Upload model

README.md

@@ -0,0 +1,61 @@
+---
+library_name: transformers
+license: other
+base_model: Qwen/Qwen2.5-32B-Instruct
+tags:
+- llama-factory
+- full
+- generated_from_trainer
+model-index:
+- name: no_pipeline_10k_32B
+  results: []
+---
+
+<!-- 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. -->
+
+# no_pipeline_10k_32B
+
+This model is a fine-tuned version of [Qwen/Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct) on the mlfoundations-dev/no_pipeline_10k dataset.
+
+## 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: 4e-05
+- train_batch_size: 1
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 64
+- gradient_accumulation_steps: 2
+- total_train_batch_size: 128
+- total_eval_batch_size: 512
+- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_ratio: 0.1
+- num_epochs: 5.0
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 4.46.1
+- Pytorch 2.6.0+cu124
+- Datasets 3.1.0
+- Tokenizers 0.20.3

all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 4.968152866242038,
+    "total_flos": 4.6317288847048704e+17,
+    "train_loss": 0.30433785139750213,
+    "train_runtime": 26201.4476,
+    "train_samples_per_second": 1.908,
+    "train_steps_per_second": 0.015
+}

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "Qwen/Qwen2.5-32B-Instruct",
+  "_name_or_path": "/leonardo_work/EUHPC_E03_068/DCFT_shared/hub/models--Qwen--Qwen2.5-32B-Instruct/snapshots/5ede1c97bbab6ce5cda5812749b4c0bdf79b18dd",
   "architectures": [
     "Qwen2ForCausalLM"
   ],

generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.46.1"
+}

model-00001-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4741f7e789cbe835ca5afd15db6e5b962c55615daf6b4c9ebcb3a9179593c3ca
+oid sha256:5f1f636f4e5f229264ac499fd1b6498b4d634e0eca7417ad74ac4370ba27b60a
 size 4891730992

model-00002-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:636c0d6d1c1ca0e93e048dcf9dae2a0bc210be51a1a22a7999903ca254d69ff2
+oid sha256:3c8221dca74000d315fc1656d1c4366e245ba02391a02a1eba22e33ba0b2a8de
 size 4876059352

model-00003-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:771abf431c41751f47a76fffe3c2aaaba1984f98ef6a9d8eacc0eb68d40ad1db
+oid sha256:201914ba32fcae76ad2cf974c3484207f19fe190408a0164379eebb8b86e1695
 size 4876059384

model-00004-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84c13970f1431cfbd9ce6b14e1ce4e12e40eeb71777c3c26a61963e1df2876d4
+oid sha256:82c112a3a8451b88bf953f38513836d1adc0e1db14a2831475ac973d4fa641a9
 size 4876059416

model-00005-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fad9b29536808b8ff32ec5cd3bb06c2eb3cf80d12380dd358e5c251cf76099e1
+oid sha256:3129e4cc601cb61a0814c6830fe510c2c9d300906baece2a426c43a1bbe113eb
 size 4876059416

model-00006-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d43232873b47fb3a984f4b6bc75a3894edc039326cb8f0009dbd64d086192469
+oid sha256:e7e40333a2d4843e5b1f3623940d74f0e8252facca31ca43abd8dae0efd8ffbb
 size 4876059416

model-00007-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:836ac301a2e83ec6c108aaece9e4cc9526c24a6d7babf9e28b19d12538a55c5e
+oid sha256:404f102e5ec35134edaaa203203bb81cff320b58693184f957347d7c678e22ce
 size 4876059416

model-00008-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e66b9ced4e2db6b7f64d8809d46aa2644813b48e8df01662eb36d0e1e9c8570
+oid sha256:34072aa1e01e6ea3b9e3e412d5865816e80bc8b65f51380cd2221e9b46d8eb1c
 size 4876059416

model-00009-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdbb35ea74ffb5d45461e74bfd99239919194910a748458792b3b87d8afb8dce
+oid sha256:5e2dd006f483441af0076cd4b3ad572f7c10ef6ac456eff0a3284fc7221eab7d
 size 4876059416

model-00010-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3226df592c7c99b4e395ce9af5358d6b105d48a758a5990a8c4aa4d32269b354
+oid sha256:c6d8c7e642ae0dfd55cf6e525a9f218392262122d6aa16f8ad480ea41e70b731
 size 4876059416

model-00011-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea2d1abb15bfaa5325c584b16f73ea84da8b035c76af656c7d3ab9cfe540f707
+oid sha256:6124bdf156fe7d99c7187e88e65d9715a8bd9746b4ba4402637a9ecfad91e43f
 size 4876059416

model-00012-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e30fe19ec6aff4e03d68eb778aa957232542322c1f0ec3f93854f0e14945788
+oid sha256:0cbcef5f9025a3c86cf248fc09be439280f6eb729db80ad661d71c8490489193
 size 4876059416

model-00013-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf4db43150a3495621e5ba85a18c7b84e8dc9984dc04432cdc143d2d63a302a6
+oid sha256:faf8c95d5b8e086c30e1d58c808ab5c7b3dae6b49e36aaf7bbb443c959c3d569
 size 4876059416

model-00014-of-00014.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1399acf53c0688d628995bfa59f8b32becce5f233e6c41ae54433f52ce7e3b7
+oid sha256:bb9704bfc526eda757df1be22bb99acd6990237efd892d4e8799d28cd08770ab
 size 2123397800

start_end.json

@@ -0,0 +1 @@
+{"start_time": "2025-04-29 10:28:29", "end_time": "2025-04-29 17:52:34"}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 4.968152866242038,
+    "total_flos": 4.6317288847048704e+17,
+    "train_loss": 0.30433785139750213,
+    "train_runtime": 26201.4476,
+    "train_samples_per_second": 1.908,
+    "train_steps_per_second": 0.015
+}

