": 5.2462499999999994e-05,
-      "loss": 2.0502,
-      "step": 1400
-    },
-    {
-      "epoch": 0.49,
-      "learning_rate": 5.62125e-05,
-      "loss": 1.8434,
-      "step": 1500
-    },
-    {
-      "epoch": 0.49,
-      "eval_loss": 1.0473320484161377,
-      "eval_runtime": 720.1235,
-      "eval_samples_per_second": 22.136,
-      "eval_steps_per_second": 1.384,
-      "eval_wer": 0.7446153648029981,
-      "step": 1500
-    },
-    {
-      "epoch": 0.52,
-      "learning_rate": 5.9962499999999994e-05,
-      "loss": 1.7339,
-      "step": 1600
-    },
-    {
-      "epoch": 0.55,
-      "learning_rate": 6.367499999999999e-05,
-      "loss": 1.6535,
-      "step": 1700
-    },
-    {
-      "epoch": 0.58,
-      "learning_rate": 6.7425e-05,
-      "loss": 1.5793,
-      "step": 1800
-    },
-    {
-      "epoch": 0.62,
-      "learning_rate": 7.1175e-05,
-      "loss": 1.5056,
-      "step": 1900
-    },
-    {
-      "epoch": 0.65,
-      "learning_rate": 7.492499999999999e-05,
-      "loss": 1.4298,
-      "step": 2000
-    },
-    {
-      "epoch": 0.65,
-      "eval_loss": 0.5728740692138672,
-      "eval_runtime": 712.5783,
-      "eval_samples_per_second": 22.371,
-      "eval_steps_per_second": 1.399,
-      "eval_wer": 0.5129521000882147,
-      "step": 2000
-    },
-    {
-      "epoch": 0.68,
-      "learning_rate": 7.325623202301054e-05,
-      "loss": 1.3592,
-      "step": 2100
-    },
-    {
-      "epoch": 0.71,
-      "learning_rate": 7.145853307766058e-05,
-      "loss": 1.2917,
-      "step": 2200
-    },
-    {
-      "epoch": 0.75,
-      "learning_rate": 6.966083413231063e-05,
-      "loss": 1.2536,
-      "step": 2300
-    },
-    {
-      "epoch": 0.78,
-      "learning_rate": 6.788111217641419e-05,
-      "loss": 1.2345,
-      "step": 2400
-    },
-    {
-      "epoch": 0.81,
-      "learning_rate": 6.608341323106423e-05,
-      "loss": 1.1937,
-      "step": 2500
-    },
-    {
-      "epoch": 0.81,
-      "eval_loss": 0.3795304000377655,
-      "eval_runtime": 716.4435,
-      "eval_samples_per_second": 22.25,
-      "eval_steps_per_second": 1.392,
-      "eval_wer": 0.34504806732645216,
-      "step": 2500
-    },
-    {
-      "epoch": 0.84,
-      "learning_rate": 6.428571428571427e-05,
-      "loss": 1.1806,
-      "step": 2600
-    },
-    {
-      "epoch": 0.87,
-      "learning_rate": 6.248801534036433e-05,
-      "loss": 1.1651,
-      "step": 2700
-    },
-    {
-      "epoch": 0.91,
-      "learning_rate": 6.069031639501438e-05,
-      "loss": 1.1455,
-      "step": 2800
-    },
-    {
-      "epoch": 0.94,
-      "learning_rate": 5.889261744966442e-05,
-      "loss": 1.1312,
-      "step": 2900
-    },
-    {
-      "epoch": 0.97,
-      "learning_rate": 5.709491850431447e-05,
-      "loss": 1.1248,
-      "step": 3000
-    },
-    {
-      "epoch": 0.97,
-      "eval_loss": 0.3320523500442505,
-      "eval_runtime": 716.2808,
-      "eval_samples_per_second": 22.255,
-      "eval_steps_per_second": 1.392,
-      "eval_wer": 0.30515830344552264,
-      "step": 3000
-    },
-    {
-      "epoch": 1.0,
-      "learning_rate": 5.5297219558964525e-05,
-      "loss": 1.1017,
-      "step": 3100
-    },
-    {
-      "epoch": 1.04,
-      "learning_rate": 5.3499520613614567e-05,
-      "loss": 1.0978,
-      "step": 3200
-    },
-    {
-      "epoch": 1.07,
-      "learning_rate": 5.1701821668264615e-05,
-      "loss": 1.0954,
-      "step": 3300
-    },
-    {
-      "epoch": 1.1,
-      "learning_rate": 4.990412272291467e-05,
-      "loss": 1.0867,
-      "step": 3400
-    },
-    {
-      "epoch": 1.13,
-      "learning_rate": 4.812440076701821e-05,
-      "loss": 1.0835,
-      "step": 3500
-    },
-    {
-      "epoch": 1.13,
-      "eval_loss": 0.3037940561771393,
-      "eval_runtime": 714.1597,
-      "eval_samples_per_second": 22.321,
-      "eval_steps_per_second": 1.396,
-      "eval_wer": 0.2805032742445413,
-      "step": 3500
-    },
-    {
-      "epoch": 1.17,
-      "learning_rate": 4.632670182166826e-05,
-      "loss": 1.0808,
-      "step": 3600
-    },
-    {
-      "epoch": 1.2,
-      "learning_rate": 4.4529002876318304e-05,
-      "loss": 1.0648,
-      "step": 3700
-    },
-    {
-      "epoch": 1.23,
-      "learning_rate": 4.273130393096836e-05,
-      "loss": 1.0541,
-      "step": 3800
-    },
-    {
-      "epoch": 1.26,
-      "learning_rate": 4.093360498561841e-05,
-      "loss": 1.0621,
-      "step": 3900
