Upload

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

config.json

@@ -0,0 +1,36 @@
+{
+  "_name_or_path": "",
+  "architectures": [
+    "XLMRobertaForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "classifier_dropout": 0.0,
+  "eos_token_id": 2,
+  "finetuning_task": "text-classification",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.0,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 8194,
+  "model_type": "xlm-roberta",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "problem_type": "regression",
+  "torch_dtype": "float32",
+  "transformers_version": "4.39.0",
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "vocab_size": 250002
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0425999578be322709ed81ed4d0b7083dcfdacc3b48ebbafeb89fe1f059f56d7
+size 2271071852

readme.md

@@ -0,0 +1,67 @@
+本模型在数据集[BAAI/IndustryCorpus2](https://huggingface.co/datasets/BAAI/IndustryCorpus2)中用来进行质量评估
+
+- 为什么要筛选低质量的数据
+
+  下面是从数据中抽取的低质量数据，可以看到这种数据对模型的学习是有害无益的
+
+  ```
+  {"text": "\\_\\__\n\nTranslated from *Chinese Journal of Biochemistry and Molecular Biology*, 2007, 23(2): 154--159 \\[译自:中国生物化学与分子生物学报\\]\n"}
+  
+  {"text": "#ifndef _IMGBMP_H_\n#define _IMGBMP_H_\n\n#ifdef __cplusplus\nextern \"C\" {\n#endif\n\nconst uint8_t bmp[]={\n\\/\\/--  调入了一幅图像：D:\\我的文档\\My Pictures\\12864-555.bmp  --*\\/\n\\/\\/--  宽度x高度=128x64  --\n0x00,0x06,0x0A,0xFE,0x0A,0xC6,0x00,0xE0,0x00,0xF0,0x00,0xF8,0x00,0x00,0x00,0x00,\n0x00,0x00,0xFE,0x7D,0xBB,0xC7,0xEF,0xEF,0xEF,0xEF,0xEF,0xEF,0xEF,0xC7,0xBB,0x7D,\n0xFE,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x08,\n0x0C,0xFE,0xFE,0x0C,0x08,0x20,0x60,0xFE,0xFE,0x60,0x20,0x00,0x00,0x00,0x78,0x48,\n0xFE,0x82,0xBA,0xBA,0x82,0xBA,0xBA,0x82,0xBA,0xBA,0x82,0xBA,0xBA,0x82,0xFE,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x01,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0xFE,0xFF,\n0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0xFF,0xFF,0x00,0x00,0xFE,0xFF,0x03,\n0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0xFF,0xFE,0x00,0x00,0x00,0x00,0xC0,0xC0,\n0xC0,0x00,0x00,0x00,0x00,0xFE,0xFF,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,\n0xFF,0xFE,0x00,0x00,0xFE,0xFF,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0x03,0xFF,\n0xFE,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0xFF,0xFF,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0xFF,0xFF,0x00,0x00,0xFF,0xFF,0x0C,\n0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0xFF,0xFF,0x00,0x00,0x00,0x00,0xE1,0xE1,\n0xE1,0x00,0x00,0x00,0x00,0xFF,0xFF,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0xFF,0xFF,0x00,0x00,0xFF,0xFF,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0x0C,0xFF,\n0xFF,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x0F,0x1F,\n0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x1F,0x0F,0x00,0x00,0x0F,0x1F,0x18,\n0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x1F,0x0F,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x0F,0x1F,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,\n0x1F,0x0F,0x00,0x00,0x0F,0x1F,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x1F,\n0x0F,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0xE2,0x92,0x8A,0x86,0x00,0x00,0x7C,0x82,0x82,0x82,0x7C,\n0x00,0xFE,0x00,0x82,0x92,0xAA,0xC6,0x00,0x00,0xC0,0xC0,0x00,0x7C,0x82,0x82,0x82,\n0x7C,0x00,0x00,0x02,0x02,0x02,0xFE,0x00,0x00,0xC0,0xC0,0x00,0x7C,0x82,0x82,0x82,\n0x7C,0x00,0x00,0xFE,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x24,0xA4,0x2E,0x24,0xE4,0x24,0x2E,0xA4,0x24,0x00,0x00,0x00,0xF8,0x4A,0x4C,\n0x48,0xF8,0x48,0x4C,0x4A,0xF8,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0xC0,0x20,0x10,0x10,\n0x10,0x10,0x20,0xC0,0x00,0x00,0xC0,0x20,0x10,0x10,0x10,0x10,0x20,0xC0,0x00,0x00,\n0x00,0x12,0x0A,0x07,0x02,0x7F,0x02,0x07,0x0A,0x12,0x00,0x00,0x00,0x0B,0x0A,0x0A,\n0x0A,0x7F,0x0A,0x0A,0x0A,0x0B,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,\n0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x1F,0x20,0x40,0x40,\n0x40,0x50,0x20,0x5F,0x80,0x00,0x1F,0x20,0x40,0x40,0x40,0x50,0x20,0x5F,0x80,0x00,\n};  \n\n\n#ifdef __cplusplus\n}\n#endif\n\n#endif \\/\\/ _IMGBMP_H_ _SSD1306_16BIT_H_\n"}
+  
+  ```
+
+- 数据构建
+
+  数据来源：随机采样预训练语料
+
+  标签构建：设计数据打分细则，借助LLM模型进行多轮打分，筛选多轮打分分差小于2的数据
+
+  数据规模：20k打分数据，中英文比例1:1
+
+  数据打分prompt
+
+  ```
+  quality_prompt = """Below is an extract from a web page. Evaluate whether the page has a high natural language value and could be useful in an naturanl language task to train a good language model using the additive 5-point scoring system described below. Points are accumulated based on the satisfaction of each criterion:
+  
+  - Zero score if the content contains only some meaningless content or private content, such as some random code, http url or copyright information, personally identifiable information, binary encoding of images.