trainer_state.json

@@ -0,0 +1,2772 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.968152866242038,
+  "eval_steps": 500,
+  "global_step": 390,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.012738853503184714,
+      "grad_norm": 3.7423072661703385,
+      "learning_rate": 1.0256410256410257e-06,
+      "loss": 0.972,
+      "step": 1
+    },
+    {
+      "epoch": 0.025477707006369428,
+      "grad_norm": 4.014331193631274,
+      "learning_rate": 2.0512820512820513e-06,
+      "loss": 0.9843,
+      "step": 2
+    },
+    {
+      "epoch": 0.03821656050955414,
+      "grad_norm": 3.8392264390505235,
+      "learning_rate": 3.0769230769230774e-06,
+      "loss": 0.9844,
+      "step": 3
+    },
+    {
+      "epoch": 0.050955414012738856,
+      "grad_norm": 3.4206021200245513,
+      "learning_rate": 4.102564102564103e-06,
+      "loss": 0.9535,
+      "step": 4
+    },
+    {
+      "epoch": 0.06369426751592357,
+      "grad_norm": 2.844532001391515,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.9244,
+      "step": 5
+    },
+    {
+      "epoch": 0.07643312101910828,
+      "grad_norm": 2.019106639258673,
+      "learning_rate": 6.153846153846155e-06,
+      "loss": 0.8528,
+      "step": 6
+    },
+    {
+      "epoch": 0.08917197452229299,
+      "grad_norm": 1.57361042148724,
+      "learning_rate": 7.17948717948718e-06,
+      "loss": 0.8202,
+      "step": 7
+    },
+    {
+      "epoch": 0.10191082802547771,
+      "grad_norm": 1.3225717537131316,
+      "learning_rate": 8.205128205128205e-06,
+      "loss": 0.806,
+      "step": 8
+    },
+    {
+      "epoch": 0.11464968152866242,
+      "grad_norm": 1.297213830671725,
+      "learning_rate": 9.230769230769232e-06,
+      "loss": 0.8505,
+      "step": 9
+    },
+    {
+      "epoch": 0.12738853503184713,
+      "grad_norm": 1.924658811896663,
+      "learning_rate": 1.0256410256410256e-05,
+      "loss": 0.7536,
+      "step": 10
+    },
+    {
+      "epoch": 0.14012738853503184,
+      "grad_norm": 1.7407165339891915,
+      "learning_rate": 1.1282051282051283e-05,
+      "loss": 0.7524,
+      "step": 11
+    },
+    {
+      "epoch": 0.15286624203821655,
+      "grad_norm": 1.3558527366110018,
+      "learning_rate": 1.230769230769231e-05,
+      "loss": 0.7382,
+      "step": 12
+    },
+    {
+      "epoch": 0.16560509554140126,
+      "grad_norm": 1.0929459173306966,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.695,
+      "step": 13
+    },
+    {
+      "epoch": 0.17834394904458598,
+      "grad_norm": 1.2749157546624197,
+      "learning_rate": 1.435897435897436e-05,
+      "loss": 0.7111,
+      "step": 14
+    },
+    {
+      "epoch": 0.1910828025477707,
+      "grad_norm": 1.0998533610546595,
+      "learning_rate": 1.5384615384615387e-05,
+      "loss": 0.7091,
+      "step": 15
+    },
+    {
+      "epoch": 0.20382165605095542,
+      "grad_norm": 0.913903420689088,
+      "learning_rate": 1.641025641025641e-05,
+      "loss": 0.6703,
+      "step": 16
+    },
+    {
+      "epoch": 0.21656050955414013,
+      "grad_norm": 0.7661216217896927,
+      "learning_rate": 1.7435897435897438e-05,
+      "loss": 0.6948,
+      "step": 17
+    },
+    {
+      "epoch": 0.22929936305732485,
+      "grad_norm": 0.7437430749983936,
+      "learning_rate": 1.8461538461538465e-05,
+      "loss": 0.6176,
+      "step": 18
+    },
+    {
+      "epoch": 0.24203821656050956,
+      "grad_norm": 0.7446006831365574,
+      "learning_rate": 1.9487179487179488e-05,
+      "loss": 0.6591,
+      "step": 19
+    },
+    {
+      "epoch": 0.25477707006369427,
+      "grad_norm": 0.6792039966401283,
+      "learning_rate": 2.0512820512820512e-05,
+      "loss": 0.6359,
+      "step": 20
+    },
+    {
+      "epoch": 0.267515923566879,
+      "grad_norm": 0.5849626006077434,
+      "learning_rate": 2.153846153846154e-05,
+      "loss": 0.6271,
+      "step": 21
+    },
+    {
+      "epoch": 0.2802547770700637,
+      "grad_norm": 0.6919741782724808,
+      "learning_rate": 2.2564102564102566e-05,
+      "loss": 0.665,
+      "step": 22
+    },
+    {
+      "epoch": 0.2929936305732484,
+      "grad_norm": 0.7570356242077189,
+      "learning_rate": 2.3589743589743593e-05,
+      "loss": 0.6547,
+      "step": 23
+    },
+    {
+      "epoch": 0.3057324840764331,
+      "grad_norm": 0.6189824489932229,
+      "learning_rate": 2.461538461538462e-05,
+      "loss": 0.6146,
+      "step": 24
+    },
+    {
+      "epoch": 0.3184713375796178,
+      "grad_norm": 0.5416486550934623,
+      "learning_rate": 2.5641025641025646e-05,
+      "loss": 0.6467,
+      "step": 25
+    },
+    {
+      "epoch": 0.33121019108280253,
+      "grad_norm": 0.6011316533516919,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 0.6231,
+      "step": 26
+    },
+    {
+      "epoch": 0.34394904458598724,
+      "grad_norm": 0.5744339851945378,
+      "learning_rate": 2.7692307692307694e-05,
+      "loss": 0.5921,
+      "step": 27
+    },
+    {
+      "epoch": 0.35668789808917195,
+      "grad_norm": 0.5427273841273952,
+      "learning_rate": 2.871794871794872e-05,
+      "loss": 0.5984,
+      "step": 28
+    },
+    {
+      "epoch": 0.36942675159235666,
+      "grad_norm": 0.6587706620684388,
+      "learning_rate": 2.9743589743589747e-05,
