Upload folder using huggingface_hub

README.md

@@ -0,0 +1,77 @@
+---
+license: bigcode-openrail-m
+base_model: bigcode/starcoderbase
+tags:
+- generated_from_trainer
+datasets:
+- fals3/methods2test_small
+metrics:
+- accuracy
+model-index:
+- name: output
+  results:
+  - task:
+      name: Causal Language Modeling
+      type: text-generation
+    dataset:
+      name: fals3/methods2test_small fm+fc+c+m+f+t+tc
+      type: fals3/methods2test_small
+      args: fm+fc+c+m+f+t+tc
+    metrics:
+    - name: Accuracy
+      type: accuracy
+      value: 0.5611769226558302
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/fals3/methods2test_small/runs/81u023ks)
+# output
+
+This model is a fine-tuned version of [bigcode/starcoderbase](https://huggingface.co/bigcode/starcoderbase) on the fals3/methods2test_small fm+fc+c+m+f+t+tc dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.6457
+- Accuracy: 0.5612
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-05
+- train_batch_size: 1
+- eval_batch_size: 1
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 6
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 48
+- total_eval_batch_size: 6
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- lr_scheduler_warmup_ratio: 0.1
+- num_epochs: 3.0
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 4.41.0.dev0
+- Pytorch 2.2.1+cu118
+- Datasets 2.17.1
+- Tokenizers 0.19.1

all_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 2.985172981878089,
+    "eval_accuracy": 0.5611769226558302,
+    "eval_loss": 0.6457484364509583,
+    "eval_runtime": 38.6111,
+    "eval_samples": 931,
+    "eval_samples_per_second": 24.112,
+    "eval_steps_per_second": 4.04,
+    "perplexity": 1.9074140733989902,
+    "total_flos": 9.99505283128492e+17,
+    "train_loss": 0.5814478738155323,
+    "train_runtime": 61267.4818,
+    "train_samples": 7283,
+    "train_samples_per_second": 0.357,
+    "train_steps_per_second": 0.007
+}

config.json

@@ -0,0 +1,39 @@
+{
+  "_name_or_path": "bigcode/starcoderbase",
+  "activation_function": "gelu",
+  "architectures": [
+    "GPTBigCodeForCausalLM"
+  ],
+  "attention_softmax_in_fp32": true,
+  "attn_pdrop": 0.1,
+  "bos_token_id": 0,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 0,
+  "inference_runner": 0,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "max_batch_size": null,
+  "max_sequence_length": null,
+  "model_type": "gpt_bigcode",
+  "multi_query": true,
+  "n_embd": 6144,
+  "n_head": 48,
+  "n_inner": 24576,
+  "n_layer": 40,
+  "n_positions": 8192,
+  "pad_key_length": true,
+  "pre_allocate_kv_cache": false,
+  "resid_pdrop": 0.1,
+  "scale_attention_softmax_in_fp32": true,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.41.0.dev0",
+  "use_cache": true,
+  "validate_runner_input": true,
+  "vocab_size": 49152
+}

eval_results.json

@@ -0,0 +1,10 @@
+{
+    "epoch": 2.985172981878089,
+    "eval_accuracy": 0.5611769226558302,
+    "eval_loss": 0.6457484364509583,
+    "eval_runtime": 38.6111,
+    "eval_samples": 931,
+    "eval_samples_per_second": 24.112,
+    "eval_steps_per_second": 4.04,
+    "perplexity": 1.9074140733989902
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "transformers_version": "4.41.0.dev0"
+}

model-00001-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b43daf1801e264b5a315c820f95a740c84e50db019fee72a0bc7caf1cd4e9a4
+size 4952185256

model-00002-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1eefd60123f2f0142a6ffdc678cb9422983fca0631ce949d6fd3fe775a77ec2e
+size 4930227792

model-00003-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:652040a6561f1411466092d96f6fc5ce6cf793f2ec6ca95f652db049c97534a2
+size 4927118464

model-00004-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e0251199744ee513dbc79e3645c71ae6782b60c2bcaae0c78855f3ed569e93b
+size 4930227864

model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdb06a918cf4226fbafeb532dd5933bda118e7f750227bd03ff047b1ea84abe9
+size 4927118464

model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36b878bc0c4babd17e338a1029b579bc9944e949d16889960b79d60f6c397218
+size 4930227864

model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ad37739796bac00e588d9ba3dcb70479a4bdff82da2a72b97b385956aacd480
+size 1437858576

model.safetensors.index.json

@@ -0,0 +1,491 @@
+{
+  "metadata": {
+    "total_size": 31034912768
+  },
+  "weight_map": {
+    "transformer.h.0.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.0.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.0.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.0.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.0.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.0.mlp.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.0.mlp.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.1.mlp.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.1.mlp.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.10.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.10.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.10.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.10.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.10.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.10.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.10.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.11.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.11.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.12.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.12.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.12.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.12.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.12.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.12.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.12.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.12.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.12.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.12.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.12.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.12.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.13.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.13.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.14.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.14.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.15.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.15.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.16.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.16.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.17.mlp.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.17.mlp.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.attn.c_attn.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.18.attn.c_attn.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.attn.c_proj.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.18.attn.c_proj.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.ln_1.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.18.ln_1.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.ln_2.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.18.ln_2.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.mlp.c_fc.bias": "model-00003-of-00007.safetensors",
+    "transformer.h.18.mlp.c_fc.weight": "model-00003-of-00007.safetensors",
+    "transformer.h.18.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.18.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.19.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.19.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.2.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.2.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.2.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.2.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.2.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.2.mlp.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.2.mlp.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.20.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.20.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.20.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.20.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.20.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.20.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.20.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.21.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.21.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.22.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.22.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.23.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.23.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.attn.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.attn.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.ln_2.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.ln_2.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.mlp.c_fc.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.mlp.c_fc.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.24.mlp.c_proj.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.24.mlp.c_proj.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.25.attn.c_attn.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.25.attn.c_attn.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.25.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.25.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.25.ln_1.bias": "model-00004-of-00007.safetensors",
+    "transformer.h.25.ln_1.weight": "model-00004-of-00007.safetensors",
+    "transformer.h.25.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.25.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.25.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.25.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.25.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.25.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.26.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.26.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.27.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.27.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.28.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.28.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.29.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.29.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.3.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.3.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.3.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.3.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.3.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.3.mlp.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.3.mlp.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.30.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.30.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.30.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.30.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.30.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.30.mlp.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.30.mlp.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.attn.c_attn.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.31.attn.c_attn.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.attn.c_proj.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.31.attn.c_proj.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.ln_1.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.31.ln_1.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.ln_2.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.31.ln_2.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.mlp.c_fc.bias": "model-00005-of-00007.safetensors",
+    "transformer.h.31.mlp.c_fc.weight": "model-00005-of-00007.safetensors",
+    "transformer.h.31.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.31.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.32.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.32.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.33.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.33.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.34.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.34.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.35.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.35.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.36.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.36.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.attn.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.attn.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.ln_2.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.ln_2.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.mlp.c_fc.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.mlp.c_fc.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.37.mlp.c_proj.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.37.mlp.c_proj.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.38.attn.c_attn.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.38.attn.c_attn.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.38.attn.c_proj.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.38.attn.c_proj.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.38.ln_1.bias": "model-00006-of-00007.safetensors",
+    "transformer.h.38.ln_1.weight": "model-00006-of-00007.safetensors",
+    "transformer.h.38.ln_2.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.38.ln_2.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.38.mlp.c_fc.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.38.mlp.c_fc.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.38.mlp.c_proj.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.38.mlp.c_proj.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.attn.c_attn.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.attn.c_attn.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.attn.c_proj.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.attn.c_proj.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.ln_1.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.ln_1.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.ln_2.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.ln_2.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.mlp.c_fc.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.mlp.c_fc.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.39.mlp.c_proj.bias": "model-00007-of-00007.safetensors",
+    "transformer.h.39.mlp.c_proj.weight": "model-00007-of-00007.safetensors",
+    "transformer.h.4.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.4.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.4.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.4.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.4.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.4.mlp.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.4.mlp.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.attn.c_attn.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.5.attn.c_attn.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.attn.c_proj.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.5.attn.c_proj.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.ln_1.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.5.ln_1.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.ln_2.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.5.ln_2.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.mlp.c_fc.bias": "model-00001-of-00007.safetensors",
+    "transformer.h.5.mlp.c_fc.weight": "model-00001-of-00007.safetensors",
+    "transformer.h.5.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.5.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.6.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.6.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.7.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.7.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.8.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.8.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.attn.c_attn.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.attn.c_attn.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.attn.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.attn.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.ln_1.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.ln_1.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.ln_2.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.ln_2.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.mlp.c_fc.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.mlp.c_fc.weight": "model-00002-of-00007.safetensors",
+    "transformer.h.9.mlp.c_proj.bias": "model-00002-of-00007.safetensors",
+    "transformer.h.9.mlp.c_proj.weight": "model-00002-of-00007.safetensors",
+    "transformer.ln_f.bias": "model-00007-of-00007.safetensors",
+    "transformer.ln_f.weight": "model-00007-of-00007.safetensors",
+    "transformer.wpe.weight": "model-00001-of-00007.safetensors",
+    "transformer.wte.weight": "model-00001-of-00007.safetensors"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,45 @@
+{
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<|endoftext|>",
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,186 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<fim_prefix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<fim_middle>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<fim_suffix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "<fim_pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "<filename>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "<gh_stars>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "7": {
+      "content": "<issue_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "8": {
+      "content": "<issue_comment>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "9": {
+      "content": "<issue_closed>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "10": {
+      "content": "<jupyter_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "11": {
+      "content": "<jupyter_text>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "12": {
+      "content": "<jupyter_code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "13": {
+      "content": "<jupyter_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "14": {
+      "content": "<empty_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "15": {
+      "content": "<commit_before>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "16": {
+      "content": "<commit_msg>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "17": {
+      "content": "<commit_after>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "18": {
+      "content": "<reponame>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "<fim_prefix>",
+    "<fim_middle>",
+    "<fim_suffix>",
+    "<fim_pad>",
+    "<filename>",
+    "<gh_stars>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>"
+  ],
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "vocab_size": 49152
+}

train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 2.985172981878089,
+    "total_flos": 9.99505283128492e+17,
+    "train_loss": 0.5814478738155323,
+    "train_runtime": 61267.4818,
+    "train_samples": 7283,
+    "train_samples_per_second": 0.357,
+    "train_steps_per_second": 0.007
+}