+  - Add 1 point if the extract provides some basic information, even if it includes some useless contents like advertisements and promotional material.
+  - Add another point if the extract is written in good style, semantically fluent, and free of repetitive content and grammatical errors.
+  - Award a third point tf the extract has relatively complete semantic content, and is written in a good and fluent style, the entire content expresses something related to the same topic, rather than a patchwork of several unrelated items.
+  - A fourth point is awarded if the extract has obvious educational or literary value, or provides a meaningful point or content, contributes to the learning of the topic, and is written in a clear and consistent style. It may be similar to a chapter in a textbook or tutorial, providing a lot of educational content, including exercises and solutions, with little to no superfluous information. The content is coherent and focused, which is valuable for structured learning.
+  - A fifth point is awarded if the extract has outstanding educational value or is of very high information density, provides very high value and meaningful content, does not contain useless information, and is well suited for teaching or knowledge transfer. It contains detailed reasoning, has an easy-to-follow writing style, and can provide deep and thorough insights.
+  
+  
+  The extract:
+  <{EXAMPLE}>.
+  
+  After examining the extract:
+  - Briefly justify your total score, up to 50 words.
+  - Conclude with the score using the format: "Quality score:  <total points>"
+  ...
+  """
+  ```
+
+- 模型训练
+
+  模型选型：与分类模型类似，我们同样使用的是0.5b规模的模型，并对比试验了beg-m3和qwen-0.5b，最终实验显示bge-m3综合表现最优
+
+  模型超参：基座bge-m3，全参数训练，lr=1e-5，batch_size=64, max_length = 2048
+
+  模型评估：在验证集上模型与GPT4对样本质量判定一致率为90%
+
+  ![image-20240919142248242](./img/quality-exp.png)
+
+- 高质量数据带来的训练收益
+
+  我们为了验证高质量的数据是否能带来更高效的训练效率，在同一基座模型下，使用从未筛选之前的50b数据中抽取出高质量数据，可以认为两个数据的分布大体是一致的，进行自回归训练.
+
+  曲线中可以看到，经过高质量数据训练的模型14B的tokens可以达到普通数据50B的模型表现，高质量的数据可以极大的提升训练效率。
+
+  ![image-20240919142732476](./img/quality_train.png)
+
+此外，高质量的数据可以作为预训练的退火阶段的数据加入到模型中，进一步拉升模型效果，为了验证这个猜测，我们在训练行业模型时候，在模型的退火阶段加入了筛选之后高质量数据和部分指令数据转成的预训练数据，可以看到极大提高了模型的表现。
+
+![cpt_two_stage](./img/cpt_two_stage.png)
+
+最后，高质量的预训练语料中包含着丰富的高价值知识性内容，可以从中提取出指令数据进一步提升指令数据的丰富度和知识性，这也催发了[Industry-Instruction](https://huggingface.co/datasets/BAAI/Industry-Instruction)项目的诞生，我们会在那里进行详细的说明。

sentencepiece.bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfc8146abe2a0488e9e2a0c56de7952f7c11ab059eca145a0a727afce0db2865
+size 5069051

special_tokens_map.json

@@ -0,0 +1,51 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_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
+  },
+  "mask_token": {
+    "content": "<mask>",
+    "lstrip": true,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6710678b12670bc442b99edc952c4d996ae309a7020c1fa0096dd245c2faf790
+size 17082821

tokenizer_config.json

@@ -0,0 +1,55 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "250001": {
+      "content": "<mask>",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "mask_token": "<mask>",
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "unk_token": "<unk>"
+}

trainer_state.json

@@ -0,0 +1,1542 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9974025974025974,
+  "eval_steps": 500,
+  "global_step": 216,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 176.64862060546875,
+      "learning_rate": 9.999998677875842e-06,
+      "loss": 11.3275,
+      "step": 1
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 130.30368041992188,
+      "learning_rate": 9.999994711504062e-06,
+      "loss": 6.8023,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 103.31938171386719,
+      "learning_rate": 9.99998810088676e-06,
+      "loss": 4.4755,
+      "step": 3
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 74.88383483886719,
+      "learning_rate": 9.999978846027432e-06,
+      "loss": 2.9368,
+      "step": 4
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 42.227046966552734,
+      "learning_rate": 9.999966946930972e-06,
+      "loss": 1.2319,
+      "step": 5
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 12.707175254821777,
+      "learning_rate": 9.999952403603674e-06,
+      "loss": 0.8505,
+      "step": 6
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 15.292159080505371,
+      "learning_rate": 9.999935216053227e-06,
+      "loss": 0.9937,
+      "step": 7
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 39.702308654785156,
+      "learning_rate": 9.999915384288723e-06,
+      "loss": 1.2296,
+      "step": 8
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 45.62909698486328,
+      "learning_rate": 9.999892908320647e-06,
+      "loss": 1.6743,
+      "step": 9
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 44.297847747802734,
+      "learning_rate": 9.999867788160888e-06,
+      "loss": 1.5054,
+      "step": 10
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 36.63005828857422,
+      "learning_rate": 9.99984002382273e-06,
+      "loss": 1.3223,
+      "step": 11
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 19.54110336303711,
+      "learning_rate": 9.999809615320857e-06,
+      "loss": 0.8807,
+      "step": 12
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 21.871042251586914,
+      "learning_rate": 9.999776562671349e-06,
+      "loss": 1.1686,
+      "step": 13
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 8.16606616973877,