+      "loss": 0.6258,
+      "step": 29
+    },
+    {
+      "epoch": 0.3821656050955414,
+      "grad_norm": 0.5241407477072185,
+      "learning_rate": 3.0769230769230774e-05,
+      "loss": 0.6057,
+      "step": 30
+    },
+    {
+      "epoch": 0.39490445859872614,
+      "grad_norm": 0.6221622861877016,
+      "learning_rate": 3.1794871794871795e-05,
+      "loss": 0.6028,
+      "step": 31
+    },
+    {
+      "epoch": 0.40764331210191085,
+      "grad_norm": 0.5873239727341781,
+      "learning_rate": 3.282051282051282e-05,
+      "loss": 0.5984,
+      "step": 32
+    },
+    {
+      "epoch": 0.42038216560509556,
+      "grad_norm": 0.5306206208481558,
+      "learning_rate": 3.384615384615385e-05,
+      "loss": 0.559,
+      "step": 33
+    },
+    {
+      "epoch": 0.43312101910828027,
+      "grad_norm": 0.5582669312616446,
+      "learning_rate": 3.4871794871794875e-05,
+      "loss": 0.6149,
+      "step": 34
+    },
+    {
+      "epoch": 0.445859872611465,
+      "grad_norm": 0.5000854344725365,
+      "learning_rate": 3.58974358974359e-05,
+      "loss": 0.6189,
+      "step": 35
+    },
+    {
+      "epoch": 0.4585987261146497,
+      "grad_norm": 0.527484918197263,
+      "learning_rate": 3.692307692307693e-05,
+      "loss": 0.6085,
+      "step": 36
+    },
+    {
+      "epoch": 0.4713375796178344,
+      "grad_norm": 0.48032262139418297,
+      "learning_rate": 3.794871794871795e-05,
+      "loss": 0.5931,
+      "step": 37
+    },
+    {
+      "epoch": 0.4840764331210191,
+      "grad_norm": 0.4279530947525605,
+      "learning_rate": 3.8974358974358976e-05,
+      "loss": 0.5781,
+      "step": 38
+    },
+    {
+      "epoch": 0.4968152866242038,
+      "grad_norm": 0.4900464691732963,
+      "learning_rate": 4e-05,
+      "loss": 0.6013,
+      "step": 39
+    },
+    {
+      "epoch": 0.5095541401273885,
+      "grad_norm": 0.4882711712476194,
+      "learning_rate": 3.9999198907597046e-05,
+      "loss": 0.5778,
+      "step": 40
+    },
+    {
+      "epoch": 0.5222929936305732,
+      "grad_norm": 0.4368372090967015,
+      "learning_rate": 3.9996795694563096e-05,
+      "loss": 0.567,
+      "step": 41
+    },
+    {
+      "epoch": 0.535031847133758,
+      "grad_norm": 0.46003529609029364,
+      "learning_rate": 3.999279055341771e-05,
+      "loss": 0.5919,
+      "step": 42
+    },
+    {
+      "epoch": 0.5477707006369427,
+      "grad_norm": 0.433443291072034,
+      "learning_rate": 3.998718380500971e-05,
+      "loss": 0.5608,
+      "step": 43
+    },
+    {
+      "epoch": 0.5605095541401274,
+      "grad_norm": 0.5579094629535033,
+      "learning_rate": 3.997997589849145e-05,
+      "loss": 0.5919,
+      "step": 44
+    },
+    {
+      "epoch": 0.5732484076433121,
+      "grad_norm": 0.5002127081158247,
+      "learning_rate": 3.9971167411282835e-05,
+      "loss": 0.6202,
+      "step": 45
+    },
+    {
+      "epoch": 0.5859872611464968,
+      "grad_norm": 0.45513028054084476,
+      "learning_rate": 3.99607590490251e-05,
+      "loss": 0.583,
+      "step": 46
+    },
+    {
+      "epoch": 0.5987261146496815,
+      "grad_norm": 0.49480268233907687,
+      "learning_rate": 3.9948751645524235e-05,
+      "loss": 0.6031,
+      "step": 47
+    },
+    {
+      "epoch": 0.6114649681528662,
+      "grad_norm": 0.4676370814108011,
+      "learning_rate": 3.9935146162684206e-05,
+      "loss": 0.5774,
+      "step": 48
+    },
+    {
+      "epoch": 0.6242038216560509,
+      "grad_norm": 0.42161416955688974,
+      "learning_rate": 3.9919943690429906e-05,
+      "loss": 0.5836,
+      "step": 49
+    },
+    {
+      "epoch": 0.6369426751592356,
+      "grad_norm": 0.4758290404818582,
+      "learning_rate": 3.9903145446619837e-05,
+      "loss": 0.5961,
+      "step": 50
+    },
+    {
+      "epoch": 0.6496815286624203,
+      "grad_norm": 0.4465799608039247,
+      "learning_rate": 3.9884752776948564e-05,
+      "loss": 0.5712,
+      "step": 51
+    },
+    {
+      "epoch": 0.6624203821656051,
+      "grad_norm": 0.4783306849945249,
+      "learning_rate": 3.9864767154838864e-05,
+      "loss": 0.5781,
+      "step": 52
+    },
+    {
+      "epoch": 0.6751592356687898,
+      "grad_norm": 0.47408070441336125,
+      "learning_rate": 3.9843190181323744e-05,
+      "loss": 0.5631,
+      "step": 53
+    },
+    {
+      "epoch": 0.6878980891719745,
+      "grad_norm": 0.44160447990297685,
+      "learning_rate": 3.982002358491817e-05,
+      "loss": 0.5793,
+      "step": 54
+    },
+    {
+      "epoch": 0.7006369426751592,
+      "grad_norm": 0.5440809599078149,
+      "learning_rate": 3.979526922148058e-05,
+      "loss": 0.5739,
+      "step": 55
+    },
+    {
+      "epoch": 0.7133757961783439,
+      "grad_norm": 0.4030353591762494,
+      "learning_rate": 3.9768929074064206e-05,
+      "loss": 0.5731,
+      "step": 56
+    },
+    {
+      "epoch": 0.7261146496815286,
+      "grad_norm": 0.47204667115023985,
+      "learning_rate": 3.9741005252758255e-05,
+      "loss": 0.5752,
+      "step": 57
+    },
+    {
+      "epoch": 0.7388535031847133,
+      "grad_norm": 0.3924967377535464,
+      "learning_rate": 3.971149999451886e-05,
+      "loss": 0.5723,
+      "step": 58
+    },
+    {
+      "epoch": 0.7515923566878981,
+      "grad_norm": 0.5036651859656847,