trainer_state.json

@@ -0,0 +1,3213 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.985172981878089,
+  "eval_steps": 500,
+  "global_step": 453,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.006589785831960461,
+      "grad_norm": 57166.33984375,
+      "learning_rate": 0.0,
+      "loss": 0.9208,
+      "step": 1
+    },
+    {
+      "epoch": 0.013179571663920923,
+      "grad_norm": 24538.927734375,
+      "learning_rate": 1.0869565217391306e-06,
+      "loss": 0.7281,
+      "step": 2
+    },
+    {
+      "epoch": 0.019769357495881382,
+      "grad_norm": 15162.59375,
+      "learning_rate": 2.173913043478261e-06,
+      "loss": 0.7891,
+      "step": 3
+    },
+    {
+      "epoch": 0.026359143327841845,
+      "grad_norm": 19609.22265625,
+      "learning_rate": 3.2608695652173914e-06,
+      "loss": 0.7988,
+      "step": 4
+    },
+    {
+      "epoch": 0.032948929159802305,
+      "grad_norm": 36443.21484375,
+      "learning_rate": 4.347826086956522e-06,
+      "loss": 0.7955,
+      "step": 5
+    },
+    {
+      "epoch": 0.039538714991762765,
+      "grad_norm": 5708.90673828125,
+      "learning_rate": 5.4347826086956525e-06,
+      "loss": 0.756,
+      "step": 6
+    },
+    {
+      "epoch": 0.04612850082372323,
+      "grad_norm": 16939.138671875,
+      "learning_rate": 6.521739130434783e-06,
+      "loss": 0.8382,
+      "step": 7
+    },
+    {
+      "epoch": 0.05271828665568369,
+      "grad_norm": 9426.6240234375,
+      "learning_rate": 7.608695652173914e-06,
+      "loss": 0.8287,
+      "step": 8
+    },
+    {
+      "epoch": 0.05930807248764415,
+      "grad_norm": 35329.375,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 0.7193,
+      "step": 9
+    },
+    {
+      "epoch": 0.06589785831960461,
+      "grad_norm": 25405.27734375,
+      "learning_rate": 9.782608695652175e-06,
+      "loss": 0.7123,
+      "step": 10
+    },
+    {
+      "epoch": 0.07248764415156507,
+      "grad_norm": 39777.97265625,
+      "learning_rate": 1.0869565217391305e-05,
+      "loss": 0.8391,
+      "step": 11
+    },
+    {
+      "epoch": 0.07907742998352553,
+      "grad_norm": 32360.591796875,
+      "learning_rate": 1.1956521739130435e-05,
+      "loss": 0.7638,
+      "step": 12
+    },
+    {
+      "epoch": 0.085667215815486,
+      "grad_norm": 20279.001953125,
+      "learning_rate": 1.3043478260869566e-05,
+      "loss": 0.8044,
+      "step": 13
+    },
+    {
+      "epoch": 0.09225700164744646,
+      "grad_norm": 21474.5,
+      "learning_rate": 1.4130434782608694e-05,
+      "loss": 0.6984,
+      "step": 14
+    },
+    {
+      "epoch": 0.09884678747940692,
+      "grad_norm": 24501.21484375,
+      "learning_rate": 1.5217391304347828e-05,
+      "loss": 0.7501,
+      "step": 15
+    },
+    {
+      "epoch": 0.10543657331136738,
+      "grad_norm": 16850.5625,
+      "learning_rate": 1.630434782608696e-05,
+      "loss": 0.7073,
+      "step": 16
+    },
+    {
+      "epoch": 0.11202635914332784,
+      "grad_norm": 18344.18359375,
+      "learning_rate": 1.739130434782609e-05,
+      "loss": 0.7623,
+      "step": 17
+    },
+    {
+      "epoch": 0.1186161449752883,
+      "grad_norm": 7098.125,
+      "learning_rate": 1.8478260869565216e-05,
+      "loss": 0.6343,
+      "step": 18
+    },
+    {
+      "epoch": 0.12520593080724876,
+      "grad_norm": 15517.8115234375,
+      "learning_rate": 1.956521739130435e-05,
+      "loss": 0.6598,
+      "step": 19
+    },
+    {
+      "epoch": 0.13179571663920922,
+      "grad_norm": 111118.3046875,
+      "learning_rate": 2.065217391304348e-05,
+      "loss": 0.6546,
+      "step": 20
+    },
+    {
+      "epoch": 0.13838550247116968,
+      "grad_norm": 22035.267578125,
+      "learning_rate": 2.173913043478261e-05,
+      "loss": 0.6148,
+      "step": 21
+    },
+    {
+      "epoch": 0.14497528830313014,
+      "grad_norm": 21264.662109375,
+      "learning_rate": 2.282608695652174e-05,
+      "loss": 0.6115,
+      "step": 22
+    },
+    {
+      "epoch": 0.1515650741350906,
+      "grad_norm": 10420.123046875,
+      "learning_rate": 2.391304347826087e-05,
+      "loss": 0.6716,
+      "step": 23
+    },
+    {
+      "epoch": 0.15815485996705106,
+      "grad_norm": 31856.8828125,
+      "learning_rate": 2.5e-05,
+      "loss": 0.6439,
+      "step": 24
+    },
+    {
+      "epoch": 0.16474464579901152,
+      "grad_norm": 33552.83984375,
+      "learning_rate": 2.608695652173913e-05,
+      "loss": 0.6781,
+      "step": 25
+    },
+    {
+      "epoch": 0.171334431630972,
+      "grad_norm": 8155.12451171875,
+      "learning_rate": 2.7173913043478262e-05,
+      "loss": 0.6042,
+      "step": 26
+    },
+    {
+      "epoch": 0.17792421746293247,
+      "grad_norm": 11283.130859375,
+      "learning_rate": 2.826086956521739e-05,
+      "loss": 0.6639,
+      "step": 27
+    },
+    {
+      "epoch": 0.18451400329489293,
+      "grad_norm": 56492.08984375,
+      "learning_rate": 2.9347826086956526e-05,
+      "loss": 0.6198,
+      "step": 28
+    },
+    {
+      "epoch": 0.19110378912685339,
+      "grad_norm": 36276.30859375,
+      "learning_rate": 3.0434782608695656e-05,
+      "loss": 0.7941,
+      "step": 29
+    },
+    {
+      "epoch": 0.19769357495881384,
+      "grad_norm": 16123.35546875,
+      "learning_rate": 3.152173913043479e-05,
+      "loss": 0.6976,
+      "step": 30
+    },
+    {
+      "epoch": 0.2042833607907743,
+      "grad_norm": 3790.1572265625,
+      "learning_rate": 3.260869565217392e-05,
+      "loss": 0.6401,
+      "step": 31
+    },
+    {
+      "epoch": 0.21087314662273476,
+      "grad_norm": 20178.978515625,
+      "learning_rate": 3.369565217391305e-05,
+      "loss": 0.7029,
+      "step": 32
+    },
+    {
+      "epoch": 0.21746293245469522,
+      "grad_norm": 12479.48046875,
+      "learning_rate": 3.478260869565218e-05,
+      "loss": 0.7024,
+      "step": 33
+    },
+    {
+      "epoch": 0.22405271828665568,
+      "grad_norm": 17013.83984375,
+      "learning_rate": 3.58695652173913e-05,
+      "loss": 0.7124,
+      "step": 34
+    },
+    {
+      "epoch": 0.23064250411861614,
+      "grad_norm": 8563.2705078125,
+      "learning_rate": 3.695652173913043e-05,
+      "loss": 0.6505,
+      "step": 35
+    },
+    {
+      "epoch": 0.2372322899505766,
+      "grad_norm": 8254.0205078125,
+      "learning_rate": 3.804347826086957e-05,
+      "loss": 0.6654,
+      "step": 36
+    },
+    {
+      "epoch": 0.24382207578253706,
+      "grad_norm": 49902.2890625,
+      "learning_rate": 3.91304347826087e-05,
+      "loss": 0.6191,
+      "step": 37
+    },
+    {
+      "epoch": 0.2504118616144975,
+      "grad_norm": 18592.572265625,
+      "learning_rate": 4.021739130434783e-05,
+      "loss": 0.6082,
+      "step": 38
+    },
+    {
+      "epoch": 0.257001647446458,
+      "grad_norm": 25148.34375,
+      "learning_rate": 4.130434782608696e-05,
+      "loss": 0.7318,
+      "step": 39
+    },
+    {
+      "epoch": 0.26359143327841844,
+      "grad_norm": 11293.0634765625,
+      "learning_rate": 4.239130434782609e-05,
+      "loss": 0.6075,
+      "step": 40
+    },
+    {
+      "epoch": 0.2701812191103789,
+      "grad_norm": 10365.61328125,
+      "learning_rate": 4.347826086956522e-05,
+      "loss": 0.5871,
+      "step": 41
+    },
+    {
+      "epoch": 0.27677100494233936,
+      "grad_norm": 6548.3740234375,
+      "learning_rate": 4.456521739130435e-05,
+      "loss": 0.5931,
+      "step": 42
+    },
+    {
+      "epoch": 0.2833607907742998,
+      "grad_norm": 17649.12109375,
+      "learning_rate": 4.565217391304348e-05,
+      "loss": 0.7389,
+      "step": 43
+    },
+    {
+      "epoch": 0.2899505766062603,
+      "grad_norm": 14737.7197265625,
+      "learning_rate": 4.673913043478261e-05,
+      "loss": 0.5972,
+      "step": 44
+    },
+    {
+      "epoch": 0.29654036243822074,
+      "grad_norm": 9264.072265625,
+      "learning_rate": 4.782608695652174e-05,
+      "loss": 0.6256,
+      "step": 45
+    },
+    {
+      "epoch": 0.3031301482701812,
+      "grad_norm": 86530.90625,
+      "learning_rate": 4.891304347826087e-05,
+      "loss": 0.6694,
+      "step": 46
+    },
+    {
+      "epoch": 0.30971993410214166,
+      "grad_norm": 10428.66015625,
+      "learning_rate": 5e-05,
+      "loss": 0.6411,
+      "step": 47
+    },
+    {
+      "epoch": 0.3163097199341021,
+      "grad_norm": 35924.46484375,
+      "learning_rate": 4.987714987714988e-05,
+      "loss": 0.7293,
+      "step": 48
+    },
+    {
+      "epoch": 0.3228995057660626,
+      "grad_norm": 9777.244140625,
+      "learning_rate": 4.9754299754299756e-05,
+      "loss": 0.6293,
+      "step": 49
+    },
+    {
+      "epoch": 0.32948929159802304,
+      "grad_norm": 19732.7421875,
+      "learning_rate": 4.963144963144963e-05,
+      "loss": 0.686,
+      "step": 50
+    },
+    {
+      "epoch": 0.33607907742998355,
+      "grad_norm": 35202.5234375,
+      "learning_rate": 4.950859950859951e-05,
+      "loss": 0.6859,
+      "step": 51
+    },
+    {
+      "epoch": 0.342668863261944,
+      "grad_norm": 12520.201171875,
+      "learning_rate": 4.9385749385749387e-05,
+      "loss": 0.6202,
+      "step": 52
+    },
+    {
+      "epoch": 0.34925864909390447,
+      "grad_norm": 34133.8671875,
+      "learning_rate": 4.926289926289926e-05,
+      "loss": 0.6566,
+      "step": 53
+    },
+    {
+      "epoch": 0.35584843492586493,
+      "grad_norm": 30663.423828125,
+      "learning_rate": 4.914004914004915e-05,
+      "loss": 0.607,
+      "step": 54
+    },
+    {
+      "epoch": 0.3624382207578254,
+      "grad_norm": 33234.12109375,
+      "learning_rate": 4.901719901719902e-05,
+      "loss": 0.5683,
+      "step": 55
+    },
+    {
+      "epoch": 0.36902800658978585,
+      "grad_norm": 11093.591796875,
+      "learning_rate": 4.8894348894348894e-05,
+      "loss": 0.6449,
+      "step": 56
+    },
+    {
+      "epoch": 0.3756177924217463,
+      "grad_norm": 10510.125,
+      "learning_rate": 4.877149877149878e-05,
+      "loss": 0.6932,
+      "step": 57
+    },
+    {
+      "epoch": 0.38220757825370677,
+      "grad_norm": 76218.2109375,
+      "learning_rate": 4.8648648648648654e-05,
+      "loss": 0.681,
+      "step": 58
+    },
+    {
+      "epoch": 0.38879736408566723,
+      "grad_norm": 25486.68359375,
+      "learning_rate": 4.8525798525798524e-05,
+      "loss": 0.577,
+      "step": 59
+    },
+    {
+      "epoch": 0.3953871499176277,
+      "grad_norm": 39751.8671875,
+      "learning_rate": 4.840294840294841e-05,
+      "loss": 0.6199,
+      "step": 60
+    },
+    {
+      "epoch": 0.40197693574958815,
+      "grad_norm": 14526.541015625,
+      "learning_rate": 4.8280098280098285e-05,
+      "loss": 0.5628,
+      "step": 61
+    },
+    {
+      "epoch": 0.4085667215815486,
+      "grad_norm": 33324.5625,
+      "learning_rate": 4.8157248157248155e-05,
+      "loss": 0.7594,
+      "step": 62
+    },
+    {
+      "epoch": 0.41515650741350907,
+      "grad_norm": 13729.7373046875,
+      "learning_rate": 4.803439803439804e-05,
+      "loss": 0.6227,
+      "step": 63
+    },
+    {
+      "epoch": 0.42174629324546953,
+      "grad_norm": 34946.58203125,
+      "learning_rate": 4.7911547911547915e-05,
+      "loss": 0.6377,
+      "step": 64
+    },
+    {
+      "epoch": 0.42833607907743,
+      "grad_norm": 7339.2587890625,
+      "learning_rate": 4.778869778869779e-05,