+      "learning_rate": 9.999740865891686e-06,
+      "loss": 0.8842,
+      "step": 14
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.233722686767578,
+      "learning_rate": 9.99970252500075e-06,
+      "loss": 0.6474,
+      "step": 15
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 16.253034591674805,
+      "learning_rate": 9.999661540018812e-06,
+      "loss": 0.8965,
+      "step": 16
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 12.703817367553711,
+      "learning_rate": 9.99961791096755e-06,
+      "loss": 0.8819,
+      "step": 17
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 14.576752662658691,
+      "learning_rate": 9.999571637870035e-06,
+      "loss": 0.8724,
+      "step": 18
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 15.657022476196289,
+      "learning_rate": 9.999522720750743e-06,
+      "loss": 0.7275,
+      "step": 19
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.936721324920654,
+      "learning_rate": 9.999471159635538e-06,
+      "loss": 0.9085,
+      "step": 20
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 12.384129524230957,
+      "learning_rate": 9.999416954551693e-06,
+      "loss": 0.5257,
+      "step": 21
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 7.639270782470703,
+      "learning_rate": 9.999360105527871e-06,
+      "loss": 0.765,
+      "step": 22
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 8.124340057373047,
+      "learning_rate": 9.999300612594139e-06,
+      "loss": 0.6521,
+      "step": 23
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 15.96132755279541,
+      "learning_rate": 9.999238475781957e-06,
+      "loss": 0.8599,
+      "step": 24
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.499070167541504,
+      "learning_rate": 9.999173695124188e-06,
+      "loss": 0.5235,
+      "step": 25
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 13.837294578552246,
+      "learning_rate": 9.999106270655093e-06,
+      "loss": 0.7003,
+      "step": 26
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 8.020848274230957,
+      "learning_rate": 9.999036202410324e-06,
+      "loss": 0.7127,
+      "step": 27
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.126096248626709,
+      "learning_rate": 9.998963490426943e-06,
+      "loss": 0.6233,
+      "step": 28
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 9.910087585449219,
+      "learning_rate": 9.998888134743398e-06,
+      "loss": 0.6812,
+      "step": 29
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 17.2391414642334,
+      "learning_rate": 9.998810135399545e-06,
+      "loss": 0.6481,
+      "step": 30
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 18.45667266845703,
+      "learning_rate": 9.998729492436635e-06,
+      "loss": 0.6672,
+      "step": 31
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 22.11907196044922,
+      "learning_rate": 9.99864620589731e-06,
+      "loss": 0.4814,
+      "step": 32
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 11.804239273071289,
+      "learning_rate": 9.99856027582562e-06,
+      "loss": 0.6971,
+      "step": 33
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.253346920013428,
+      "learning_rate": 9.998471702267007e-06,
+      "loss": 0.4941,
+      "step": 34
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 9.442488670349121,
+      "learning_rate": 9.998380485268317e-06,
+      "loss": 0.5258,
+      "step": 35
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 10.5572509765625,
+      "learning_rate": 9.998286624877786e-06,
+      "loss": 0.5023,
+      "step": 36
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 11.808622360229492,
+      "learning_rate": 9.998190121145056e-06,
+      "loss": 0.5166,
+      "step": 37
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.973758697509766,
+      "learning_rate": 9.99809097412116e-06,
+      "loss": 0.4458,
+      "step": 38
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 7.20625114440918,
+      "learning_rate": 9.997989183858531e-06,
+      "loss": 0.4228,
+      "step": 39
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 13.362131118774414,
+      "learning_rate": 9.997884750411004e-06,
+      "loss": 0.681,
+      "step": 40
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 9.089489936828613,
+      "learning_rate": 9.997777673833807e-06,
+      "loss": 0.4588,
+      "step": 41
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.178775787353516,
+      "learning_rate": 9.997667954183566e-06,
+      "loss": 0.483,
+      "step": 42
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 8.83432674407959,
+      "learning_rate": 9.997555591518307e-06,
+      "loss": 0.4187,
+      "step": 43
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 14.531988143920898,
+      "learning_rate": 9.997440585897455e-06,
+      "loss": 0.4184,
+      "step": 44
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 7.901028633117676,
+      "learning_rate": 9.997322937381829e-06,
+      "loss": 0.7002,
+      "step": 45
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.49567985534668,
+      "learning_rate": 9.997202646033649e-06,
+      "loss": 0.4165,
+      "step": 46
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 32.977596282958984,
+      "learning_rate": 9.997079711916527e-06,
+      "loss": 0.512,
+      "step": 47
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 7.730500221252441,
+      "learning_rate": 9.99695413509548e-06,
+      "loss": 0.5751,