+      "learning_rate": 3.9680415662989806e-05,
+      "loss": 0.5816,
+      "step": 59
+    },
+    {
+      "epoch": 0.7643312101910829,
+      "grad_norm": 0.44083451558702347,
+      "learning_rate": 3.9647754748313294e-05,
+      "loss": 0.5878,
+      "step": 60
+    },
+    {
+      "epoch": 0.7770700636942676,
+      "grad_norm": 0.4848776951212216,
+      "learning_rate": 3.96135198669304e-05,
+      "loss": 0.5733,
+      "step": 61
+    },
+    {
+      "epoch": 0.7898089171974523,
+      "grad_norm": 0.4525502560691572,
+      "learning_rate": 3.957771376137144e-05,
+      "loss": 0.5558,
+      "step": 62
+    },
+    {
+      "epoch": 0.802547770700637,
+      "grad_norm": 0.4368149673012582,
+      "learning_rate": 3.954033930003634e-05,
+      "loss": 0.5766,
+      "step": 63
+    },
+    {
+      "epoch": 0.8152866242038217,
+      "grad_norm": 0.4530850844628041,
+      "learning_rate": 3.9501399476964806e-05,
+      "loss": 0.567,
+      "step": 64
+    },
+    {
+      "epoch": 0.8280254777070064,
+      "grad_norm": 0.4520029723511962,
+      "learning_rate": 3.946089741159648e-05,
+      "loss": 0.5747,
+      "step": 65
+    },
+    {
+      "epoch": 0.8407643312101911,
+      "grad_norm": 0.4458792601241768,
+      "learning_rate": 3.9418836348521045e-05,
+      "loss": 0.5567,
+      "step": 66
+    },
+    {
+      "epoch": 0.8535031847133758,
+      "grad_norm": 0.49824351969919184,
+      "learning_rate": 3.937521965721831e-05,
+      "loss": 0.5707,
+      "step": 67
+    },
+    {
+      "epoch": 0.8662420382165605,
+      "grad_norm": 0.5483292524184726,
+      "learning_rate": 3.933005083178828e-05,
+      "loss": 0.5947,
+      "step": 68
+    },
+    {
+      "epoch": 0.8789808917197452,
+      "grad_norm": 0.48089214203643743,
+      "learning_rate": 3.928333349067125e-05,
+      "loss": 0.5389,
+      "step": 69
+    },
+    {
+      "epoch": 0.89171974522293,
+      "grad_norm": 0.49454234464079616,
+      "learning_rate": 3.923507137635792e-05,
+      "loss": 0.5396,
+      "step": 70
+    },
+    {
+      "epoch": 0.9044585987261147,
+      "grad_norm": 0.43160503319375604,
+      "learning_rate": 3.9185268355089606e-05,
+      "loss": 0.5714,
+      "step": 71
+    },
+    {
+      "epoch": 0.9171974522292994,
+      "grad_norm": 0.4419560432757935,
+      "learning_rate": 3.913392841654851e-05,
+      "loss": 0.5623,
+      "step": 72
+    },
+    {
+      "epoch": 0.9299363057324841,
+      "grad_norm": 0.44484601437802485,
+      "learning_rate": 3.9081055673538093e-05,
+      "loss": 0.5606,
+      "step": 73
+    },
+    {
+      "epoch": 0.9426751592356688,
+      "grad_norm": 0.4085811628730557,
+      "learning_rate": 3.902665436165364e-05,
+      "loss": 0.5457,
+      "step": 74
+    },
+    {
+      "epoch": 0.9554140127388535,
+      "grad_norm": 0.4101366746857519,
+      "learning_rate": 3.897072883894291e-05,
+      "loss": 0.6062,
+      "step": 75
+    },
+    {
+      "epoch": 0.9681528662420382,
+      "grad_norm": 0.4532752791503924,
+      "learning_rate": 3.8913283585557054e-05,
+      "loss": 0.5861,
+      "step": 76
+    },
+    {
+      "epoch": 0.9808917197452229,
+      "grad_norm": 0.3783522323934715,
+      "learning_rate": 3.885432320339167e-05,
+      "loss": 0.5515,
+      "step": 77
+    },
+    {
+      "epoch": 0.9936305732484076,
+      "grad_norm": 0.4477233989258556,
+      "learning_rate": 3.879385241571817e-05,
+      "loss": 0.5707,
+      "step": 78
+    },
+    {
+      "epoch": 1.0063694267515924,
+      "grad_norm": 0.4629761435051509,
+      "learning_rate": 3.873187606680543e-05,
+      "loss": 0.4782,
+      "step": 79
+    },
+    {
+      "epoch": 1.019108280254777,
+      "grad_norm": 0.4840535828792894,
+      "learning_rate": 3.866839912153168e-05,
+      "loss": 0.418,
+      "step": 80
+    },
+    {
+      "epoch": 1.0318471337579618,
+      "grad_norm": 0.5036326924866047,
+      "learning_rate": 3.860342666498677e-05,
+      "loss": 0.3971,
+      "step": 81
+    },
+    {
+      "epoch": 1.0445859872611465,
+      "grad_norm": 0.6228426093815846,
+      "learning_rate": 3.853696390206484e-05,
+      "loss": 0.3766,
+      "step": 82
+    },
+    {
+      "epoch": 1.0573248407643312,
+      "grad_norm": 0.5561176865138412,
+      "learning_rate": 3.846901615704734e-05,
+      "loss": 0.3907,
+      "step": 83
+    },
+    {
+      "epoch": 1.070063694267516,
+      "grad_norm": 0.5545736824358007,
+      "learning_rate": 3.839958887317649e-05,
+      "loss": 0.4015,
+      "step": 84
+    },
+    {
+      "epoch": 1.0828025477707006,
+      "grad_norm": 0.5837935204920232,
+      "learning_rate": 3.832868761221926e-05,
+      "loss": 0.4234,
+      "step": 85
+    },
+    {
+      "epoch": 1.0955414012738853,
+      "grad_norm": 0.492326405924025,
+      "learning_rate": 3.825631805402182e-05,
+      "loss": 0.3696,
+      "step": 86
+    },
+    {
+      "epoch": 1.10828025477707,
+      "grad_norm": 0.5209834439614793,
+      "learning_rate": 3.818248599605448e-05,
+      "loss": 0.4019,
+      "step": 87
+    },
+    {
+      "epoch": 1.1210191082802548,
+      "grad_norm": 0.4885566025553211,
+      "learning_rate": 3.810719735294731e-05,
+      "loss": 0.3943,
+      "step": 88
+    },
+    {