+      "loss": 0.7474,
+      "step": 65
+    },
+    {
+      "epoch": 0.43492586490939045,
+      "grad_norm": 61978.32421875,
+      "learning_rate": 4.766584766584767e-05,
+      "loss": 0.7136,
+      "step": 66
+    },
+    {
+      "epoch": 0.4415156507413509,
+      "grad_norm": 18481.29296875,
+      "learning_rate": 4.7542997542997546e-05,
+      "loss": 0.64,
+      "step": 67
+    },
+    {
+      "epoch": 0.44810543657331137,
+      "grad_norm": 11941.44140625,
+      "learning_rate": 4.742014742014742e-05,
+      "loss": 0.6273,
+      "step": 68
+    },
+    {
+      "epoch": 0.4546952224052718,
+      "grad_norm": 70506.828125,
+      "learning_rate": 4.72972972972973e-05,
+      "loss": 0.7494,
+      "step": 69
+    },
+    {
+      "epoch": 0.4612850082372323,
+      "grad_norm": 8327.736328125,
+      "learning_rate": 4.7174447174447176e-05,
+      "loss": 0.6674,
+      "step": 70
+    },
+    {
+      "epoch": 0.46787479406919275,
+      "grad_norm": 13985.94140625,
+      "learning_rate": 4.705159705159705e-05,
+      "loss": 0.6474,
+      "step": 71
+    },
+    {
+      "epoch": 0.4744645799011532,
+      "grad_norm": 9354.6904296875,
+      "learning_rate": 4.692874692874693e-05,
+      "loss": 0.663,
+      "step": 72
+    },
+    {
+      "epoch": 0.48105436573311366,
+      "grad_norm": 30927.59765625,
+      "learning_rate": 4.680589680589681e-05,
+      "loss": 0.6372,
+      "step": 73
+    },
+    {
+      "epoch": 0.4876441515650741,
+      "grad_norm": 32827.7109375,
+      "learning_rate": 4.6683046683046684e-05,
+      "loss": 0.623,
+      "step": 74
+    },
+    {
+      "epoch": 0.4942339373970346,
+      "grad_norm": 30631.48046875,
+      "learning_rate": 4.656019656019656e-05,
+      "loss": 0.6329,
+      "step": 75
+    },
+    {
+      "epoch": 0.500823723228995,
+      "grad_norm": 23105.517578125,
+      "learning_rate": 4.6437346437346444e-05,
+      "loss": 0.719,
+      "step": 76
+    },
+    {
+      "epoch": 0.5074135090609555,
+      "grad_norm": 21652.029296875,
+      "learning_rate": 4.6314496314496314e-05,
+      "loss": 0.7059,
+      "step": 77
+    },
+    {
+      "epoch": 0.514003294892916,
+      "grad_norm": 15986.599609375,
+      "learning_rate": 4.619164619164619e-05,
+      "loss": 0.6683,
+      "step": 78
+    },
+    {
+      "epoch": 0.5205930807248764,
+      "grad_norm": 20643.82421875,
+      "learning_rate": 4.6068796068796074e-05,
+      "loss": 0.7123,
+      "step": 79
+    },
+    {
+      "epoch": 0.5271828665568369,
+      "grad_norm": 24529.80078125,
+      "learning_rate": 4.594594594594595e-05,
+      "loss": 0.6836,
+      "step": 80
+    },
+    {
+      "epoch": 0.5337726523887973,
+      "grad_norm": 19665.833984375,
+      "learning_rate": 4.582309582309582e-05,
+      "loss": 0.6292,
+      "step": 81
+    },
+    {
+      "epoch": 0.5403624382207578,
+      "grad_norm": 35243.45703125,
+      "learning_rate": 4.5700245700245705e-05,
+      "loss": 0.6891,
+      "step": 82
+    },
+    {
+      "epoch": 0.5469522240527183,
+      "grad_norm": 15323.2197265625,
+      "learning_rate": 4.557739557739558e-05,
+      "loss": 0.6287,
+      "step": 83
+    },
+    {
+      "epoch": 0.5535420098846787,
+      "grad_norm": 21972.83203125,
+      "learning_rate": 4.545454545454546e-05,
+      "loss": 0.676,
+      "step": 84
+    },
+    {
+      "epoch": 0.5601317957166392,
+      "grad_norm": 38221.46484375,
+      "learning_rate": 4.5331695331695335e-05,
+      "loss": 0.7015,
+      "step": 85
+    },
+    {
+      "epoch": 0.5667215815485996,
+      "grad_norm": 17371.90625,
+      "learning_rate": 4.520884520884521e-05,
+      "loss": 0.6665,
+      "step": 86
+    },
+    {
+      "epoch": 0.5733113673805601,
+      "grad_norm": 23913.32421875,
+      "learning_rate": 4.508599508599509e-05,
+      "loss": 0.6641,
+      "step": 87
+    },
+    {
+      "epoch": 0.5799011532125206,
+      "grad_norm": 14188.1494140625,
+      "learning_rate": 4.4963144963144966e-05,
+      "loss": 0.6153,
+      "step": 88
+    },
+    {
+      "epoch": 0.586490939044481,
+      "grad_norm": 15268.400390625,
+      "learning_rate": 4.484029484029484e-05,
+      "loss": 0.7959,
+      "step": 89
+    },
+    {
+      "epoch": 0.5930807248764415,
+      "grad_norm": 85963.09375,
+      "learning_rate": 4.471744471744472e-05,
+      "loss": 0.7464,
+      "step": 90
+    },
+    {
+      "epoch": 0.5996705107084019,
+      "grad_norm": 48107.12890625,
+      "learning_rate": 4.4594594594594596e-05,
+      "loss": 0.6248,
+      "step": 91
+    },
+    {
+      "epoch": 0.6062602965403624,
+      "grad_norm": 32619.958984375,
+      "learning_rate": 4.447174447174447e-05,
+      "loss": 0.6176,
+      "step": 92
+    },
+    {
+      "epoch": 0.6128500823723229,
+      "grad_norm": 13193.05078125,
+      "learning_rate": 4.434889434889435e-05,
+      "loss": 0.7016,
+      "step": 93
+    },
+    {
+      "epoch": 0.6194398682042833,
+      "grad_norm": 242728.328125,
+      "learning_rate": 4.422604422604423e-05,
+      "loss": 0.6053,
+      "step": 94
+    },
+    {
+      "epoch": 0.6260296540362438,
+      "grad_norm": 13395.2021484375,
+      "learning_rate": 4.4103194103194104e-05,
+      "loss": 0.6552,
+      "step": 95
+    },
+    {
+      "epoch": 0.6326194398682042,
+      "grad_norm": 26586.83984375,
+      "learning_rate": 4.398034398034398e-05,
+      "loss": 0.6451,
+      "step": 96
+    },
+    {
+      "epoch": 0.6392092257001647,
+      "grad_norm": 20530.763671875,
+      "learning_rate": 4.385749385749386e-05,
+      "loss": 0.6142,
+      "step": 97
+    },
+    {
+      "epoch": 0.6457990115321252,
+      "grad_norm": 120099.6875,
+      "learning_rate": 4.373464373464374e-05,
+      "loss": 0.7366,
+      "step": 98
+    },
+    {
+      "epoch": 0.6523887973640856,
+      "grad_norm": 32992.8359375,
+      "learning_rate": 4.361179361179362e-05,
+      "loss": 0.6341,
+      "step": 99
+    },
+    {
+      "epoch": 0.6589785831960461,
+      "grad_norm": 114159.0625,
+      "learning_rate": 4.348894348894349e-05,
+      "loss": 0.6375,
+      "step": 100
+    },
+    {
+      "epoch": 0.6655683690280065,
+      "grad_norm": 59862.4453125,
+      "learning_rate": 4.336609336609337e-05,
+      "loss": 0.642,
+      "step": 101
+    },
+    {
+      "epoch": 0.6721581548599671,
+      "grad_norm": 22114.5078125,
+      "learning_rate": 4.324324324324325e-05,
+      "loss": 0.6883,
+      "step": 102
+    },
+    {
+      "epoch": 0.6787479406919276,
+      "grad_norm": 63335.8203125,
+      "learning_rate": 4.312039312039312e-05,
+      "loss": 0.695,
+      "step": 103
+    },
+    {
+      "epoch": 0.685337726523888,
+      "grad_norm": 8014.02001953125,
+      "learning_rate": 4.2997542997543e-05,
+      "loss": 0.668,
+      "step": 104
+    },
+    {
+      "epoch": 0.6919275123558485,
+      "grad_norm": 35121.1640625,
+      "learning_rate": 4.287469287469288e-05,
+      "loss": 0.5207,
+      "step": 105
+    },
+    {
+      "epoch": 0.6985172981878089,
+      "grad_norm": 16844.70703125,
+      "learning_rate": 4.2751842751842756e-05,
+      "loss": 0.7609,
+      "step": 106
+    },
+    {
+      "epoch": 0.7051070840197694,
+      "grad_norm": 12384.4345703125,
+      "learning_rate": 4.262899262899263e-05,
+      "loss": 0.6501,
+      "step": 107
+    },
+    {
+      "epoch": 0.7116968698517299,
+      "grad_norm": 5228.11572265625,
+      "learning_rate": 4.250614250614251e-05,
+      "loss": 0.6624,
+      "step": 108
+    },
+    {
+      "epoch": 0.7182866556836903,
+      "grad_norm": 112343.34375,
+      "learning_rate": 4.2383292383292386e-05,
+      "loss": 0.6711,
+      "step": 109
+    },
+    {
+      "epoch": 0.7248764415156508,
+      "grad_norm": 11115.724609375,
+      "learning_rate": 4.226044226044226e-05,
+      "loss": 0.668,
+      "step": 110
+    },
+    {
+      "epoch": 0.7314662273476112,
+      "grad_norm": 91624.140625,
+      "learning_rate": 4.213759213759214e-05,
+      "loss": 0.6913,
+      "step": 111
+    },
+    {
+      "epoch": 0.7380560131795717,
+      "grad_norm": 6046.64794921875,
+      "learning_rate": 4.2014742014742017e-05,
+      "loss": 0.6699,
+      "step": 112
+    },
+    {
+      "epoch": 0.7446457990115322,
+      "grad_norm": 26374.94921875,
+      "learning_rate": 4.189189189189189e-05,
+      "loss": 0.7146,
+      "step": 113
+    },
+    {
+      "epoch": 0.7512355848434926,
+      "grad_norm": 27798.625,
+      "learning_rate": 4.176904176904177e-05,
+      "loss": 0.6188,
+      "step": 114
+    },
+    {
+      "epoch": 0.7578253706754531,
+      "grad_norm": 8118.2197265625,
+      "learning_rate": 4.164619164619165e-05,
+      "loss": 0.6177,
+      "step": 115
+    },
+    {
+      "epoch": 0.7644151565074135,
+      "grad_norm": 40166.359375,
+      "learning_rate": 4.1523341523341524e-05,
+      "loss": 0.66,
+      "step": 116
+    },
+    {
+      "epoch": 0.771004942339374,
+      "grad_norm": 16126.8427734375,
+      "learning_rate": 4.14004914004914e-05,
+      "loss": 0.7109,
+      "step": 117
+    },
+    {
+      "epoch": 0.7775947281713345,
+      "grad_norm": 18022.84765625,
+      "learning_rate": 4.127764127764128e-05,
+      "loss": 0.6224,
+      "step": 118
+    },
+    {
+      "epoch": 0.7841845140032949,
+      "grad_norm": 15072.46484375,
+      "learning_rate": 4.1154791154791154e-05,
+      "loss": 0.5862,
+      "step": 119
+    },
+    {
+      "epoch": 0.7907742998352554,
+      "grad_norm": 11486.779296875,
+      "learning_rate": 4.103194103194104e-05,
+      "loss": 0.591,
+      "step": 120
+    },
+    {
+      "epoch": 0.7973640856672158,
+      "grad_norm": 12171.77734375,
+      "learning_rate": 4.0909090909090915e-05,
+      "loss": 0.6598,
+      "step": 121
+    },
+    {
+      "epoch": 0.8039538714991763,
+      "grad_norm": 22686.841796875,
+      "learning_rate": 4.0786240786240785e-05,
+      "loss": 0.6059,
+      "step": 122
+    },
+    {
+      "epoch": 0.8105436573311368,
+      "grad_norm": 27070.658203125,
+      "learning_rate": 4.066339066339067e-05,
+      "loss": 0.5981,
+      "step": 123
+    },
+    {
+      "epoch": 0.8171334431630972,
+      "grad_norm": 15005.7265625,
+      "learning_rate": 4.0540540540540545e-05,
+      "loss": 0.6305,
+      "step": 124
+    },
+    {
+      "epoch": 0.8237232289950577,
+      "grad_norm": 11356.5380859375,
+      "learning_rate": 4.0417690417690415e-05,
+      "loss": 0.6736,
+      "step": 125
+    },
+    {
+      "epoch": 0.8303130148270181,
+      "grad_norm": 78888.9296875,
+      "learning_rate": 4.02948402948403e-05,
+      "loss": 0.6746,
+      "step": 126
+    },
+    {
+      "epoch": 0.8369028006589786,
+      "grad_norm": 8765.6591796875,
+      "learning_rate": 4.0171990171990176e-05,
+      "loss": 0.6088,
+      "step": 127
+    },
+    {
+      "epoch": 0.8434925864909391,
+      "grad_norm": 27692.36328125,
+      "learning_rate": 4.004914004914005e-05,
+      "loss": 0.7365,
+      "step": 128
+    },
+    {
+      "epoch": 0.8500823723228995,
+      "grad_norm": 27458.75,
+      "learning_rate": 3.992628992628993e-05,
+      "loss": 0.6885,
+      "step": 129
+    },
+    {
+      "epoch": 0.85667215815486,
+      "grad_norm": 32011.431640625,
+      "learning_rate": 3.9803439803439806e-05,
+      "loss": 0.6128,
+      "step": 130
+    },