+      "step": 48
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 7.134888172149658,
+      "learning_rate": 9.996825915636918e-06,
+      "loss": 0.5055,
+      "step": 49
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 11.032304763793945,
+      "learning_rate": 9.996695053608651e-06,
+      "loss": 0.5691,
+      "step": 50
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 28.639476776123047,
+      "learning_rate": 9.996561549079886e-06,
+      "loss": 0.4184,
+      "step": 51
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 8.934990882873535,
+      "learning_rate": 9.996425402121224e-06,
+      "loss": 0.4129,
+      "step": 52
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 8.567286491394043,
+      "learning_rate": 9.996286612804666e-06,
+      "loss": 0.5257,
+      "step": 53
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 12.532737731933594,
+      "learning_rate": 9.996145181203616e-06,
+      "loss": 0.4802,
+      "step": 54
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 7.696206092834473,
+      "learning_rate": 9.996001107392864e-06,
+      "loss": 0.5477,
+      "step": 55
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.3067426681518555,
+      "learning_rate": 9.995854391448607e-06,
+      "loss": 0.4384,
+      "step": 56
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.36694860458374,
+      "learning_rate": 9.995705033448435e-06,
+      "loss": 0.483,
+      "step": 57
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 16.650466918945312,
+      "learning_rate": 9.995553033471335e-06,
+      "loss": 0.5578,
+      "step": 58
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 11.836296081542969,
+      "learning_rate": 9.995398391597693e-06,
+      "loss": 0.4419,
+      "step": 59
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.363118648529053,
+      "learning_rate": 9.99524110790929e-06,
+      "loss": 0.424,
+      "step": 60
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 7.7522101402282715,
+      "learning_rate": 9.995081182489306e-06,
+      "loss": 0.4357,
+      "step": 61
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.991522312164307,
+      "learning_rate": 9.99491861542232e-06,
+      "loss": 0.437,
+      "step": 62
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 14.417398452758789,
+      "learning_rate": 9.994753406794303e-06,
+      "loss": 0.6347,
+      "step": 63
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 4.412823677062988,
+      "learning_rate": 9.994585556692624e-06,
+      "loss": 0.2691,
+      "step": 64
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 14.479527473449707,
+      "learning_rate": 9.994415065206053e-06,
+      "loss": 0.4691,
+      "step": 65
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 7.585235595703125,
+      "learning_rate": 9.994241932424755e-06,
+      "loss": 0.4724,
+      "step": 66
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.731332302093506,
+      "learning_rate": 9.99406615844029e-06,
+      "loss": 0.5277,
+      "step": 67
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.436354637145996,
+      "learning_rate": 9.993887743345613e-06,
+      "loss": 0.4632,
+      "step": 68
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.3371992111206055,
+      "learning_rate": 9.993706687235085e-06,
+      "loss": 0.3623,
+      "step": 69
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 10.071105003356934,
+      "learning_rate": 9.993522990204453e-06,
+      "loss": 0.4256,
+      "step": 70
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 12.001503944396973,
+      "learning_rate": 9.993336652350867e-06,
+      "loss": 0.5697,
+      "step": 71
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 10.96662712097168,
+      "learning_rate": 9.993147673772869e-06,
+      "loss": 0.3635,
+      "step": 72
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 8.96445083618164,
+      "learning_rate": 9.992956054570405e-06,
+      "loss": 0.3633,
+      "step": 73
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 10.597489356994629,
+      "learning_rate": 9.99276179484481e-06,
+      "loss": 0.5179,
+      "step": 74
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 37.88457107543945,
+      "learning_rate": 9.992564894698816e-06,
+      "loss": 0.3815,
+      "step": 75
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 10.796125411987305,
+      "learning_rate": 9.992365354236557e-06,
+      "loss": 0.3985,
+      "step": 76
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.9689860343933105,
+      "learning_rate": 9.992163173563559e-06,
+      "loss": 0.3025,
+      "step": 77
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 9.331859588623047,
+      "learning_rate": 9.991958352786744e-06,
+      "loss": 0.3846,
+      "step": 78
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 8.02157211303711,
+      "learning_rate": 9.991750892014433e-06,
+      "loss": 0.421,
+      "step": 79
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 18.640714645385742,
+      "learning_rate": 9.991540791356342e-06,
+      "loss": 0.6442,
+      "step": 80
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.119135856628418,
+      "learning_rate": 9.99132805092358e-06,
+      "loss": 0.4399,
+      "step": 81
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 9.806747436523438,
+      "learning_rate": 9.991112670828657e-06,
+      "loss": 0.3378,
+      "step": 82
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 7.447767734527588,