+      "epoch": 1.1337579617834395,
+      "grad_norm": 0.47450364241586784,
+      "learning_rate": 3.8030458156016326e-05,
+      "loss": 0.3907,
+      "step": 89
+    },
+    {
+      "epoch": 1.1464968152866242,
+      "grad_norm": 0.4445266649179879,
+      "learning_rate": 3.795227455278029e-05,
+      "loss": 0.411,
+      "step": 90
+    },
+    {
+      "epoch": 1.1592356687898089,
+      "grad_norm": 0.46006224511432137,
+      "learning_rate": 3.787265280646825e-05,
+      "loss": 0.3926,
+      "step": 91
+    },
+    {
+      "epoch": 1.1719745222929936,
+      "grad_norm": 0.4270153987174869,
+      "learning_rate": 3.7791599295517825e-05,
+      "loss": 0.3655,
+      "step": 92
+    },
+    {
+      "epoch": 1.1847133757961783,
+      "grad_norm": 0.4878817140154042,
+      "learning_rate": 3.7709120513064196e-05,
+      "loss": 0.3798,
+      "step": 93
+    },
+    {
+      "epoch": 1.197452229299363,
+      "grad_norm": 0.46867955444134074,
+      "learning_rate": 3.762522306641998e-05,
+      "loss": 0.3631,
+      "step": 94
+    },
+    {
+      "epoch": 1.2101910828025477,
+      "grad_norm": 0.4419027945938377,
+      "learning_rate": 3.7539913676545874e-05,
+      "loss": 0.3884,
+      "step": 95
+    },
+    {
+      "epoch": 1.2229299363057324,
+      "grad_norm": 0.40849750356307674,
+      "learning_rate": 3.745319917751229e-05,
+      "loss": 0.3933,
+      "step": 96
+    },
+    {
+      "epoch": 1.2356687898089171,
+      "grad_norm": 0.5655828636490249,
+      "learning_rate": 3.736508651595188e-05,
+      "loss": 0.3992,
+      "step": 97
+    },
+    {
+      "epoch": 1.2484076433121019,
+      "grad_norm": 0.4781160371870933,
+      "learning_rate": 3.727558275050301e-05,
+      "loss": 0.4138,
+      "step": 98
+    },
+    {
+      "epoch": 1.2611464968152866,
+      "grad_norm": 0.4213212240808343,
+      "learning_rate": 3.718469505124434e-05,
+      "loss": 0.3848,
+      "step": 99
+    },
+    {
+      "epoch": 1.2738853503184713,
+      "grad_norm": 0.4274617938587094,
+      "learning_rate": 3.709243069912041e-05,
+      "loss": 0.3763,
+      "step": 100
+    },
+    {
+      "epoch": 1.286624203821656,
+      "grad_norm": 0.4220383272407581,
+      "learning_rate": 3.699879708535838e-05,
+      "loss": 0.378,
+      "step": 101
+    },
+    {
+      "epoch": 1.2993630573248407,
+      "grad_norm": 0.4171375820693096,
+      "learning_rate": 3.69038017108759e-05,
+      "loss": 0.41,
+      "step": 102
+    },
+    {
+      "epoch": 1.3121019108280254,
+      "grad_norm": 0.4382305912659545,
+      "learning_rate": 3.680745218568026e-05,
+      "loss": 0.4059,
+      "step": 103
+    },
+    {
+      "epoch": 1.3248407643312101,
+      "grad_norm": 0.4165670382998542,
+      "learning_rate": 3.6709756228258735e-05,
+      "loss": 0.3844,
+      "step": 104
+    },
+    {
+      "epoch": 1.3375796178343948,
+      "grad_norm": 0.4406466854897885,
+      "learning_rate": 3.6610721664960236e-05,
+      "loss": 0.4006,
+      "step": 105
+    },
+    {
+      "epoch": 1.3503184713375795,
+      "grad_norm": 0.4481397864298152,
+      "learning_rate": 3.65103564293684e-05,
+      "loss": 0.384,
+      "step": 106
+    },
+    {
+      "epoch": 1.3630573248407643,
+      "grad_norm": 0.43576569305511864,
+      "learning_rate": 3.640866856166601e-05,
+      "loss": 0.3816,
+      "step": 107
+    },
+    {
+      "epoch": 1.3757961783439492,
+      "grad_norm": 0.41729038843610433,
+      "learning_rate": 3.6305666207990886e-05,
+      "loss": 0.417,
+      "step": 108
+    },
+    {
+      "epoch": 1.388535031847134,
+      "grad_norm": 0.4880218086268087,
+      "learning_rate": 3.6201357619783336e-05,
+      "loss": 0.4033,
+      "step": 109
+    },
+    {
+      "epoch": 1.4012738853503186,
+      "grad_norm": 0.41047644449018167,
+      "learning_rate": 3.609575115312511e-05,
+      "loss": 0.3797,
+      "step": 110
+    },
+    {
+      "epoch": 1.4140127388535033,
+      "grad_norm": 0.4822480986437211,
+      "learning_rate": 3.598885526807003e-05,
+      "loss": 0.4076,
+      "step": 111
+    },
+    {
+      "epoch": 1.426751592356688,
+      "grad_norm": 0.4459969963783435,
+      "learning_rate": 3.5880678527966224e-05,
+      "loss": 0.384,
+      "step": 112
+    },
+    {
+      "epoch": 1.4394904458598727,
+      "grad_norm": 0.4108885261851228,
+      "learning_rate": 3.577122959877017e-05,
+      "loss": 0.3775,
+      "step": 113
+    },
+    {
+      "epoch": 1.4522292993630574,
+      "grad_norm": 0.4264177407871251,
+      "learning_rate": 3.566051724835245e-05,
+      "loss": 0.3723,
+      "step": 114
+    },
+    {
+      "epoch": 1.4649681528662422,
+      "grad_norm": 0.46254534003867953,
+      "learning_rate": 3.554855034579532e-05,
+      "loss": 0.3905,
+      "step": 115
+    },
+    {
+      "epoch": 1.4777070063694269,
+      "grad_norm": 0.43159939406328157,
+      "learning_rate": 3.5435337860682304e-05,
+      "loss": 0.3968,
+      "step": 116
+    },
+    {
+      "epoch": 1.4904458598726116,
+      "grad_norm": 0.40378318420157067,
+      "learning_rate": 3.532088886237956e-05,
+      "loss": 0.3986,
+      "step": 117
+    },
+    {
+      "epoch": 1.5031847133757963,
+      "grad_norm": 0.4010856819942095,
+      "learning_rate": 3.520521251930941e-05,