+    {
+      "epoch": 0.8632619439868204,
+      "grad_norm": 8888.3623046875,
+      "learning_rate": 3.968058968058968e-05,
+      "loss": 0.8316,
+      "step": 131
+    },
+    {
+      "epoch": 0.8698517298187809,
+      "grad_norm": 31571.5,
+      "learning_rate": 3.955773955773956e-05,
+      "loss": 0.6086,
+      "step": 132
+    },
+    {
+      "epoch": 0.8764415156507414,
+      "grad_norm": 7211.0869140625,
+      "learning_rate": 3.943488943488944e-05,
+      "loss": 0.6343,
+      "step": 133
+    },
+    {
+      "epoch": 0.8830313014827018,
+      "grad_norm": 28377.615234375,
+      "learning_rate": 3.9312039312039314e-05,
+      "loss": 0.6234,
+      "step": 134
+    },
+    {
+      "epoch": 0.8896210873146623,
+      "grad_norm": 19913.7734375,
+      "learning_rate": 3.918918918918919e-05,
+      "loss": 0.7022,
+      "step": 135
+    },
+    {
+      "epoch": 0.8962108731466227,
+      "grad_norm": 38258.0546875,
+      "learning_rate": 3.906633906633907e-05,
+      "loss": 0.553,
+      "step": 136
+    },
+    {
+      "epoch": 0.9028006589785832,
+      "grad_norm": 18686.283203125,
+      "learning_rate": 3.8943488943488944e-05,
+      "loss": 0.6194,
+      "step": 137
+    },
+    {
+      "epoch": 0.9093904448105437,
+      "grad_norm": 25210.80078125,
+      "learning_rate": 3.882063882063882e-05,
+      "loss": 0.6567,
+      "step": 138
+    },
+    {
+      "epoch": 0.9159802306425041,
+      "grad_norm": 24367.556640625,
+      "learning_rate": 3.86977886977887e-05,
+      "loss": 0.686,
+      "step": 139
+    },
+    {
+      "epoch": 0.9225700164744646,
+      "grad_norm": 135211.953125,
+      "learning_rate": 3.857493857493858e-05,
+      "loss": 0.5756,
+      "step": 140
+    },
+    {
+      "epoch": 0.929159802306425,
+      "grad_norm": 18591.775390625,
+      "learning_rate": 3.845208845208845e-05,
+      "loss": 0.6784,
+      "step": 141
+    },
+    {
+      "epoch": 0.9357495881383855,
+      "grad_norm": 13147.021484375,
+      "learning_rate": 3.8329238329238335e-05,
+      "loss": 0.6022,
+      "step": 142
+    },
+    {
+      "epoch": 0.942339373970346,
+      "grad_norm": 19533.595703125,
+      "learning_rate": 3.820638820638821e-05,
+      "loss": 0.5408,
+      "step": 143
+    },
+    {
+      "epoch": 0.9489291598023064,
+      "grad_norm": 7753.4462890625,
+      "learning_rate": 3.808353808353808e-05,
+      "loss": 0.6628,
+      "step": 144
+    },
+    {
+      "epoch": 0.9555189456342669,
+      "grad_norm": 44759.55859375,
+      "learning_rate": 3.7960687960687965e-05,
+      "loss": 0.5822,
+      "step": 145
+    },
+    {
+      "epoch": 0.9621087314662273,
+      "grad_norm": 7180.6044921875,
+      "learning_rate": 3.783783783783784e-05,
+      "loss": 0.5473,
+      "step": 146
+    },
+    {
+      "epoch": 0.9686985172981878,
+      "grad_norm": 37214.40625,
+      "learning_rate": 3.771498771498771e-05,
+      "loss": 0.6619,
+      "step": 147
+    },
+    {
+      "epoch": 0.9752883031301482,
+      "grad_norm": 14762.1357421875,
+      "learning_rate": 3.7592137592137596e-05,
+      "loss": 0.6526,
+      "step": 148
+    },
+    {
+      "epoch": 0.9818780889621087,
+      "grad_norm": 67078.1875,
+      "learning_rate": 3.746928746928747e-05,
+      "loss": 0.72,
+      "step": 149
+    },
+    {
+      "epoch": 0.9884678747940692,
+      "grad_norm": 21195.07421875,
+      "learning_rate": 3.734643734643735e-05,
+      "loss": 0.6566,
+      "step": 150
+    },
+    {
+      "epoch": 0.9950576606260296,
+      "grad_norm": 22123.794921875,
+      "learning_rate": 3.7223587223587226e-05,
+      "loss": 0.647,
+      "step": 151
+    },
+    {
+      "epoch": 1.00164744645799,
+      "grad_norm": 9963.392578125,
+      "learning_rate": 3.71007371007371e-05,
+      "loss": 0.5923,
+      "step": 152
+    },
+    {
+      "epoch": 1.0082372322899507,
+      "grad_norm": 10004.744140625,
+      "learning_rate": 3.697788697788698e-05,
+      "loss": 0.5811,
+      "step": 153
+    },
+    {
+      "epoch": 1.014827018121911,
+      "grad_norm": 5818.58935546875,
+      "learning_rate": 3.685503685503686e-05,
+      "loss": 0.5744,
+      "step": 154
+    },
+    {
+      "epoch": 1.0214168039538716,
+      "grad_norm": 12603.640625,
+      "learning_rate": 3.6732186732186734e-05,
+      "loss": 0.6278,
+      "step": 155
+    },
+    {
+      "epoch": 1.028006589785832,
+      "grad_norm": 23099.1796875,
+      "learning_rate": 3.660933660933661e-05,
+      "loss": 0.6001,
+      "step": 156
+    },
+    {
+      "epoch": 1.0345963756177925,
+      "grad_norm": 19207.27734375,
+      "learning_rate": 3.648648648648649e-05,
+      "loss": 0.5701,
+      "step": 157
+    },
+    {
+      "epoch": 1.0411861614497528,
+      "grad_norm": 10040.673828125,
+      "learning_rate": 3.6363636363636364e-05,
+      "loss": 0.6095,
+      "step": 158
+    },
+    {
+      "epoch": 1.0477759472817134,
+      "grad_norm": 13896.298828125,
+      "learning_rate": 3.624078624078625e-05,
+      "loss": 0.5673,
+      "step": 159
+    },
+    {
+      "epoch": 1.0543657331136738,
+      "grad_norm": 10502.177734375,
+      "learning_rate": 3.611793611793612e-05,
+      "loss": 0.5662,
+      "step": 160
+    },
+    {
+      "epoch": 1.0609555189456343,
+      "grad_norm": 6510.84033203125,
+      "learning_rate": 3.5995085995085995e-05,
+      "loss": 0.5095,
+      "step": 161
+    },
+    {
+      "epoch": 1.0675453047775947,
+      "grad_norm": 18709.1640625,
+      "learning_rate": 3.587223587223588e-05,
+      "loss": 0.5437,
+      "step": 162
+    },
+    {
+      "epoch": 1.0741350906095553,
+      "grad_norm": 6607.08935546875,
+      "learning_rate": 3.574938574938575e-05,
+      "loss": 0.5686,
+      "step": 163
+    },
+    {
+      "epoch": 1.0807248764415156,
+      "grad_norm": 11358.892578125,
+      "learning_rate": 3.562653562653563e-05,
+      "loss": 0.585,
+      "step": 164
+    },
+    {
+      "epoch": 1.0873146622734762,
+      "grad_norm": 24775.66796875,
+      "learning_rate": 3.550368550368551e-05,
+      "loss": 0.5248,
+      "step": 165
+    },
+    {
+      "epoch": 1.0939044481054365,
+      "grad_norm": 13074.2919921875,
+      "learning_rate": 3.538083538083538e-05,
+      "loss": 0.5206,
+      "step": 166
+    },
+    {
+      "epoch": 1.100494233937397,
+      "grad_norm": 10716.3486328125,
+      "learning_rate": 3.525798525798526e-05,
+      "loss": 0.6356,
+      "step": 167
+    },
+    {
+      "epoch": 1.1070840197693574,
+      "grad_norm": 19648.953125,
+      "learning_rate": 3.513513513513514e-05,
+      "loss": 0.5096,
+      "step": 168
+    },
+    {
+      "epoch": 1.113673805601318,
+      "grad_norm": 10499.521484375,
+      "learning_rate": 3.501228501228501e-05,
+      "loss": 0.6298,
+      "step": 169
+    },
+    {
+      "epoch": 1.1202635914332784,
+      "grad_norm": 38415.08984375,
+      "learning_rate": 3.488943488943489e-05,
+      "loss": 0.6272,
+      "step": 170
+    },
+    {
+      "epoch": 1.126853377265239,
+      "grad_norm": 18025.1328125,
+      "learning_rate": 3.476658476658477e-05,
+      "loss": 0.609,
+      "step": 171
+    },
+    {
+      "epoch": 1.1334431630971993,
+      "grad_norm": 17254.71484375,
+      "learning_rate": 3.4643734643734647e-05,
+      "loss": 0.5771,
+      "step": 172
+    },
+    {
+      "epoch": 1.1400329489291599,
+      "grad_norm": 56900.84375,
+      "learning_rate": 3.452088452088452e-05,
+      "loss": 0.4617,
+      "step": 173
+    },
+    {
+      "epoch": 1.1466227347611202,
+      "grad_norm": 94377.71875,
+      "learning_rate": 3.43980343980344e-05,
+      "loss": 0.5262,
+      "step": 174
+    },
+    {
+      "epoch": 1.1532125205930808,
+      "grad_norm": 26418.802734375,
+      "learning_rate": 3.427518427518428e-05,
+      "loss": 0.5388,
+      "step": 175
+    },
+    {
+      "epoch": 1.1598023064250411,
+      "grad_norm": 39558.36328125,
+      "learning_rate": 3.4152334152334154e-05,
+      "loss": 0.6323,
+      "step": 176
+    },
+    {
+      "epoch": 1.1663920922570017,
+      "grad_norm": 12363.81640625,
+      "learning_rate": 3.402948402948403e-05,
+      "loss": 0.4672,
+      "step": 177
+    },
+    {
+      "epoch": 1.172981878088962,
+      "grad_norm": 4902.41748046875,
+      "learning_rate": 3.390663390663391e-05,
+      "loss": 0.6218,
+      "step": 178
+    },
+    {
+      "epoch": 1.1795716639209226,
+      "grad_norm": 18641.564453125,
+      "learning_rate": 3.3783783783783784e-05,
+      "loss": 0.5324,
+      "step": 179
+    },
+    {
+      "epoch": 1.186161449752883,
+      "grad_norm": 20244.51171875,
+      "learning_rate": 3.366093366093366e-05,
+      "loss": 0.5641,
+      "step": 180
+    },
+    {
+      "epoch": 1.1927512355848435,
+      "grad_norm": 10752.59765625,
+      "learning_rate": 3.3538083538083545e-05,
+      "loss": 0.5537,
+      "step": 181
+    },
+    {
+      "epoch": 1.1993410214168039,
+      "grad_norm": 16433.798828125,
+      "learning_rate": 3.3415233415233415e-05,
+      "loss": 0.6038,
+      "step": 182
+    },
+    {
+      "epoch": 1.2059308072487644,
+      "grad_norm": 65011.72265625,
+      "learning_rate": 3.329238329238329e-05,
+      "loss": 0.71,
+      "step": 183
+    },
+    {
+      "epoch": 1.2125205930807248,
+      "grad_norm": 139034.671875,
+      "learning_rate": 3.3169533169533175e-05,
+      "loss": 0.5752,
+      "step": 184
+    },
+    {
+      "epoch": 1.2191103789126854,
+      "grad_norm": 6198.3046875,
+      "learning_rate": 3.3046683046683045e-05,
+      "loss": 0.6047,
+      "step": 185
+    },
+    {
+      "epoch": 1.2257001647446457,
+      "grad_norm": 32208.603515625,
+      "learning_rate": 3.292383292383293e-05,
+      "loss": 0.548,
+      "step": 186
+    },
+    {
+      "epoch": 1.2322899505766063,
+      "grad_norm": 83243.2578125,
+      "learning_rate": 3.2800982800982806e-05,
+      "loss": 0.6223,
+      "step": 187
+    },
+    {
+      "epoch": 1.2388797364085666,
+      "grad_norm": 11003.818359375,
+      "learning_rate": 3.2678132678132676e-05,
+      "loss": 0.5597,
+      "step": 188
+    },
+    {
+      "epoch": 1.2454695222405272,
+      "grad_norm": 30586.80078125,
+      "learning_rate": 3.255528255528256e-05,
+      "loss": 0.5653,
+      "step": 189
+    },
+    {
+      "epoch": 1.2520593080724876,
+      "grad_norm": 36942.6171875,
+      "learning_rate": 3.2432432432432436e-05,
+      "loss": 0.4991,
+      "step": 190
+    },
+    {
+      "epoch": 1.2586490939044481,
+      "grad_norm": 34296.30859375,
+      "learning_rate": 3.2309582309582306e-05,
+      "loss": 0.5868,
+      "step": 191
+    },
+    {
+      "epoch": 1.2652388797364087,
+      "grad_norm": 10427.1875,
+      "learning_rate": 3.218673218673219e-05,
+      "loss": 0.6163,
+      "step": 192
+    },
+    {
+      "epoch": 1.271828665568369,
+      "grad_norm": 50948.59375,
+      "learning_rate": 3.206388206388207e-05,
+      "loss": 0.5254,
+      "step": 193
+    },
+    {
+      "epoch": 1.2784184514003294,
+      "grad_norm": 5948.72802734375,
+      "learning_rate": 3.1941031941031943e-05,
+      "loss": 0.5176,
+      "step": 194
+    },
+    {
+      "epoch": 1.28500823723229,
+      "grad_norm": 24703.78515625,
+      "learning_rate": 3.181818181818182e-05,
+      "loss": 0.5382,
+      "step": 195
+    },
+    {
+      "epoch": 1.2915980230642505,