+      "learning_rate": 9.990894651185475e-06,
+      "loss": 0.4278,
+      "step": 83
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 28.66916847229004,
+      "learning_rate": 9.990673992109335e-06,
+      "loss": 0.4325,
+      "step": 84
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 13.08967399597168,
+      "learning_rate": 9.990450693716933e-06,
+      "loss": 0.3821,
+      "step": 85
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 11.561846733093262,
+      "learning_rate": 9.99022475612636e-06,
+      "loss": 0.4719,
+      "step": 86
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 11.134137153625488,
+      "learning_rate": 9.9899961794571e-06,
+      "loss": 0.4789,
+      "step": 87
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.945651531219482,
+      "learning_rate": 9.989764963830038e-06,
+      "loss": 0.5293,
+      "step": 88
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.415275573730469,
+      "learning_rate": 9.989531109367454e-06,
+      "loss": 0.489,
+      "step": 89
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 14.867259979248047,
+      "learning_rate": 9.989294616193018e-06,
+      "loss": 0.7109,
+      "step": 90
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 18.615076065063477,
+      "learning_rate": 9.989055484431803e-06,
+      "loss": 0.521,
+      "step": 91
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 16.754840850830078,
+      "learning_rate": 9.988813714210272e-06,
+      "loss": 0.4094,
+      "step": 92
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.629899024963379,
+      "learning_rate": 9.988569305656286e-06,
+      "loss": 0.4783,
+      "step": 93
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 8.339197158813477,
+      "learning_rate": 9.988322258899099e-06,
+      "loss": 0.4288,
+      "step": 94
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 8.170295715332031,
+      "learning_rate": 9.988072574069363e-06,
+      "loss": 0.4073,
+      "step": 95
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 9.301414489746094,
+      "learning_rate": 9.987820251299121e-06,
+      "loss": 0.5135,
+      "step": 96
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 24.828275680541992,
+      "learning_rate": 9.98756529072182e-06,
+      "loss": 0.5708,
+      "step": 97
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 8.960503578186035,
+      "learning_rate": 9.987307692472288e-06,
+      "loss": 0.4161,
+      "step": 98
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 8.696434020996094,
+      "learning_rate": 9.98704745668676e-06,
+      "loss": 0.4182,
+      "step": 99
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 9.047277450561523,
+      "learning_rate": 9.986784583502863e-06,
+      "loss": 0.3782,
+      "step": 100
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 15.280583381652832,
+      "learning_rate": 9.986519073059613e-06,
+      "loss": 0.5169,
+      "step": 101
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 11.528756141662598,
+      "learning_rate": 9.986250925497429e-06,
+      "loss": 0.549,
+      "step": 102
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.328488349914551,
+      "learning_rate": 9.98598014095812e-06,
+      "loss": 0.4786,
+      "step": 103
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.289353847503662,
+      "learning_rate": 9.985706719584888e-06,
+      "loss": 0.3513,
+      "step": 104
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 10.800015449523926,
+      "learning_rate": 9.985430661522333e-06,
+      "loss": 0.4872,
+      "step": 105
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 13.80408000946045,
+      "learning_rate": 9.985151966916451e-06,
+      "loss": 0.389,
+      "step": 106
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 13.14911937713623,
+      "learning_rate": 9.984870635914625e-06,
+      "loss": 0.4488,
+      "step": 107
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 8.940996170043945,
+      "learning_rate": 9.984586668665641e-06,
+      "loss": 0.389,
+      "step": 108
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 7.7654924392700195,
+      "learning_rate": 9.984300065319673e-06,
+      "loss": 0.3816,
+      "step": 109
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.716012954711914,
+      "learning_rate": 9.984010826028289e-06,
+      "loss": 0.3724,
+      "step": 110
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 16.961824417114258,
+      "learning_rate": 9.983718950944457e-06,
+      "loss": 0.5116,
+      "step": 111
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 8.058645248413086,
+      "learning_rate": 9.98342444022253e-06,
+      "loss": 0.4868,
+      "step": 112
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 7.157467365264893,
+      "learning_rate": 9.983127294018267e-06,
+      "loss": 0.3658,
+      "step": 113
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.30387544631958,
+      "learning_rate": 9.982827512488809e-06,
+      "loss": 0.422,
+      "step": 114
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.9396748542785645,
+      "learning_rate": 9.982525095792694e-06,
+      "loss": 0.3137,
+      "step": 115
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 9.931533813476562,
+      "learning_rate": 9.98222004408986e-06,
+      "loss": 0.3283,
+      "step": 116
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 8.188447952270508,
+      "learning_rate": 9.981912357541628e-06,