+      "loss": 0.4018,
+      "step": 118
+    },
+    {
+      "epoch": 1.515923566878981,
+      "grad_norm": 0.3953489181151893,
+      "learning_rate": 3.5088318098215805e-05,
+      "loss": 0.3877,
+      "step": 119
+    },
+    {
+      "epoch": 1.5286624203821657,
+      "grad_norm": 0.4360231716627973,
+      "learning_rate": 3.497021496342203e-05,
+      "loss": 0.3619,
+      "step": 120
+    },
+    {
+      "epoch": 1.5414012738853504,
+      "grad_norm": 0.42883040333822336,
+      "learning_rate": 3.485091257608047e-05,
+      "loss": 0.3917,
+      "step": 121
+    },
+    {
+      "epoch": 1.5541401273885351,
+      "grad_norm": 0.3949540394409347,
+      "learning_rate": 3.473042049341474e-05,
+      "loss": 0.3795,
+      "step": 122
+    },
+    {
+      "epoch": 1.5668789808917198,
+      "grad_norm": 0.4613117397583663,
+      "learning_rate": 3.4608748367954064e-05,
+      "loss": 0.3996,
+      "step": 123
+    },
+    {
+      "epoch": 1.5796178343949046,
+      "grad_norm": 0.3602038452810683,
+      "learning_rate": 3.4485905946759965e-05,
+      "loss": 0.3716,
+      "step": 124
+    },
+    {
+      "epoch": 1.5923566878980893,
+      "grad_norm": 0.3931259570104483,
+      "learning_rate": 3.4361903070645484e-05,
+      "loss": 0.3847,
+      "step": 125
+    },
+    {
+      "epoch": 1.605095541401274,
+      "grad_norm": 0.4433362238715447,
+      "learning_rate": 3.423674967338681e-05,
+      "loss": 0.3824,
+      "step": 126
+    },
+    {
+      "epoch": 1.6178343949044587,
+      "grad_norm": 0.3897182981017782,
+      "learning_rate": 3.411045578092754e-05,
+      "loss": 0.382,
+      "step": 127
+    },
+    {
+      "epoch": 1.6305732484076434,
+      "grad_norm": 0.39580662710002257,
+      "learning_rate": 3.398303151057543e-05,
+      "loss": 0.3872,
+      "step": 128
+    },
+    {
+      "epoch": 1.643312101910828,
+      "grad_norm": 0.36389859578115363,
+      "learning_rate": 3.385448707019199e-05,
+      "loss": 0.393,
+      "step": 129
+    },
+    {
+      "epoch": 1.6560509554140128,
+      "grad_norm": 0.3760822336320154,
+      "learning_rate": 3.372483275737468e-05,
+      "loss": 0.4038,
+      "step": 130
+    },
+    {
+      "epoch": 1.6687898089171975,
+      "grad_norm": 0.40124915655150517,
+      "learning_rate": 3.359407895863199e-05,
+      "loss": 0.3828,
+      "step": 131
+    },
+    {
+      "epoch": 1.6815286624203822,
+      "grad_norm": 0.3492681667679969,
+      "learning_rate": 3.34622361485514e-05,
+      "loss": 0.3988,
+      "step": 132
+    },
+    {
+      "epoch": 1.694267515923567,
+      "grad_norm": 0.4359639458909479,
+      "learning_rate": 3.332931488896029e-05,
+      "loss": 0.4123,
+      "step": 133
+    },
+    {
+      "epoch": 1.7070063694267517,
+      "grad_norm": 0.35765716673817444,
+      "learning_rate": 3.319532582807977e-05,
+      "loss": 0.4025,
+      "step": 134
+    },
+    {
+      "epoch": 1.7197452229299364,
+      "grad_norm": 0.4181343414020294,
+      "learning_rate": 3.30602796996717e-05,
+      "loss": 0.3988,
+      "step": 135
+    },
+    {
+      "epoch": 1.732484076433121,
+      "grad_norm": 0.3671754645675639,
+      "learning_rate": 3.2924187322178865e-05,
+      "loss": 0.3866,
+      "step": 136
+    },
+    {
+      "epoch": 1.7452229299363058,
+      "grad_norm": 0.40706707942452913,
+      "learning_rate": 3.278705959785821e-05,
+      "loss": 0.4071,
+      "step": 137
+    },
+    {
+      "epoch": 1.7579617834394905,
+      "grad_norm": 0.40132966074561627,
+      "learning_rate": 3.2648907511907544e-05,
+      "loss": 0.3915,
+      "step": 138
+    },
+    {
+      "epoch": 1.7707006369426752,
+      "grad_norm": 0.356092810738192,
+      "learning_rate": 3.250974213158555e-05,
+      "loss": 0.3881,
+      "step": 139
+    },
+    {
+      "epoch": 1.78343949044586,
+      "grad_norm": 0.4088495881297229,
+      "learning_rate": 3.23695746053251e-05,
+      "loss": 0.3956,
+      "step": 140
+    },
+    {
+      "epoch": 1.7961783439490446,
+      "grad_norm": 0.3568174986894459,
+      "learning_rate": 3.222841616184025e-05,
+      "loss": 0.4264,
+      "step": 141
+    },
+    {
+      "epoch": 1.8089171974522293,
+      "grad_norm": 0.3828743980988108,
+      "learning_rate": 3.208627810922665e-05,
+      "loss": 0.3934,
+      "step": 142
+    },
+    {
+      "epoch": 1.821656050955414,
+      "grad_norm": 0.40695432826523725,
+      "learning_rate": 3.194317183405573e-05,
+      "loss": 0.3538,
+      "step": 143
+    },
+    {
+      "epoch": 1.8343949044585988,
+      "grad_norm": 0.34980415470555404,
+      "learning_rate": 3.1799108800462466e-05,
+      "loss": 0.3751,
+      "step": 144
+    },
+    {
+      "epoch": 1.8471337579617835,
+      "grad_norm": 0.4171552280293834,
+      "learning_rate": 3.1654100549227024e-05,
+      "loss": 0.3823,
+      "step": 145
+    },
+    {
+      "epoch": 1.8598726114649682,
+      "grad_norm": 0.3856839605835017,
+      "learning_rate": 3.1508158696850275e-05,
+      "loss": 0.3732,
+      "step": 146
+    },
+    {
+      "epoch": 1.872611464968153,
+      "grad_norm": 0.3534737799953714,
+      "learning_rate": 3.136129493462312e-05,
+      "loss": 0.4172,
+      "step": 147
+    },
+    {
+      "epoch": 1.8853503184713376,