+      "grad_norm": 31584.517578125,
+      "learning_rate": 3.16953316953317e-05,
+      "loss": 0.5676,
+      "step": 196
+    },
+    {
+      "epoch": 1.2981878088962109,
+      "grad_norm": 15364.802734375,
+      "learning_rate": 3.1572481572481574e-05,
+      "loss": 0.5168,
+      "step": 197
+    },
+    {
+      "epoch": 1.3047775947281712,
+      "grad_norm": 20843.013671875,
+      "learning_rate": 3.144963144963145e-05,
+      "loss": 0.5424,
+      "step": 198
+    },
+    {
+      "epoch": 1.3113673805601318,
+      "grad_norm": 17939.484375,
+      "learning_rate": 3.132678132678133e-05,
+      "loss": 0.5653,
+      "step": 199
+    },
+    {
+      "epoch": 1.3179571663920924,
+      "grad_norm": 9113.7607421875,
+      "learning_rate": 3.120393120393121e-05,
+      "loss": 0.5648,
+      "step": 200
+    },
+    {
+      "epoch": 1.3245469522240527,
+      "grad_norm": 32489.458984375,
+      "learning_rate": 3.108108108108108e-05,
+      "loss": 0.5858,
+      "step": 201
+    },
+    {
+      "epoch": 1.331136738056013,
+      "grad_norm": 23361.166015625,
+      "learning_rate": 3.095823095823096e-05,
+      "loss": 0.5749,
+      "step": 202
+    },
+    {
+      "epoch": 1.3377265238879736,
+      "grad_norm": 12417.0478515625,
+      "learning_rate": 3.083538083538084e-05,
+      "loss": 0.5019,
+      "step": 203
+    },
+    {
+      "epoch": 1.3443163097199342,
+      "grad_norm": 24939.2265625,
+      "learning_rate": 3.071253071253071e-05,
+      "loss": 0.5861,
+      "step": 204
+    },
+    {
+      "epoch": 1.3509060955518946,
+      "grad_norm": 20826.5859375,
+      "learning_rate": 3.058968058968059e-05,
+      "loss": 0.591,
+      "step": 205
+    },
+    {
+      "epoch": 1.357495881383855,
+      "grad_norm": 13685.1572265625,
+      "learning_rate": 3.046683046683047e-05,
+      "loss": 0.6588,
+      "step": 206
+    },
+    {
+      "epoch": 1.3640856672158155,
+      "grad_norm": 26390.5625,
+      "learning_rate": 3.0343980343980342e-05,
+      "loss": 0.5532,
+      "step": 207
+    },
+    {
+      "epoch": 1.370675453047776,
+      "grad_norm": 13378.560546875,
+      "learning_rate": 3.0221130221130222e-05,
+      "loss": 0.5108,
+      "step": 208
+    },
+    {
+      "epoch": 1.3772652388797364,
+      "grad_norm": 8095.1689453125,
+      "learning_rate": 3.0098280098280103e-05,
+      "loss": 0.5522,
+      "step": 209
+    },
+    {
+      "epoch": 1.3838550247116967,
+      "grad_norm": 11471.375,
+      "learning_rate": 2.9975429975429976e-05,
+      "loss": 0.4777,
+      "step": 210
+    },
+    {
+      "epoch": 1.3904448105436573,
+      "grad_norm": 11452.013671875,
+      "learning_rate": 2.9852579852579853e-05,
+      "loss": 0.5945,
+      "step": 211
+    },
+    {
+      "epoch": 1.3970345963756179,
+      "grad_norm": 31670.03125,
+      "learning_rate": 2.9729729729729733e-05,
+      "loss": 0.6015,
+      "step": 212
+    },
+    {
+      "epoch": 1.4036243822075782,
+      "grad_norm": 22501.869140625,
+      "learning_rate": 2.9606879606879607e-05,
+      "loss": 0.4813,
+      "step": 213
+    },
+    {
+      "epoch": 1.4102141680395386,
+      "grad_norm": 102328.46875,
+      "learning_rate": 2.9484029484029483e-05,
+      "loss": 0.6696,
+      "step": 214
+    },
+    {
+      "epoch": 1.4168039538714992,
+      "grad_norm": 26580.23828125,
+      "learning_rate": 2.9361179361179364e-05,
+      "loss": 0.4692,
+      "step": 215
+    },
+    {
+      "epoch": 1.4233937397034597,
+      "grad_norm": 12255.455078125,
+      "learning_rate": 2.9238329238329237e-05,
+      "loss": 0.5184,
+      "step": 216
+    },
+    {
+      "epoch": 1.42998352553542,
+      "grad_norm": 13490.662109375,
+      "learning_rate": 2.9115479115479117e-05,
+      "loss": 0.5556,
+      "step": 217
+    },
+    {
+      "epoch": 1.4365733113673804,
+      "grad_norm": 14122.8046875,
+      "learning_rate": 2.8992628992628994e-05,
+      "loss": 0.5259,
+      "step": 218
+    },
+    {
+      "epoch": 1.443163097199341,
+      "grad_norm": 14578.7783203125,
+      "learning_rate": 2.8869778869778868e-05,
+      "loss": 0.5878,
+      "step": 219
+    },
+    {
+      "epoch": 1.4497528830313016,
+      "grad_norm": 36417.54296875,
+      "learning_rate": 2.8746928746928748e-05,
+      "loss": 0.6328,
+      "step": 220
+    },
+    {
+      "epoch": 1.456342668863262,
+      "grad_norm": 24717.400390625,
+      "learning_rate": 2.8624078624078625e-05,
+      "loss": 0.5597,
+      "step": 221
+    },
+    {
+      "epoch": 1.4629324546952225,
+      "grad_norm": 42796.0390625,
+      "learning_rate": 2.8501228501228505e-05,
+      "loss": 0.6242,
+      "step": 222
+    },
+    {
+      "epoch": 1.4695222405271828,
+      "grad_norm": 11775.6015625,
+      "learning_rate": 2.8378378378378378e-05,
+      "loss": 0.5561,
+      "step": 223
+    },
+    {
+      "epoch": 1.4761120263591434,
+      "grad_norm": 11294.82421875,
+      "learning_rate": 2.825552825552826e-05,
+      "loss": 0.6068,
+      "step": 224
+    },
+    {
+      "epoch": 1.4827018121911038,
+      "grad_norm": 164647.03125,
+      "learning_rate": 2.8132678132678135e-05,
+      "loss": 0.5618,
+      "step": 225
+    },
+    {
+      "epoch": 1.4892915980230643,
+      "grad_norm": 19815.3203125,
+      "learning_rate": 2.800982800982801e-05,
+      "loss": 0.5637,
+      "step": 226
+    },
+    {
+      "epoch": 1.4958813838550247,
+      "grad_norm": 87994.859375,
+      "learning_rate": 2.788697788697789e-05,
+      "loss": 0.6073,
+      "step": 227
+    },
+    {
+      "epoch": 1.5024711696869852,
+      "grad_norm": 9418.3603515625,
+      "learning_rate": 2.776412776412777e-05,
+      "loss": 0.5177,
+      "step": 228
+    },
+    {
+      "epoch": 1.5090609555189456,
+      "grad_norm": 293888.625,
+      "learning_rate": 2.764127764127764e-05,
+      "loss": 0.6276,
+      "step": 229
+    },
+    {
+      "epoch": 1.515650741350906,
+      "grad_norm": 10872.537109375,
+      "learning_rate": 2.751842751842752e-05,
+      "loss": 0.5746,
+      "step": 230
+    },
+    {
+      "epoch": 1.5222405271828665,
+      "grad_norm": 60597.2109375,
+      "learning_rate": 2.73955773955774e-05,
+      "loss": 0.5874,
+      "step": 231
+    },
+    {
+      "epoch": 1.528830313014827,
+      "grad_norm": 53855.8515625,
+      "learning_rate": 2.7272727272727273e-05,
+      "loss": 0.5431,
+      "step": 232
+    },
+    {
+      "epoch": 1.5354200988467874,
+      "grad_norm": 17162.5859375,
+      "learning_rate": 2.714987714987715e-05,
+      "loss": 0.5703,
+      "step": 233
+    },
+    {
+      "epoch": 1.5420098846787478,
+      "grad_norm": 12093.095703125,
+      "learning_rate": 2.702702702702703e-05,
+      "loss": 0.5776,
+      "step": 234
+    },
+    {
+      "epoch": 1.5485996705107083,
+      "grad_norm": 33037.2421875,
+      "learning_rate": 2.6904176904176904e-05,
+      "loss": 0.5018,
+      "step": 235
+    },
+    {
+      "epoch": 1.555189456342669,
+      "grad_norm": 18464.041015625,
+      "learning_rate": 2.678132678132678e-05,
+      "loss": 0.5508,
+      "step": 236
+    },
+    {
+      "epoch": 1.5617792421746293,
+      "grad_norm": 33591.078125,
+      "learning_rate": 2.665847665847666e-05,
+      "loss": 0.5522,
+      "step": 237
+    },
+    {
+      "epoch": 1.5683690280065898,
+      "grad_norm": 13077.189453125,
+      "learning_rate": 2.6535626535626534e-05,
+      "loss": 0.4632,
+      "step": 238
+    },
+    {
+      "epoch": 1.5749588138385504,
+      "grad_norm": 8347.5576171875,
+      "learning_rate": 2.6412776412776414e-05,
+      "loss": 0.6278,
+      "step": 239
+    },
+    {
+      "epoch": 1.5815485996705108,
+      "grad_norm": 15981.6953125,
+      "learning_rate": 2.628992628992629e-05,
+      "loss": 0.496,
+      "step": 240
+    },
+    {
+      "epoch": 1.588138385502471,
+      "grad_norm": 8308.947265625,
+      "learning_rate": 2.616707616707617e-05,
+      "loss": 0.659,
+      "step": 241
+    },
+    {
+      "epoch": 1.5947281713344317,
+      "grad_norm": 16152.51171875,
+      "learning_rate": 2.6044226044226045e-05,
+      "loss": 0.5447,
+      "step": 242
+    },
+    {
+      "epoch": 1.6013179571663922,
+      "grad_norm": 10750.263671875,
+      "learning_rate": 2.5921375921375925e-05,
+      "loss": 0.6014,
+      "step": 243
+    },
+    {
+      "epoch": 1.6079077429983526,
+      "grad_norm": 16181.7021484375,
+      "learning_rate": 2.5798525798525802e-05,
+      "loss": 0.5084,
+      "step": 244
+    },
+    {
+      "epoch": 1.614497528830313,
+      "grad_norm": 17315.806640625,
+      "learning_rate": 2.5675675675675675e-05,
+      "loss": 0.6028,
+      "step": 245
+    },
+    {
+      "epoch": 1.6210873146622735,
+      "grad_norm": 35364.375,
+      "learning_rate": 2.5552825552825555e-05,
+      "loss": 0.565,
+      "step": 246
+    },
+    {
+      "epoch": 1.627677100494234,
+      "grad_norm": 7373.3349609375,
+      "learning_rate": 2.5429975429975432e-05,
+      "loss": 0.5849,
+      "step": 247
+    },
+    {
+      "epoch": 1.6342668863261944,
+      "grad_norm": 10823.8154296875,
+      "learning_rate": 2.5307125307125306e-05,
+      "loss": 0.4575,
+      "step": 248
+    },
+    {
+      "epoch": 1.6408566721581548,
+      "grad_norm": 15178.8916015625,
+      "learning_rate": 2.5184275184275186e-05,
+      "loss": 0.4679,
+      "step": 249
+    },
+    {
+      "epoch": 1.6474464579901154,
+      "grad_norm": 10164.9697265625,
+      "learning_rate": 2.5061425061425066e-05,
+      "loss": 0.4315,
+      "step": 250
+    },
+    {
+      "epoch": 1.654036243822076,
+      "grad_norm": 8609.7099609375,
+      "learning_rate": 2.493857493857494e-05,
+      "loss": 0.6248,
+      "step": 251
+    },
+    {
+      "epoch": 1.6606260296540363,
+      "grad_norm": 40045.09375,
+      "learning_rate": 2.4815724815724816e-05,
+      "loss": 0.5335,
+      "step": 252
+    },
+    {
+      "epoch": 1.6672158154859966,
+      "grad_norm": 17241.337890625,
+      "learning_rate": 2.4692874692874693e-05,
+      "loss": 0.6546,
+      "step": 253
+    },
+    {
+      "epoch": 1.6738056013179572,
+      "grad_norm": 23707.69921875,
+      "learning_rate": 2.4570024570024573e-05,
+      "loss": 0.5672,
+      "step": 254
+    },
+    {
+      "epoch": 1.6803953871499178,
+      "grad_norm": 9953.140625,
+      "learning_rate": 2.4447174447174447e-05,
+      "loss": 0.5867,
+      "step": 255
+    },
+    {
+      "epoch": 1.6869851729818781,
+      "grad_norm": 29007.3125,
+      "learning_rate": 2.4324324324324327e-05,
+      "loss": 0.5345,
+      "step": 256
+    },
+    {
+      "epoch": 1.6935749588138385,
+      "grad_norm": 9451.4765625,
+      "learning_rate": 2.4201474201474204e-05,
+      "loss": 0.6363,
+      "step": 257
+    },
+    {
+      "epoch": 1.700164744645799,
+      "grad_norm": 36570.16796875,
+      "learning_rate": 2.4078624078624077e-05,
+      "loss": 0.5335,
+      "step": 258
+    },
+    {
+      "epoch": 1.7067545304777596,
+      "grad_norm": 48013.99609375,
+      "learning_rate": 2.3955773955773958e-05,
+      "loss": 0.5234,
+      "step": 259
+    },
+    {
+      "epoch": 1.71334431630972,
+      "grad_norm": 43459.08203125,
+      "learning_rate": 2.3832923832923834e-05,
+      "loss": 0.5699,
+      "step": 260
+    },
+    {
+      "epoch": 1.7199341021416803,