+      "loss": 0.3709,
+      "step": 117
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 7.596179485321045,
+      "learning_rate": 9.981602036310722e-06,
+      "loss": 0.4277,
+      "step": 118
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.1218767166137695,
+      "learning_rate": 9.981289080561253e-06,
+      "loss": 0.338,
+      "step": 119
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 12.419051170349121,
+      "learning_rate": 9.980973490458728e-06,
+      "loss": 0.3293,
+      "step": 120
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 11.043360710144043,
+      "learning_rate": 9.980655266170049e-06,
+      "loss": 0.4129,
+      "step": 121
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 20.372882843017578,
+      "learning_rate": 9.980334407863506e-06,
+      "loss": 0.4247,
+      "step": 122
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 7.637713432312012,
+      "learning_rate": 9.980010915708785e-06,
+      "loss": 0.3427,
+      "step": 123
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 8.155780792236328,
+      "learning_rate": 9.979684789876965e-06,
+      "loss": 0.3592,
+      "step": 124
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 12.149063110351562,
+      "learning_rate": 9.97935603054052e-06,
+      "loss": 0.3915,
+      "step": 125
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 18.71912956237793,
+      "learning_rate": 9.979024637873309e-06,
+      "loss": 0.5225,
+      "step": 126
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 11.921587944030762,
+      "learning_rate": 9.978690612050594e-06,
+      "loss": 0.3912,
+      "step": 127
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 11.384403228759766,
+      "learning_rate": 9.978353953249023e-06,
+      "loss": 0.4185,
+      "step": 128
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 15.501189231872559,
+      "learning_rate": 9.978014661646637e-06,
+      "loss": 0.2584,
+      "step": 129
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.86824893951416,
+      "learning_rate": 9.97767273742287e-06,
+      "loss": 0.2435,
+      "step": 130
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 10.46956729888916,
+      "learning_rate": 9.97732818075855e-06,
+      "loss": 0.2893,
+      "step": 131
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 8.8429594039917,
+      "learning_rate": 9.976980991835896e-06,
+      "loss": 0.3392,
+      "step": 132
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 10.744973182678223,
+      "learning_rate": 9.976631170838516e-06,
+      "loss": 0.381,
+      "step": 133
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 11.55877685546875,
+      "learning_rate": 9.976278717951414e-06,
+      "loss": 0.5731,
+      "step": 134
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.640347480773926,
+      "learning_rate": 9.975923633360985e-06,
+      "loss": 0.3342,
+      "step": 135
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.5956926345825195,
+      "learning_rate": 9.975565917255017e-06,
+      "loss": 0.257,
+      "step": 136
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 9.520169258117676,
+      "learning_rate": 9.975205569822683e-06,
+      "loss": 0.4152,
+      "step": 137
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 12.1378755569458,
+      "learning_rate": 9.974842591254559e-06,
+      "loss": 0.398,
+      "step": 138
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 16.691791534423828,
+      "learning_rate": 9.974476981742602e-06,
+      "loss": 0.4328,
+      "step": 139
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.700257778167725,
+      "learning_rate": 9.974108741480167e-06,
+      "loss": 0.301,
+      "step": 140
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 7.227187156677246,
+      "learning_rate": 9.973737870661995e-06,
+      "loss": 0.3648,
+      "step": 141
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.791788578033447,
+      "learning_rate": 9.973364369484222e-06,
+      "loss": 0.4057,
+      "step": 142
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 12.814080238342285,
+      "learning_rate": 9.972988238144373e-06,
+      "loss": 0.3357,
+      "step": 143
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.2968316078186035,
+      "learning_rate": 9.972609476841368e-06,
+      "loss": 0.3597,
+      "step": 144
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 8.382381439208984,
+      "learning_rate": 9.972228085775512e-06,
+      "loss": 0.276,
+      "step": 145
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 13.944960594177246,
+      "learning_rate": 9.971844065148504e-06,
+      "loss": 0.327,
+      "step": 146
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 18.251310348510742,
+      "learning_rate": 9.971457415163435e-06,
+      "loss": 0.5018,
+      "step": 147
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 16.452730178833008,
+      "learning_rate": 9.971068136024781e-06,
+      "loss": 0.5458,
+      "step": 148
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.843854904174805,
+      "learning_rate": 9.970676227938416e-06,
+      "loss": 0.3638,
+      "step": 149
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 7.829722881317139,
+      "learning_rate": 9.970281691111598e-06,
+      "loss": 0.2724,
+      "step": 150
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 10.796073913574219,
+      "learning_rate": 9.96988452575298e-06,