+      "grad_norm": 0.40251325811058225,
+      "learning_rate": 3.121352102768998e-05,
+      "loss": 0.3775,
+      "step": 148
+    },
+    {
+      "epoch": 1.8980891719745223,
+      "grad_norm": 0.39323479639785724,
+      "learning_rate": 3.106484881410628e-05,
+      "loss": 0.3929,
+      "step": 149
+    },
+    {
+      "epoch": 1.910828025477707,
+      "grad_norm": 0.3948424894576429,
+      "learning_rate": 3.091529020389009e-05,
+      "loss": 0.4098,
+      "step": 150
+    },
+    {
+      "epoch": 1.9235668789808917,
+      "grad_norm": 0.3798104391658232,
+      "learning_rate": 3.076485717806808e-05,
+      "loss": 0.3865,
+      "step": 151
+    },
+    {
+      "epoch": 1.9363057324840764,
+      "grad_norm": 0.42664388139031734,
+      "learning_rate": 3.061356178771564e-05,
+      "loss": 0.4049,
+      "step": 152
+    },
+    {
+      "epoch": 1.9490445859872612,
+      "grad_norm": 0.38754913155372867,
+      "learning_rate": 3.0461416152991555e-05,
+      "loss": 0.3863,
+      "step": 153
+    },
+    {
+      "epoch": 1.9617834394904459,
+      "grad_norm": 0.41477873550604283,
+      "learning_rate": 3.0308432462167045e-05,
+      "loss": 0.3887,
+      "step": 154
+    },
+    {
+      "epoch": 1.9745222929936306,
+      "grad_norm": 0.4140640968780079,
+      "learning_rate": 3.015462297064936e-05,
+      "loss": 0.3936,
+      "step": 155
+    },
+    {
+      "epoch": 1.9872611464968153,
+      "grad_norm": 0.38731104462679833,
+      "learning_rate": 3.0000000000000004e-05,
+      "loss": 0.383,
+      "step": 156
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 0.3728294872032267,
+      "learning_rate": 2.98445759369477e-05,
+      "loss": 0.3533,
+      "step": 157
+    },
+    {
+      "epoch": 2.0127388535031847,
+      "grad_norm": 0.5326083329193402,
+      "learning_rate": 2.9688363232396056e-05,
+      "loss": 0.2746,
+      "step": 158
+    },
+    {
+      "epoch": 2.0254777070063694,
+      "grad_norm": 0.4241037663468422,
+      "learning_rate": 2.9531374400426158e-05,
+      "loss": 0.2451,
+      "step": 159
+    },
+    {
+      "epoch": 2.038216560509554,
+      "grad_norm": 0.5238076547115884,
+      "learning_rate": 2.9373622017294075e-05,
+      "loss": 0.2426,
+      "step": 160
+    },
+    {
+      "epoch": 2.050955414012739,
+      "grad_norm": 0.43750634458426324,
+      "learning_rate": 2.9215118720423375e-05,
+      "loss": 0.2461,
+      "step": 161
+    },
+    {
+      "epoch": 2.0636942675159236,
+      "grad_norm": 0.5426569564960375,
+      "learning_rate": 2.9055877207392752e-05,
+      "loss": 0.2396,
+      "step": 162
+    },
+    {
+      "epoch": 2.0764331210191083,
+      "grad_norm": 0.4603442806084417,
+      "learning_rate": 2.8895910234918828e-05,
+      "loss": 0.2423,
+      "step": 163
+    },
+    {
+      "epoch": 2.089171974522293,
+      "grad_norm": 0.4783694185949688,
+      "learning_rate": 2.873523061783426e-05,
+      "loss": 0.2522,
+      "step": 164
+    },
+    {
+      "epoch": 2.1019108280254777,
+      "grad_norm": 0.4155346213843112,
+      "learning_rate": 2.8573851228061084e-05,
+      "loss": 0.2384,
+      "step": 165
+    },
+    {
+      "epoch": 2.1146496815286624,
+      "grad_norm": 0.4032696957971972,
+      "learning_rate": 2.8411784993579633e-05,
+      "loss": 0.2505,
+      "step": 166
+    },
+    {
+      "epoch": 2.127388535031847,
+      "grad_norm": 0.3898038039651499,
+      "learning_rate": 2.8249044897392814e-05,
+      "loss": 0.2701,
+      "step": 167
+    },
+    {
+      "epoch": 2.140127388535032,
+      "grad_norm": 0.40575399978730575,
+      "learning_rate": 2.80856439764861e-05,
+      "loss": 0.2305,
+      "step": 168
+    },
+    {
+      "epoch": 2.1528662420382165,
+      "grad_norm": 0.37004856559051524,
+      "learning_rate": 2.792159532078314e-05,
+      "loss": 0.2451,
+      "step": 169
+    },
+    {
+      "epoch": 2.1656050955414012,
+      "grad_norm": 0.39808666444917457,
+      "learning_rate": 2.77569120720971e-05,
+      "loss": 0.258,
+      "step": 170
+    },
+    {
+      "epoch": 2.178343949044586,
+      "grad_norm": 0.36955071620464847,
+      "learning_rate": 2.7591607423077932e-05,
+      "loss": 0.2292,
+      "step": 171
+    },
+    {
+      "epoch": 2.1910828025477707,
+      "grad_norm": 0.336518551638344,
+      "learning_rate": 2.7425694616155474e-05,
+      "loss": 0.2477,
+      "step": 172
+    },
+    {
+      "epoch": 2.2038216560509554,
+      "grad_norm": 0.4140957079330811,
+      "learning_rate": 2.7259186942478656e-05,
+      "loss": 0.2255,
+      "step": 173
+    },
+    {
+      "epoch": 2.21656050955414,
+      "grad_norm": 0.32308874630904505,
+      "learning_rate": 2.7092097740850712e-05,
+      "loss": 0.237,
+      "step": 174
+    },
+    {
+      "epoch": 2.229299363057325,
+      "grad_norm": 0.38917894801947195,
+      "learning_rate": 2.692444039666066e-05,
+      "loss": 0.2514,
+      "step": 175
+    },
+    {
+      "epoch": 2.2420382165605095,
+      "grad_norm": 0.32626408176829913,
+      "learning_rate": 2.6756228340810946e-05,
+      "loss": 0.2317,
+      "step": 176
+    },
+    {
+      "epoch": 2.254777070063694,
+      "grad_norm": 0.34013561380822094,
+      "learning_rate": 2.6587475048641596e-05,