+      "grad_norm": 27575.84375,
+      "learning_rate": 2.371007371007371e-05,
+      "loss": 0.5577,
+      "step": 261
+    },
+    {
+      "epoch": 1.7265238879736409,
+      "grad_norm": 6892.0205078125,
+      "learning_rate": 2.3587223587223588e-05,
+      "loss": 0.6098,
+      "step": 262
+    },
+    {
+      "epoch": 1.7331136738056014,
+      "grad_norm": 79846.03125,
+      "learning_rate": 2.3464373464373465e-05,
+      "loss": 0.5509,
+      "step": 263
+    },
+    {
+      "epoch": 1.7397034596375618,
+      "grad_norm": 13488.93359375,
+      "learning_rate": 2.3341523341523342e-05,
+      "loss": 0.5872,
+      "step": 264
+    },
+    {
+      "epoch": 1.7462932454695221,
+      "grad_norm": 20397.814453125,
+      "learning_rate": 2.3218673218673222e-05,
+      "loss": 0.5657,
+      "step": 265
+    },
+    {
+      "epoch": 1.7528830313014827,
+      "grad_norm": 14773.505859375,
+      "learning_rate": 2.3095823095823095e-05,
+      "loss": 0.5319,
+      "step": 266
+    },
+    {
+      "epoch": 1.7594728171334433,
+      "grad_norm": 59875.83203125,
+      "learning_rate": 2.2972972972972976e-05,
+      "loss": 0.6762,
+      "step": 267
+    },
+    {
+      "epoch": 1.7660626029654036,
+      "grad_norm": 29569.642578125,
+      "learning_rate": 2.2850122850122852e-05,
+      "loss": 0.7493,
+      "step": 268
+    },
+    {
+      "epoch": 1.772652388797364,
+      "grad_norm": 96513.0625,
+      "learning_rate": 2.272727272727273e-05,
+      "loss": 0.5796,
+      "step": 269
+    },
+    {
+      "epoch": 1.7792421746293245,
+      "grad_norm": 33220.875,
+      "learning_rate": 2.2604422604422606e-05,
+      "loss": 0.5513,
+      "step": 270
+    },
+    {
+      "epoch": 1.7858319604612851,
+      "grad_norm": 9378.62890625,
+      "learning_rate": 2.2481572481572483e-05,
+      "loss": 0.5444,
+      "step": 271
+    },
+    {
+      "epoch": 1.7924217462932455,
+      "grad_norm": 8672.4189453125,
+      "learning_rate": 2.235872235872236e-05,
+      "loss": 0.5997,
+      "step": 272
+    },
+    {
+      "epoch": 1.7990115321252058,
+      "grad_norm": 75445.5390625,
+      "learning_rate": 2.2235872235872237e-05,
+      "loss": 0.5383,
+      "step": 273
+    },
+    {
+      "epoch": 1.8056013179571664,
+      "grad_norm": 12022.205078125,
+      "learning_rate": 2.2113022113022113e-05,
+      "loss": 0.6333,
+      "step": 274
+    },
+    {
+      "epoch": 1.812191103789127,
+      "grad_norm": 28632.75390625,
+      "learning_rate": 2.199017199017199e-05,
+      "loss": 0.5491,
+      "step": 275
+    },
+    {
+      "epoch": 1.8187808896210873,
+      "grad_norm": 24272.568359375,
+      "learning_rate": 2.186732186732187e-05,
+      "loss": 0.5039,
+      "step": 276
+    },
+    {
+      "epoch": 1.8253706754530477,
+      "grad_norm": 9478.4189453125,
+      "learning_rate": 2.1744471744471744e-05,
+      "loss": 0.4395,
+      "step": 277
+    },
+    {
+      "epoch": 1.8319604612850082,
+      "grad_norm": 10953.66796875,
+      "learning_rate": 2.1621621621621624e-05,
+      "loss": 0.6367,
+      "step": 278
+    },
+    {
+      "epoch": 1.8385502471169688,
+      "grad_norm": 18094.830078125,
+      "learning_rate": 2.14987714987715e-05,
+      "loss": 0.5477,
+      "step": 279
+    },
+    {
+      "epoch": 1.8451400329489291,
+      "grad_norm": 18395.52734375,
+      "learning_rate": 2.1375921375921378e-05,
+      "loss": 0.5301,
+      "step": 280
+    },
+    {
+      "epoch": 1.8517298187808895,
+      "grad_norm": 104393.421875,
+      "learning_rate": 2.1253071253071255e-05,
+      "loss": 0.581,
+      "step": 281
+    },
+    {
+      "epoch": 1.85831960461285,
+      "grad_norm": 10463.978515625,
+      "learning_rate": 2.113022113022113e-05,
+      "loss": 0.5688,
+      "step": 282
+    },
+    {
+      "epoch": 1.8649093904448106,
+      "grad_norm": 7907.69189453125,
+      "learning_rate": 2.1007371007371008e-05,
+      "loss": 0.5534,
+      "step": 283
+    },
+    {
+      "epoch": 1.871499176276771,
+      "grad_norm": 12477.5712890625,
+      "learning_rate": 2.0884520884520885e-05,
+      "loss": 0.5412,
+      "step": 284
+    },
+    {
+      "epoch": 1.8780889621087313,
+      "grad_norm": 22077.701171875,
+      "learning_rate": 2.0761670761670762e-05,
+      "loss": 0.6015,
+      "step": 285
+    },
+    {
+      "epoch": 1.884678747940692,
+      "grad_norm": 29665.4296875,
+      "learning_rate": 2.063882063882064e-05,
+      "loss": 0.6349,
+      "step": 286
+    },
+    {
+      "epoch": 1.8912685337726525,
+      "grad_norm": 8550.171875,
+      "learning_rate": 2.051597051597052e-05,
+      "loss": 0.5846,
+      "step": 287
+    },
+    {
+      "epoch": 1.8978583196046128,
+      "grad_norm": 34534.80859375,
+      "learning_rate": 2.0393120393120392e-05,
+      "loss": 0.6063,
+      "step": 288
+    },
+    {
+      "epoch": 1.9044481054365732,
+      "grad_norm": 25177.533203125,
+      "learning_rate": 2.0270270270270273e-05,
+      "loss": 0.5608,
+      "step": 289
+    },
+    {
+      "epoch": 1.9110378912685337,
+      "grad_norm": 12976.5283203125,
+      "learning_rate": 2.014742014742015e-05,
+      "loss": 0.6424,
+      "step": 290
+    },
+    {
+      "epoch": 1.9176276771004943,
+      "grad_norm": 36851.8359375,
+      "learning_rate": 2.0024570024570026e-05,
+      "loss": 0.6051,
+      "step": 291
+    },
+    {
+      "epoch": 1.9242174629324547,
+      "grad_norm": 14700.5849609375,
+      "learning_rate": 1.9901719901719903e-05,
+      "loss": 0.6738,
+      "step": 292
+    },
+    {
+      "epoch": 1.930807248764415,
+      "grad_norm": 10047.1728515625,
+      "learning_rate": 1.977886977886978e-05,
+      "loss": 0.5817,
+      "step": 293
+    },
+    {
+      "epoch": 1.9373970345963756,
+      "grad_norm": 18526.845703125,
+      "learning_rate": 1.9656019656019657e-05,
+      "loss": 0.5037,
+      "step": 294
+    },
+    {
+      "epoch": 1.9439868204283361,
+      "grad_norm": 51555.40234375,
+      "learning_rate": 1.9533169533169534e-05,
+      "loss": 0.7556,
+      "step": 295
+    },
+    {
+      "epoch": 1.9505766062602965,
+      "grad_norm": 16909.0625,
+      "learning_rate": 1.941031941031941e-05,
+      "loss": 0.5487,
+      "step": 296
+    },
+    {
+      "epoch": 1.9571663920922568,
+      "grad_norm": 22686.32421875,
+      "learning_rate": 1.928746928746929e-05,
+      "loss": 0.617,
+      "step": 297
+    },
+    {
+      "epoch": 1.9637561779242174,
+      "grad_norm": 68786.3359375,
+      "learning_rate": 1.9164619164619167e-05,
+      "loss": 0.5581,
+      "step": 298
+    },
+    {
+      "epoch": 1.970345963756178,
+      "grad_norm": 44203.34375,
+      "learning_rate": 1.904176904176904e-05,
+      "loss": 0.5491,
+      "step": 299
+    },
+    {
+      "epoch": 1.9769357495881383,
+      "grad_norm": 39032.02734375,
+      "learning_rate": 1.891891891891892e-05,
+      "loss": 0.5876,
+      "step": 300
+    },
+    {
+      "epoch": 1.9835255354200987,
+      "grad_norm": 15117.4921875,
+      "learning_rate": 1.8796068796068798e-05,
+      "loss": 0.6322,
+      "step": 301
+    },
+    {
+      "epoch": 1.9901153212520593,
+      "grad_norm": 17359.947265625,
+      "learning_rate": 1.8673218673218675e-05,
+      "loss": 0.5526,
+      "step": 302
+    },
+    {
+      "epoch": 1.9967051070840198,
+      "grad_norm": 26160.140625,
+      "learning_rate": 1.855036855036855e-05,
+      "loss": 0.4606,
+      "step": 303
+    },
+    {
+      "epoch": 2.00329489291598,
+      "grad_norm": 27996.744140625,
+      "learning_rate": 1.842751842751843e-05,
+      "loss": 0.5775,
+      "step": 304
+    },
+    {
+      "epoch": 2.0098846787479405,
+      "grad_norm": 10826.375,
+      "learning_rate": 1.8304668304668305e-05,
+      "loss": 0.4791,
+      "step": 305
+    },
+    {
+      "epoch": 2.0164744645799013,
+      "grad_norm": 7867.39697265625,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 0.6314,
+      "step": 306
+    },
+    {
+      "epoch": 2.0230642504118617,
+      "grad_norm": 13956.0439453125,
+      "learning_rate": 1.805896805896806e-05,
+      "loss": 0.4655,
+      "step": 307
+    },
+    {
+      "epoch": 2.029654036243822,
+      "grad_norm": 14681.5576171875,
+      "learning_rate": 1.793611793611794e-05,
+      "loss": 0.536,
+      "step": 308
+    },
+    {
+      "epoch": 2.0362438220757824,
+      "grad_norm": 12553.2109375,
+      "learning_rate": 1.7813267813267816e-05,
+      "loss": 0.5533,
+      "step": 309
+    },
+    {
+      "epoch": 2.042833607907743,
+      "grad_norm": 27893.552734375,
+      "learning_rate": 1.769041769041769e-05,
+      "loss": 0.4929,
+      "step": 310
+    },
+    {
+      "epoch": 2.0494233937397035,
+      "grad_norm": 16818.6171875,
+      "learning_rate": 1.756756756756757e-05,
+      "loss": 0.5123,
+      "step": 311
+    },
+    {
+      "epoch": 2.056013179571664,
+      "grad_norm": 16560.30859375,
+      "learning_rate": 1.7444717444717446e-05,
+      "loss": 0.5537,
+      "step": 312
+    },
+    {
+      "epoch": 2.062602965403624,
+      "grad_norm": 10146.822265625,
+      "learning_rate": 1.7321867321867323e-05,
+      "loss": 0.5042,
+      "step": 313
+    },
+    {
+      "epoch": 2.069192751235585,
+      "grad_norm": 15292.8125,
+      "learning_rate": 1.71990171990172e-05,
+      "loss": 0.4717,
+      "step": 314
+    },
+    {
+      "epoch": 2.0757825370675453,
+      "grad_norm": 89878.0078125,
+      "learning_rate": 1.7076167076167077e-05,
+      "loss": 0.5159,
+      "step": 315
+    },
+    {
+      "epoch": 2.0823723228995057,
+      "grad_norm": 23665.125,
+      "learning_rate": 1.6953316953316954e-05,
+      "loss": 0.4446,
+      "step": 316
+    },
+    {
+      "epoch": 2.088962108731466,
+      "grad_norm": 15712.8779296875,
+      "learning_rate": 1.683046683046683e-05,
+      "loss": 0.594,
+      "step": 317
+    },
+    {
+      "epoch": 2.095551894563427,
+      "grad_norm": 50007.99609375,
+      "learning_rate": 1.6707616707616707e-05,
+      "loss": 0.5259,
+      "step": 318
+    },
+    {
+      "epoch": 2.102141680395387,
+      "grad_norm": 30487.50390625,
+      "learning_rate": 1.6584766584766588e-05,
+      "loss": 0.5166,
+      "step": 319
+    },
+    {
+      "epoch": 2.1087314662273475,
+      "grad_norm": 29152.314453125,
+      "learning_rate": 1.6461916461916464e-05,
+      "loss": 0.5055,
+      "step": 320
+    },
+    {
+      "epoch": 2.115321252059308,
+      "grad_norm": 22360.861328125,
+      "learning_rate": 1.6339066339066338e-05,
+      "loss": 0.4797,
+      "step": 321
+    },
+    {
+      "epoch": 2.1219110378912687,
+      "grad_norm": 45466.05078125,
+      "learning_rate": 1.6216216216216218e-05,
+      "loss": 0.4505,
+      "step": 322
+    },
+    {
+      "epoch": 2.128500823723229,
+      "grad_norm": 18740.75390625,
+      "learning_rate": 1.6093366093366095e-05,
+      "loss": 0.4849,
+      "step": 323
+    },
+    {
+      "epoch": 2.1350906095551894,
+      "grad_norm": 9168.376953125,
+      "learning_rate": 1.5970515970515972e-05,
+      "loss": 0.4991,
+      "step": 324
+    },
+    {
+      "epoch": 2.1416803953871497,
+      "grad_norm": 7032.57373046875,
+      "learning_rate": 1.584766584766585e-05,
+      "loss": 0.5332,
+      "step": 325
+    },
+    {
+      "epoch": 2.1482701812191105,
+      "grad_norm": 14946.6484375,