+      "loss": 0.2902,
+      "step": 151
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 8.34335994720459,
+      "learning_rate": 9.9694847320726e-06,
+      "loss": 0.3067,
+      "step": 152
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.335442066192627,
+      "learning_rate": 9.96908231028189e-06,
+      "loss": 0.3904,
+      "step": 153
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.972692489624023,
+      "learning_rate": 9.968677260593673e-06,
+      "loss": 0.3659,
+      "step": 154
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.458805561065674,
+      "learning_rate": 9.968269583222155e-06,
+      "loss": 0.5201,
+      "step": 155
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 8.518513679504395,
+      "learning_rate": 9.967859278382939e-06,
+      "loss": 0.3055,
+      "step": 156
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 3.929286479949951,
+      "learning_rate": 9.967446346293013e-06,
+      "loss": 0.2564,
+      "step": 157
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 8.628119468688965,
+      "learning_rate": 9.967030787170757e-06,
+      "loss": 0.2978,
+      "step": 158
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.673137187957764,
+      "learning_rate": 9.96661260123594e-06,
+      "loss": 0.4875,
+      "step": 159
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 5.047264099121094,
+      "learning_rate": 9.966191788709716e-06,
+      "loss": 0.3381,
+      "step": 160
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 4.241339206695557,
+      "learning_rate": 9.965768349814635e-06,
+      "loss": 0.2596,
+      "step": 161
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 8.526716232299805,
+      "learning_rate": 9.965342284774633e-06,
+      "loss": 0.2951,
+      "step": 162
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 8.26424503326416,
+      "learning_rate": 9.964913593815032e-06,
+      "loss": 0.4233,
+      "step": 163
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 9.364177703857422,
+      "learning_rate": 9.964482277162547e-06,
+      "loss": 0.4115,
+      "step": 164
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.392032146453857,
+      "learning_rate": 9.964048335045276e-06,
+      "loss": 0.2874,
+      "step": 165
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.744924545288086,
+      "learning_rate": 9.963611767692714e-06,
+      "loss": 0.3342,
+      "step": 166
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.376719951629639,
+      "learning_rate": 9.963172575335735e-06,
+      "loss": 0.3293,
+      "step": 167
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 10.256063461303711,
+      "learning_rate": 9.962730758206612e-06,
+      "loss": 0.3503,
+      "step": 168
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 4.126972198486328,
+      "learning_rate": 9.962286316538992e-06,
+      "loss": 0.1529,
+      "step": 169
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.471617698669434,
+      "learning_rate": 9.961839250567925e-06,
+      "loss": 0.3363,
+      "step": 170
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.28734827041626,
+      "learning_rate": 9.961389560529835e-06,
+      "loss": 0.3863,
+      "step": 171
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 11.83276653289795,
+      "learning_rate": 9.960937246662546e-06,
+      "loss": 0.3687,
+      "step": 172
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 7.140760898590088,
+      "learning_rate": 9.960482309205262e-06,
+      "loss": 0.3402,
+      "step": 173
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 10.465855598449707,
+      "learning_rate": 9.960024748398576e-06,
+      "loss": 0.3292,
+      "step": 174
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 12.055079460144043,
+      "learning_rate": 9.959564564484469e-06,
+      "loss": 0.3464,
+      "step": 175
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 10.145856857299805,
+      "learning_rate": 9.959101757706308e-06,
+      "loss": 0.2928,
+      "step": 176
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 13.143282890319824,
+      "learning_rate": 9.958636328308852e-06,
+      "loss": 0.3878,
+      "step": 177
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 8.187761306762695,
+      "learning_rate": 9.95816827653824e-06,
+      "loss": 0.4443,
+      "step": 178
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 8.577767372131348,
+      "learning_rate": 9.957697602642002e-06,
+      "loss": 0.5333,
+      "step": 179
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.963403224945068,
+      "learning_rate": 9.957224306869053e-06,
+      "loss": 0.3212,
+      "step": 180
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 5.820532321929932,
+      "learning_rate": 9.956748389469697e-06,
+      "loss": 0.4086,
+      "step": 181
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 11.523386001586914,
+      "learning_rate": 9.95626985069562e-06,
+      "loss": 0.391,
+      "step": 182
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 13.058838844299316,
+      "learning_rate": 9.9557886907999e-06,
+      "loss": 0.438,
+      "step": 183
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.208215236663818,
+      "learning_rate": 9.955304910036993e-06,
+      "loss": 0.4455,
+      "step": 184
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.638975620269775,
+      "learning_rate": 9.954818508662754e-06,
+      "loss": 0.5551,
+      "step": 185