-    },
-    {
-      "epoch": 1.3,
-      "learning_rate": 3.913590604026845e-05,
-      "loss": 1.0479,
-      "step": 4000
-    },
-    {
-      "epoch": 1.3,
-      "eval_loss": 0.2910499572753906,
-      "eval_runtime": 718.1665,
-      "eval_samples_per_second": 22.197,
-      "eval_steps_per_second": 1.388,
-      "eval_wer": 0.26888727197800427,
-      "step": 4000
-    },
-    {
-      "epoch": 1.33,
-      "learning_rate": 3.7338207094918506e-05,
-      "loss": 1.0428,
-      "step": 4100
-    },
-    {
-      "epoch": 1.36,
-      "learning_rate": 3.555848513902205e-05,
-      "loss": 1.047,
-      "step": 4200
-    },
-    {
-      "epoch": 1.39,
-      "learning_rate": 3.37607861936721e-05,
-      "loss": 1.0397,
-      "step": 4300
-    },
-    {
-      "epoch": 1.43,
-      "learning_rate": 3.1963087248322145e-05,
-      "loss": 1.0347,
-      "step": 4400
-    },
-    {
-      "epoch": 1.46,
-      "learning_rate": 3.0165388302972194e-05,
-      "loss": 1.0413,
-      "step": 4500
-    },
-    {
-      "epoch": 1.46,
-      "eval_loss": 0.27976545691490173,
-      "eval_runtime": 713.7382,
-      "eval_samples_per_second": 22.335,
-      "eval_steps_per_second": 1.397,
-      "eval_wer": 0.2592995627901586,
-      "step": 4500
-    },
-    {
-      "epoch": 1.49,
-      "learning_rate": 2.836768935762224e-05,
-      "loss": 1.0238,
-      "step": 4600
-    },
-    {
-      "epoch": 1.52,
-      "learning_rate": 2.656999041227229e-05,
-      "loss": 1.0269,
-      "step": 4700
-    },
-    {
-      "epoch": 1.56,
-      "learning_rate": 2.4772291466922337e-05,
-      "loss": 1.021,
-      "step": 4800
-    },
-    {
-      "epoch": 1.59,
-      "learning_rate": 2.2974592521572386e-05,
-      "loss": 1.0186,
-      "step": 4900
-    },
-    {
-      "epoch": 1.62,
-      "learning_rate": 2.1176893576222434e-05,
-      "loss": 1.014,
-      "step": 5000
-    },
-    {
-      "epoch": 1.62,
-      "eval_loss": 0.27265554666519165,
-      "eval_runtime": 707.3075,
-      "eval_samples_per_second": 22.538,
-      "eval_steps_per_second": 1.41,
-      "eval_wer": 0.25117351242409997,
-      "step": 5000
-    },
-    {
-      "epoch": 1.65,
-      "learning_rate": 1.9379194630872483e-05,
-      "loss": 1.0074,
-      "step": 5100
-    },
-    {
-      "epoch": 1.68,
-      "learning_rate": 1.759947267497603e-05,
-      "loss": 1.0193,
-      "step": 5200
-    },
-    {
-      "epoch": 1.72,
-      "learning_rate": 1.5801773729626078e-05,
-      "loss": 1.0044,
-      "step": 5300
-    },
-    {
-      "epoch": 1.75,
-      "learning_rate": 1.4004074784276125e-05,
-      "loss": 1.0005,
-      "step": 5400
-    },
-    {
-      "epoch": 1.78,
-      "learning_rate": 1.2206375838926173e-05,
-      "loss": 1.004,
-      "step": 5500
-    },
-    {
-      "epoch": 1.78,
-      "eval_loss": 0.26460376381874084,
-      "eval_runtime": 719.6956,
-      "eval_samples_per_second": 22.15,
-      "eval_steps_per_second": 1.385,
-      "eval_wer": 0.2470782921128375,
-      "step": 5500
-    },
-    {
-      "epoch": 1.81,
-      "learning_rate": 1.0408676893576222e-05,
-      "loss": 1.0048,
-      "step": 5600
-    },
-    {
-      "epoch": 1.85,
-      "learning_rate": 8.610977948226269e-06,
-      "loss": 0.9988,
-      "step": 5700
-    },
-    {
-      "epoch": 1.88,
-      "learning_rate": 6.813279002876318e-06,
-      "loss": 0.9919,
-      "step": 5800
-    },
-    {
-      "epoch": 1.91,
-      "learning_rate": 5.015580057526366e-06,
-      "loss": 0.9886,
-      "step": 5900
-    },
-    {
-      "epoch": 1.94,
-      "learning_rate": 3.217881112176414e-06,
-      "loss": 0.9949,
-      "step": 6000
-    },
-    {
-      "epoch": 1.94,
-      "eval_loss": 0.26187804341316223,
-      "eval_runtime": 717.4473,
-      "eval_samples_per_second": 22.219,
-      "eval_steps_per_second": 1.39,
-      "eval_wer": 0.24574541380398318,
-      "step": 6000
-    },
-    {
-      "epoch": 1.98,
-      "learning_rate": 1.4201821668264622e-06,
-      "loss": 0.9931,
-      "step": 6100
-    },
-    {
-      "epoch": 2.0,
-      "step": 6172,
-      "total_flos": 1.1573983785360925e+20,
-      "train_loss": 1.788894365302016,
-      "train_runtime": 52105.5599,
-      "train_samples_per_second": 15.163,
-      "train_steps_per_second": 0.118
-    }
-  ],
-  "max_steps": 6172,
-  "num_train_epochs": 2,
-  "total_flos": 1.1573983785360925e+20,
-  "trial_name": null,
-  "trial_params": null
-}