+      "learning_rate": 1.5724815724815725e-05,
+      "loss": 0.4797,
+      "step": 326
+    },
+    {
+      "epoch": 2.154859967051071,
+      "grad_norm": 10197.732421875,
+      "learning_rate": 1.5601965601965606e-05,
+      "loss": 0.4698,
+      "step": 327
+    },
+    {
+      "epoch": 2.161449752883031,
+      "grad_norm": 11579.7890625,
+      "learning_rate": 1.547911547911548e-05,
+      "loss": 0.4995,
+      "step": 328
+    },
+    {
+      "epoch": 2.168039538714992,
+      "grad_norm": 26515.12890625,
+      "learning_rate": 1.5356265356265356e-05,
+      "loss": 0.4931,
+      "step": 329
+    },
+    {
+      "epoch": 2.1746293245469523,
+      "grad_norm": 16076.119140625,
+      "learning_rate": 1.5233415233415234e-05,
+      "loss": 0.5051,
+      "step": 330
+    },
+    {
+      "epoch": 2.1812191103789127,
+      "grad_norm": 12420.37890625,
+      "learning_rate": 1.5110565110565111e-05,
+      "loss": 0.5227,
+      "step": 331
+    },
+    {
+      "epoch": 2.187808896210873,
+      "grad_norm": 17015.359375,
+      "learning_rate": 1.4987714987714988e-05,
+      "loss": 0.5115,
+      "step": 332
+    },
+    {
+      "epoch": 2.1943986820428334,
+      "grad_norm": 13335.0068359375,
+      "learning_rate": 1.4864864864864867e-05,
+      "loss": 0.5835,
+      "step": 333
+    },
+    {
+      "epoch": 2.200988467874794,
+      "grad_norm": 41513.1015625,
+      "learning_rate": 1.4742014742014742e-05,
+      "loss": 0.437,
+      "step": 334
+    },
+    {
+      "epoch": 2.2075782537067545,
+      "grad_norm": 11096.04296875,
+      "learning_rate": 1.4619164619164619e-05,
+      "loss": 0.4619,
+      "step": 335
+    },
+    {
+      "epoch": 2.214168039538715,
+      "grad_norm": 22811.95703125,
+      "learning_rate": 1.4496314496314497e-05,
+      "loss": 0.521,
+      "step": 336
+    },
+    {
+      "epoch": 2.2207578253706757,
+      "grad_norm": 96701.7109375,
+      "learning_rate": 1.4373464373464374e-05,
+      "loss": 0.5015,
+      "step": 337
+    },
+    {
+      "epoch": 2.227347611202636,
+      "grad_norm": 28764.791015625,
+      "learning_rate": 1.4250614250614252e-05,
+      "loss": 0.5669,
+      "step": 338
+    },
+    {
+      "epoch": 2.2339373970345964,
+      "grad_norm": 9551.7666015625,
+      "learning_rate": 1.412776412776413e-05,
+      "loss": 0.5331,
+      "step": 339
+    },
+    {
+      "epoch": 2.2405271828665567,
+      "grad_norm": 24150.939453125,
+      "learning_rate": 1.4004914004914004e-05,
+      "loss": 0.6009,
+      "step": 340
+    },
+    {
+      "epoch": 2.247116968698517,
+      "grad_norm": 6333.8935546875,
+      "learning_rate": 1.3882063882063885e-05,
+      "loss": 0.4918,
+      "step": 341
+    },
+    {
+      "epoch": 2.253706754530478,
+      "grad_norm": 12934.755859375,
+      "learning_rate": 1.375921375921376e-05,
+      "loss": 0.4723,
+      "step": 342
+    },
+    {
+      "epoch": 2.260296540362438,
+      "grad_norm": 35915.16015625,
+      "learning_rate": 1.3636363636363637e-05,
+      "loss": 0.5468,
+      "step": 343
+    },
+    {
+      "epoch": 2.2668863261943986,
+      "grad_norm": 32075.701171875,
+      "learning_rate": 1.3513513513513515e-05,
+      "loss": 0.4188,
+      "step": 344
+    },
+    {
+      "epoch": 2.2734761120263594,
+      "grad_norm": 34175.6171875,
+      "learning_rate": 1.339066339066339e-05,
+      "loss": 0.5332,
+      "step": 345
+    },
+    {
+      "epoch": 2.2800658978583197,
+      "grad_norm": 27540.275390625,
+      "learning_rate": 1.3267813267813267e-05,
+      "loss": 0.4866,
+      "step": 346
+    },
+    {
+      "epoch": 2.28665568369028,
+      "grad_norm": 8475.587890625,
+      "learning_rate": 1.3144963144963146e-05,
+      "loss": 0.4882,
+      "step": 347
+    },
+    {
+      "epoch": 2.2932454695222404,
+      "grad_norm": 31548.888671875,
+      "learning_rate": 1.3022113022113022e-05,
+      "loss": 0.4984,
+      "step": 348
+    },
+    {
+      "epoch": 2.2998352553542007,
+      "grad_norm": 13258.5400390625,
+      "learning_rate": 1.2899262899262901e-05,
+      "loss": 0.4839,
+      "step": 349
+    },
+    {
+      "epoch": 2.3064250411861615,
+      "grad_norm": 38574.671875,
+      "learning_rate": 1.2776412776412778e-05,
+      "loss": 0.4749,
+      "step": 350
+    },
+    {
+      "epoch": 2.313014827018122,
+      "grad_norm": 11923.9541015625,
+      "learning_rate": 1.2653562653562653e-05,
+      "loss": 0.4684,
+      "step": 351
+    },
+    {
+      "epoch": 2.3196046128500822,
+      "grad_norm": 26359.25390625,
+      "learning_rate": 1.2530712530712533e-05,
+      "loss": 0.544,
+      "step": 352
+    },
+    {
+      "epoch": 2.326194398682043,
+      "grad_norm": 13086.6357421875,
+      "learning_rate": 1.2407862407862408e-05,
+      "loss": 0.4722,
+      "step": 353
+    },
+    {
+      "epoch": 2.3327841845140034,
+      "grad_norm": 15811.302734375,
+      "learning_rate": 1.2285012285012287e-05,
+      "loss": 0.4928,
+      "step": 354
+    },
+    {
+      "epoch": 2.3393739703459637,
+      "grad_norm": 6917.1416015625,
+      "learning_rate": 1.2162162162162164e-05,
+      "loss": 0.5691,
+      "step": 355
+    },
+    {
+      "epoch": 2.345963756177924,
+      "grad_norm": 41444.01171875,
+      "learning_rate": 1.2039312039312039e-05,
+      "loss": 0.5405,
+      "step": 356
+    },
+    {
+      "epoch": 2.352553542009885,
+      "grad_norm": 139438.609375,
+      "learning_rate": 1.1916461916461917e-05,
+      "loss": 0.5219,
+      "step": 357
+    },
+    {
+      "epoch": 2.359143327841845,
+      "grad_norm": 9378.14453125,
+      "learning_rate": 1.1793611793611794e-05,
+      "loss": 0.4549,
+      "step": 358
+    },
+    {
+      "epoch": 2.3657331136738056,
+      "grad_norm": 15612.345703125,
+      "learning_rate": 1.1670761670761671e-05,
+      "loss": 0.551,
+      "step": 359
+    },
+    {
+      "epoch": 2.372322899505766,
+      "grad_norm": 24843.34375,
+      "learning_rate": 1.1547911547911548e-05,
+      "loss": 0.4928,
+      "step": 360
+    },
+    {
+      "epoch": 2.3789126853377267,
+      "grad_norm": 27883.3359375,
+      "learning_rate": 1.1425061425061426e-05,
+      "loss": 0.5312,
+      "step": 361
+    },
+    {
+      "epoch": 2.385502471169687,
+      "grad_norm": 29997.705078125,
+      "learning_rate": 1.1302211302211303e-05,
+      "loss": 0.5811,
+      "step": 362
+    },
+    {
+      "epoch": 2.3920922570016474,
+      "grad_norm": 15760.669921875,
+      "learning_rate": 1.117936117936118e-05,
+      "loss": 0.5359,
+      "step": 363
+    },
+    {
+      "epoch": 2.3986820428336078,
+      "grad_norm": 75893.109375,
+      "learning_rate": 1.1056511056511057e-05,
+      "loss": 0.4696,
+      "step": 364
+    },
+    {
+      "epoch": 2.4052718286655685,
+      "grad_norm": 20869.072265625,
+      "learning_rate": 1.0933660933660935e-05,
+      "loss": 0.5342,
+      "step": 365
+    },
+    {
+      "epoch": 2.411861614497529,
+      "grad_norm": 15102.4658203125,
+      "learning_rate": 1.0810810810810812e-05,
+      "loss": 0.505,
+      "step": 366
+    },
+    {
+      "epoch": 2.4184514003294892,
+      "grad_norm": 24776.064453125,
+      "learning_rate": 1.0687960687960689e-05,
+      "loss": 0.5192,
+      "step": 367
+    },
+    {
+      "epoch": 2.4250411861614496,
+      "grad_norm": 17849.11328125,
+      "learning_rate": 1.0565110565110566e-05,
+      "loss": 0.5461,
+      "step": 368
+    },
+    {
+      "epoch": 2.4316309719934104,
+      "grad_norm": 55960.76953125,
+      "learning_rate": 1.0442260442260443e-05,
+      "loss": 0.5075,
+      "step": 369
+    },
+    {
+      "epoch": 2.4382207578253707,
+      "grad_norm": 19820.607421875,
+      "learning_rate": 1.031941031941032e-05,
+      "loss": 0.5661,
+      "step": 370
+    },
+    {
+      "epoch": 2.444810543657331,
+      "grad_norm": 14240.271484375,
+      "learning_rate": 1.0196560196560196e-05,
+      "loss": 0.508,
+      "step": 371
+    },
+    {
+      "epoch": 2.4514003294892914,
+      "grad_norm": 134865.75,
+      "learning_rate": 1.0073710073710075e-05,
+      "loss": 0.5145,
+      "step": 372
+    },
+    {
+      "epoch": 2.4579901153212522,
+      "grad_norm": 26314.51953125,
+      "learning_rate": 9.950859950859952e-06,
+      "loss": 0.5024,
+      "step": 373
+    },
+    {
+      "epoch": 2.4645799011532126,
+      "grad_norm": 8372.9169921875,
+      "learning_rate": 9.828009828009828e-06,
+      "loss": 0.6207,
+      "step": 374
+    },
+    {
+      "epoch": 2.471169686985173,
+      "grad_norm": 9625.1865234375,
+      "learning_rate": 9.705159705159705e-06,
+      "loss": 0.4774,
+      "step": 375
+    },
+    {
+      "epoch": 2.4777594728171333,
+      "grad_norm": 12379.4033203125,
+      "learning_rate": 9.582309582309584e-06,
+      "loss": 0.4996,
+      "step": 376
+    },
+    {
+      "epoch": 2.484349258649094,
+      "grad_norm": 7536.5390625,
+      "learning_rate": 9.45945945945946e-06,
+      "loss": 0.5236,
+      "step": 377
+    },
+    {
+      "epoch": 2.4909390444810544,
+      "grad_norm": 28774.794921875,
+      "learning_rate": 9.336609336609337e-06,
+      "loss": 0.4793,
+      "step": 378
+    },
+    {
+      "epoch": 2.4975288303130148,
+      "grad_norm": 26588.126953125,
+      "learning_rate": 9.213759213759214e-06,
+      "loss": 0.4341,
+      "step": 379
+    },
+    {
+      "epoch": 2.504118616144975,
+      "grad_norm": 20777.30078125,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.5086,
+      "step": 380
+    },
+    {
+      "epoch": 2.510708401976936,
+      "grad_norm": 11463.2158203125,
+      "learning_rate": 8.96805896805897e-06,
+      "loss": 0.5769,
+      "step": 381
+    },
+    {
+      "epoch": 2.5172981878088962,
+      "grad_norm": 39708.0390625,
+      "learning_rate": 8.845208845208845e-06,
+      "loss": 0.5116,
+      "step": 382
+    },
+    {
+      "epoch": 2.5238879736408566,
+      "grad_norm": 20928.3671875,
+      "learning_rate": 8.722358722358723e-06,
+      "loss": 0.4997,
+      "step": 383
+    },
+    {
+      "epoch": 2.5304777594728174,
+      "grad_norm": 16851.19140625,
+      "learning_rate": 8.5995085995086e-06,
+      "loss": 0.5821,
+      "step": 384
+    },
+    {
+      "epoch": 2.5370675453047777,
+      "grad_norm": 12078.3798828125,
+      "learning_rate": 8.476658476658477e-06,
+      "loss": 0.5221,
+      "step": 385
+    },
+    {
+      "epoch": 2.543657331136738,
+      "grad_norm": 66684.203125,
+      "learning_rate": 8.353808353808354e-06,
+      "loss": 0.5621,
+      "step": 386
+    },
+    {
+      "epoch": 2.5502471169686984,
+      "grad_norm": 5581.15234375,
+      "learning_rate": 8.230958230958232e-06,
+      "loss": 0.4489,
+      "step": 387
+    },
+    {
+      "epoch": 2.556836902800659,
+      "grad_norm": 23770.3359375,
+      "learning_rate": 8.108108108108109e-06,
+      "loss": 0.5072,
+      "step": 388
+    },
+    {
+      "epoch": 2.5634266886326196,
+      "grad_norm": 15222.48046875,
+      "learning_rate": 7.985257985257986e-06,
+      "loss": 0.4957,
+      "step": 389
+    },
+    {
+      "epoch": 2.57001647446458,
+      "grad_norm": 14970.568359375,
+      "learning_rate": 7.862407862407863e-06,
+      "loss": 0.4914,
+      "step": 390
+    },
+    {
+      "epoch": 2.5766062602965403,
+      "grad_norm": 6360.3046875,
+      "learning_rate": 7.73955773955774e-06,