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.3545966148376465,
+      "learning_rate": 9.954329486934411e-06,
+      "loss": 0.3912,
+      "step": 186
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.2744140625,
+      "learning_rate": 9.953837845110585e-06,
+      "loss": 0.4615,
+      "step": 187
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.212122440338135,
+      "learning_rate": 9.95334358345128e-06,
+      "loss": 0.3151,
+      "step": 188
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 11.323850631713867,
+      "learning_rate": 9.952846702217886e-06,
+      "loss": 0.4894,
+      "step": 189
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 10.906036376953125,
+      "learning_rate": 9.952347201673181e-06,
+      "loss": 0.4583,
+      "step": 190
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 18.138147354125977,
+      "learning_rate": 9.95184508208132e-06,
+      "loss": 0.5014,
+      "step": 191
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 9.336864471435547,
+      "learning_rate": 9.951340343707852e-06,
+      "loss": 0.4421,
+      "step": 192
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.270864963531494,
+      "learning_rate": 9.95083298681971e-06,
+      "loss": 0.2916,
+      "step": 193
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 10.805733680725098,
+      "learning_rate": 9.950323011685206e-06,
+      "loss": 0.4854,
+      "step": 194
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.747642517089844,
+      "learning_rate": 9.94981041857404e-06,
+      "loss": 0.2705,
+      "step": 195
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.886074542999268,
+      "learning_rate": 9.9492952077573e-06,
+      "loss": 0.2599,
+      "step": 196
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 18.723520278930664,
+      "learning_rate": 9.948777379507453e-06,
+      "loss": 0.5256,
+      "step": 197
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.4744873046875,
+      "learning_rate": 9.948256934098353e-06,
+      "loss": 0.3145,
+      "step": 198
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.738812446594238,
+      "learning_rate": 9.947733871805235e-06,
+      "loss": 0.324,
+      "step": 199
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.052412986755371,
+      "learning_rate": 9.947208192904722e-06,
+      "loss": 0.4534,
+      "step": 200
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 4.8173508644104,
+      "learning_rate": 9.946679897674823e-06,
+      "loss": 0.372,
+      "step": 201
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.492809772491455,
+      "learning_rate": 9.94614898639492e-06,
+      "loss": 0.5313,
+      "step": 202
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.617343425750732,
+      "learning_rate": 9.945615459345789e-06,
+      "loss": 0.4958,
+      "step": 203
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 9.660953521728516,
+      "learning_rate": 9.945079316809585e-06,
+      "loss": 0.3639,
+      "step": 204
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 8.472088813781738,
+      "learning_rate": 9.944540559069847e-06,
+      "loss": 0.3534,
+      "step": 205
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 10.049768447875977,
+      "learning_rate": 9.943999186411496e-06,
+      "loss": 0.4788,
+      "step": 206
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 6.580361366271973,
+      "learning_rate": 9.943455199120836e-06,
+      "loss": 0.3854,
+      "step": 207
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 5.525043487548828,
+      "learning_rate": 9.942908597485558e-06,
+      "loss": 0.3237,
+      "step": 208
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 8.51529598236084,
+      "learning_rate": 9.94235938179473e-06,
+      "loss": 0.2758,
+      "step": 209
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 11.293745994567871,
+      "learning_rate": 9.941807552338805e-06,
+      "loss": 0.319,
+      "step": 210
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 7.503701686859131,
+      "learning_rate": 9.941253109409615e-06,
+      "loss": 0.4216,
+      "step": 211
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 8.137164115905762,
+      "learning_rate": 9.94069605330038e-06,
+      "loss": 0.391,
+      "step": 212
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 5.768956661224365,
+      "learning_rate": 9.940136384305699e-06,
+      "loss": 0.3896,
+      "step": 213
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 6.318339824676514,
+      "learning_rate": 9.939574102721552e-06,
+      "loss": 0.3058,
+      "step": 214
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 5.358760356903076,
+      "learning_rate": 9.939009208845301e-06,
+      "loss": 0.3874,
+      "step": 215
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 7.182950973510742,
+      "learning_rate": 9.938441702975689e-06,
+      "loss": 0.3189,
+      "step": 216
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.3517548441886902,
+      "eval_mse": 0.35175487080665485,
+      "eval_runtime": 13.1487,
+      "eval_samples_per_second": 55.519,
+      "eval_steps_per_second": 27.759,
+      "step": 216
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 4320,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 20,
+  "save_steps": 500,
+  "total_flos": 1.7659591890305024e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e4f435f92db07a89f731afa28e06346a52fda7a409a10c393433d610d8bc101
+size 5240