+      "loss": 0.4492,
+      "step": 391
+    },
+    {
+      "epoch": 2.583196046128501,
+      "grad_norm": 15164.78515625,
+      "learning_rate": 7.616707616707617e-06,
+      "loss": 0.4749,
+      "step": 392
+    },
+    {
+      "epoch": 2.5897858319604614,
+      "grad_norm": 22485.28125,
+      "learning_rate": 7.493857493857494e-06,
+      "loss": 0.5383,
+      "step": 393
+    },
+    {
+      "epoch": 2.5963756177924218,
+      "grad_norm": 30391.03515625,
+      "learning_rate": 7.371007371007371e-06,
+      "loss": 0.4506,
+      "step": 394
+    },
+    {
+      "epoch": 2.602965403624382,
+      "grad_norm": 15713.5595703125,
+      "learning_rate": 7.2481572481572485e-06,
+      "loss": 0.5178,
+      "step": 395
+    },
+    {
+      "epoch": 2.6095551894563425,
+      "grad_norm": 24513.958984375,
+      "learning_rate": 7.125307125307126e-06,
+      "loss": 0.4911,
+      "step": 396
+    },
+    {
+      "epoch": 2.6161449752883033,
+      "grad_norm": 21254.240234375,
+      "learning_rate": 7.002457002457002e-06,
+      "loss": 0.4883,
+      "step": 397
+    },
+    {
+      "epoch": 2.6227347611202636,
+      "grad_norm": 25682.513671875,
+      "learning_rate": 6.87960687960688e-06,
+      "loss": 0.5677,
+      "step": 398
+    },
+    {
+      "epoch": 2.629324546952224,
+      "grad_norm": 10649.7880859375,
+      "learning_rate": 6.7567567567567575e-06,
+      "loss": 0.4163,
+      "step": 399
+    },
+    {
+      "epoch": 2.6359143327841847,
+      "grad_norm": 11895.4677734375,
+      "learning_rate": 6.6339066339066335e-06,
+      "loss": 0.4617,
+      "step": 400
+    },
+    {
+      "epoch": 2.642504118616145,
+      "grad_norm": 71847.7421875,
+      "learning_rate": 6.511056511056511e-06,
+      "loss": 0.4855,
+      "step": 401
+    },
+    {
+      "epoch": 2.6490939044481054,
+      "grad_norm": 12087.333984375,
+      "learning_rate": 6.388206388206389e-06,
+      "loss": 0.5488,
+      "step": 402
+    },
+    {
+      "epoch": 2.655683690280066,
+      "grad_norm": 22596.474609375,
+      "learning_rate": 6.2653562653562665e-06,
+      "loss": 0.4533,
+      "step": 403
+    },
+    {
+      "epoch": 2.662273476112026,
+      "grad_norm": 13353.603515625,
+      "learning_rate": 6.142506142506143e-06,
+      "loss": 0.4916,
+      "step": 404
+    },
+    {
+      "epoch": 2.668863261943987,
+      "grad_norm": 28952.79296875,
+      "learning_rate": 6.019656019656019e-06,
+      "loss": 0.4664,
+      "step": 405
+    },
+    {
+      "epoch": 2.6754530477759473,
+      "grad_norm": 18681.466796875,
+      "learning_rate": 5.896805896805897e-06,
+      "loss": 0.5339,
+      "step": 406
+    },
+    {
+      "epoch": 2.6820428336079076,
+      "grad_norm": 55370.97265625,
+      "learning_rate": 5.773955773955774e-06,
+      "loss": 0.5157,
+      "step": 407
+    },
+    {
+      "epoch": 2.6886326194398684,
+      "grad_norm": 18563.974609375,
+      "learning_rate": 5.6511056511056515e-06,
+      "loss": 0.4533,
+      "step": 408
+    },
+    {
+      "epoch": 2.6952224052718288,
+      "grad_norm": 17806.52734375,
+      "learning_rate": 5.528255528255528e-06,
+      "loss": 0.5011,
+      "step": 409
+    },
+    {
+      "epoch": 2.701812191103789,
+      "grad_norm": 192237.5625,
+      "learning_rate": 5.405405405405406e-06,
+      "loss": 0.4728,
+      "step": 410
+    },
+    {
+      "epoch": 2.7084019769357495,
+      "grad_norm": 7906.73388671875,
+      "learning_rate": 5.282555282555283e-06,
+      "loss": 0.504,
+      "step": 411
+    },
+    {
+      "epoch": 2.71499176276771,
+      "grad_norm": 22805.4296875,
+      "learning_rate": 5.15970515970516e-06,
+      "loss": 0.5509,
+      "step": 412
+    },
+    {
+      "epoch": 2.7215815485996706,
+      "grad_norm": 12200.3623046875,
+      "learning_rate": 5.036855036855037e-06,
+      "loss": 0.4984,
+      "step": 413
+    },
+    {
+      "epoch": 2.728171334431631,
+      "grad_norm": 14147.498046875,
+      "learning_rate": 4.914004914004914e-06,
+      "loss": 0.4063,
+      "step": 414
+    },
+    {
+      "epoch": 2.7347611202635913,
+      "grad_norm": 11302.486328125,
+      "learning_rate": 4.791154791154792e-06,
+      "loss": 0.5305,
+      "step": 415
+    },
+    {
+      "epoch": 2.741350906095552,
+      "grad_norm": 12493.9912109375,
+      "learning_rate": 4.668304668304669e-06,
+      "loss": 0.5437,
+      "step": 416
+    },
+    {
+      "epoch": 2.7479406919275124,
+      "grad_norm": 12336.576171875,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.61,
+      "step": 417
+    },
+    {
+      "epoch": 2.754530477759473,
+      "grad_norm": 131413.265625,
+      "learning_rate": 4.422604422604422e-06,
+      "loss": 0.4386,
+      "step": 418
+    },
+    {
+      "epoch": 2.761120263591433,
+      "grad_norm": 6280.30810546875,
+      "learning_rate": 4.2997542997543e-06,
+      "loss": 0.4827,
+      "step": 419
+    },
+    {
+      "epoch": 2.7677100494233935,
+      "grad_norm": 22697.05859375,
+      "learning_rate": 4.176904176904177e-06,
+      "loss": 0.5122,
+      "step": 420
+    },
+    {
+      "epoch": 2.7742998352553543,
+      "grad_norm": 24444.3203125,
+      "learning_rate": 4.0540540540540545e-06,
+      "loss": 0.483,
+      "step": 421
+    },
+    {
+      "epoch": 2.7808896210873146,
+      "grad_norm": 11774.201171875,
+      "learning_rate": 3.931203931203931e-06,
+      "loss": 0.5757,
+      "step": 422
+    },
+    {
+      "epoch": 2.787479406919275,
+      "grad_norm": 16244.38671875,
+      "learning_rate": 3.8083538083538086e-06,
+      "loss": 0.4689,
+      "step": 423
+    },
+    {
+      "epoch": 2.7940691927512358,
+      "grad_norm": 22801.099609375,
+      "learning_rate": 3.6855036855036854e-06,
+      "loss": 0.4832,
+      "step": 424
+    },
+    {
+      "epoch": 2.800658978583196,
+      "grad_norm": 12143.181640625,
+      "learning_rate": 3.562653562653563e-06,
+      "loss": 0.5305,
+      "step": 425
+    },
+    {
+      "epoch": 2.8072487644151565,
+      "grad_norm": 37724.21875,
+      "learning_rate": 3.43980343980344e-06,
+      "loss": 0.6104,
+      "step": 426
+    },
+    {
+      "epoch": 2.813838550247117,
+      "grad_norm": 15959.6953125,
+      "learning_rate": 3.3169533169533168e-06,
+      "loss": 0.4719,
+      "step": 427
+    },
+    {
+      "epoch": 2.820428336079077,
+      "grad_norm": 16792.8984375,
+      "learning_rate": 3.1941031941031944e-06,
+      "loss": 0.5533,
+      "step": 428
+    },
+    {
+      "epoch": 2.827018121911038,
+      "grad_norm": 11308.0615234375,
+      "learning_rate": 3.0712530712530717e-06,
+      "loss": 0.5628,
+      "step": 429
+    },
+    {
+      "epoch": 2.8336079077429983,
+      "grad_norm": 23545.369140625,
+      "learning_rate": 2.9484029484029485e-06,
+      "loss": 0.5278,
+      "step": 430
+    },
+    {
+      "epoch": 2.8401976935749587,
+      "grad_norm": 33122.05859375,
+      "learning_rate": 2.8255528255528258e-06,
+      "loss": 0.4413,
+      "step": 431
+    },
+    {
+      "epoch": 2.8467874794069195,
+      "grad_norm": 22283.62890625,
+      "learning_rate": 2.702702702702703e-06,
+      "loss": 0.5699,
+      "step": 432
+    },
+    {
+      "epoch": 2.85337726523888,
+      "grad_norm": 13194.3056640625,
+      "learning_rate": 2.57985257985258e-06,
+      "loss": 0.5302,
+      "step": 433
+    },
+    {
+      "epoch": 2.85996705107084,
+      "grad_norm": 20634.77734375,
+      "learning_rate": 2.457002457002457e-06,
+      "loss": 0.5518,
+      "step": 434
+    },
+    {
+      "epoch": 2.8665568369028005,
+      "grad_norm": 15800.359375,
+      "learning_rate": 2.3341523341523343e-06,
+      "loss": 0.5346,
+      "step": 435
+    },
+    {
+      "epoch": 2.873146622734761,
+      "grad_norm": 15633.765625,
+      "learning_rate": 2.211302211302211e-06,
+      "loss": 0.4884,
+      "step": 436
+    },
+    {
+      "epoch": 2.8797364085667216,
+      "grad_norm": 31339.26953125,
+      "learning_rate": 2.0884520884520884e-06,
+      "loss": 0.4925,
+      "step": 437
+    },
+    {
+      "epoch": 2.886326194398682,
+      "grad_norm": 43050.4765625,
+      "learning_rate": 1.9656019656019657e-06,
+      "loss": 0.5312,
+      "step": 438
+    },
+    {
+      "epoch": 2.892915980230643,
+      "grad_norm": 17257.248046875,
+      "learning_rate": 1.8427518427518427e-06,
+      "loss": 0.4499,
+      "step": 439
+    },
+    {
+      "epoch": 2.899505766062603,
+      "grad_norm": 53676.23046875,
+      "learning_rate": 1.71990171990172e-06,
+      "loss": 0.4375,
+      "step": 440
+    },
+    {
+      "epoch": 2.9060955518945635,
+      "grad_norm": 31634.296875,
+      "learning_rate": 1.5970515970515972e-06,
+      "loss": 0.4394,
+      "step": 441
+    },
+    {
+      "epoch": 2.912685337726524,
+      "grad_norm": 22569.1171875,
+      "learning_rate": 1.4742014742014743e-06,
+      "loss": 0.4388,
+      "step": 442
+    },
+    {
+      "epoch": 2.919275123558484,
+      "grad_norm": 66667.6796875,
+      "learning_rate": 1.3513513513513515e-06,
+      "loss": 0.4113,
+      "step": 443
+    },
+    {
+      "epoch": 2.925864909390445,
+      "grad_norm": 7823.2880859375,
+      "learning_rate": 1.2285012285012285e-06,
+      "loss": 0.4316,
+      "step": 444
+    },
+    {
+      "epoch": 2.9324546952224053,
+      "grad_norm": 8603.0859375,
+      "learning_rate": 1.1056511056511056e-06,
+      "loss": 0.4213,
+      "step": 445
+    },
+    {
+      "epoch": 2.9390444810543657,
+      "grad_norm": 26335.857421875,
+      "learning_rate": 9.828009828009828e-07,
+      "loss": 0.4523,
+      "step": 446
+    },
+    {
+      "epoch": 2.9456342668863265,
+      "grad_norm": 18206.806640625,
+      "learning_rate": 8.5995085995086e-07,
+      "loss": 0.4656,
+      "step": 447
+    },
+    {
+      "epoch": 2.952224052718287,
+      "grad_norm": 9823.25,
+      "learning_rate": 7.371007371007371e-07,
+      "loss": 0.5816,
+      "step": 448
+    },
+    {
+      "epoch": 2.958813838550247,
+      "grad_norm": 15623.380859375,
+      "learning_rate": 6.142506142506143e-07,
+      "loss": 0.4464,
+      "step": 449
+    },
+    {
+      "epoch": 2.9654036243822075,
+      "grad_norm": 22244.39453125,
+      "learning_rate": 4.914004914004914e-07,
+      "loss": 0.5091,
+      "step": 450
+    },
+    {
+      "epoch": 2.971993410214168,
+      "grad_norm": 7978.923828125,
+      "learning_rate": 3.6855036855036856e-07,
+      "loss": 0.554,
+      "step": 451
+    },
+    {
+      "epoch": 2.9785831960461286,
+      "grad_norm": 20999.830078125,
+      "learning_rate": 2.457002457002457e-07,
+      "loss": 0.5075,
+      "step": 452
+    },
+    {
+      "epoch": 2.985172981878089,
+      "grad_norm": 22284.568359375,
+      "learning_rate": 1.2285012285012285e-07,
+      "loss": 0.4835,
+      "step": 453
+    },
+    {
+      "epoch": 2.985172981878089,
+      "step": 453,
+      "total_flos": 9.99505283128492e+17,
+      "train_loss": 0.5814478738155323,
+      "train_runtime": 61267.4818,
+      "train_samples_per_second": 0.357,
+      "train_steps_per_second": 0.007
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 453,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.99505283128492e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1634dfcf83cf0563995cda7b7f393bceca79544c47d4c2bb8bc37e8acffa028
+size 6840