+      "loss": 0.2553,
+      "step": 177
+    },
+    {
+      "epoch": 2.267515923566879,
+      "grad_norm": 0.37726290374312765,
+      "learning_rate": 2.6418194038850634e-05,
+      "loss": 0.2438,
+      "step": 178
+    },
+    {
+      "epoch": 2.2802547770700636,
+      "grad_norm": 0.3524165115081538,
+      "learning_rate": 2.624839887241115e-05,
+      "loss": 0.231,
+      "step": 179
+    },
+    {
+      "epoch": 2.2929936305732483,
+      "grad_norm": 0.32629627151791846,
+      "learning_rate": 2.607810315148494e-05,
+      "loss": 0.2278,
+      "step": 180
+    },
+    {
+      "epoch": 2.305732484076433,
+      "grad_norm": 0.41583661942751143,
+      "learning_rate": 2.5907320518332827e-05,
+      "loss": 0.242,
+      "step": 181
+    },
+    {
+      "epoch": 2.3184713375796178,
+      "grad_norm": 0.30326537675872595,
+      "learning_rate": 2.5736064654221808e-05,
+      "loss": 0.2488,
+      "step": 182
+    },
+    {
+      "epoch": 2.3312101910828025,
+      "grad_norm": 0.3469983421489293,
+      "learning_rate": 2.5564349278329056e-05,
+      "loss": 0.2512,
+      "step": 183
+    },
+    {
+      "epoch": 2.343949044585987,
+      "grad_norm": 0.34347379957192603,
+      "learning_rate": 2.539218814664288e-05,
+      "loss": 0.2477,
+      "step": 184
+    },
+    {
+      "epoch": 2.356687898089172,
+      "grad_norm": 0.28641537582695287,
+      "learning_rate": 2.521959505086075e-05,
+      "loss": 0.2362,
+      "step": 185
+    },
+    {
+      "epoch": 2.3694267515923566,
+      "grad_norm": 0.3690167657528722,
+      "learning_rate": 2.5046583817284437e-05,
+      "loss": 0.2587,
+      "step": 186
+    },
+    {
+      "epoch": 2.3821656050955413,
+      "grad_norm": 0.2961653899116515,
+      "learning_rate": 2.487316830571244e-05,
+      "loss": 0.2311,
+      "step": 187
+    },
+    {
+      "epoch": 2.394904458598726,
+      "grad_norm": 0.29521846638552274,
+      "learning_rate": 2.4699362408329646e-05,
+      "loss": 0.2327,
+      "step": 188
+    },
+    {
+      "epoch": 2.4076433121019107,
+      "grad_norm": 0.29770180340093805,
+      "learning_rate": 2.4525180048594452e-05,
+      "loss": 0.2544,
+      "step": 189
+    },
+    {
+      "epoch": 2.4203821656050954,
+      "grad_norm": 0.2913325444490102,
+      "learning_rate": 2.435063518012335e-05,
+      "loss": 0.2247,
+      "step": 190
+    },
+    {
+      "epoch": 2.43312101910828,
+      "grad_norm": 0.3384396420749506,
+      "learning_rate": 2.4175741785573177e-05,
+      "loss": 0.2379,
+      "step": 191
+    },
+    {
+      "epoch": 2.445859872611465,
+      "grad_norm": 0.30159462621181515,
+      "learning_rate": 2.4000513875520892e-05,
+      "loss": 0.2152,
+      "step": 192
+    },
+    {
+      "epoch": 2.4585987261146496,
+      "grad_norm": 0.29567483557148394,
+      "learning_rate": 2.3824965487341247e-05,
+      "loss": 0.2402,
+      "step": 193
+    },
+    {
+      "epoch": 2.4713375796178343,
+      "grad_norm": 0.2769989858820714,
+      "learning_rate": 2.3649110684082258e-05,
+      "loss": 0.2597,
+      "step": 194
+    },
+    {
+      "epoch": 2.484076433121019,
+      "grad_norm": 0.29336394029366775,
+      "learning_rate": 2.3472963553338614e-05,
+      "loss": 0.2273,
+      "step": 195
+    },
+    {
+      "epoch": 2.4968152866242037,
+      "grad_norm": 0.2838756230989672,
+      "learning_rate": 2.3296538206123134e-05,
+      "loss": 0.2389,
+      "step": 196
+    },
+    {
+      "epoch": 2.5095541401273884,
+      "grad_norm": 0.2626418361521103,
+      "learning_rate": 2.311984877573636e-05,
+      "loss": 0.2352,
+      "step": 197
+    },
+    {
+      "epoch": 2.522292993630573,
+      "grad_norm": 0.26718958075760085,
+      "learning_rate": 2.2942909416634326e-05,
+      "loss": 0.2357,
+      "step": 198
+    },
+    {
+      "epoch": 2.535031847133758,
+      "grad_norm": 0.2845956555682719,
+      "learning_rate": 2.2765734303294666e-05,
+      "loss": 0.2371,
+      "step": 199
+    },
+    {
+      "epoch": 2.5477707006369426,
+      "grad_norm": 0.2640776304490595,
+      "learning_rate": 2.2588337629081107e-05,
+      "loss": 0.2384,
+      "step": 200
+    },
+    {
+      "epoch": 2.5605095541401273,
+      "grad_norm": 0.2848898116027439,
+      "learning_rate": 2.2410733605106462e-05,
+      "loss": 0.2217,
+      "step": 201
+    },
+    {
+      "epoch": 2.573248407643312,
+      "grad_norm": 0.2812237619598167,
+      "learning_rate": 2.2232936459094158e-05,
+      "loss": 0.2403,
+      "step": 202
+    },
+    {
+      "epoch": 2.5859872611464967,
+      "grad_norm": 0.29409588083070315,
+      "learning_rate": 2.205496043423849e-05,
+      "loss": 0.2266,
+      "step": 203
+    },
+    {
+      "epoch": 2.5987261146496814,
+      "grad_norm": 0.32256548006968067,
+      "learning_rate": 2.1876819788063586e-05,
+      "loss": 0.2547,
+      "step": 204
+    },
+    {
+      "epoch": 2.611464968152866,
+      "grad_norm": 0.32874439082160606,
+      "learning_rate": 2.16985287912813e-05,
+      "loss": 0.2261,
+      "step": 205
+    },
+    {
+      "epoch": 2.624203821656051,
+      "grad_norm": 0.3266191143621023,
+      "learning_rate": 2.1520101726647922e-05,
+      "loss": 0.2321,
+      "step": 206
+    },
+    {
+      "epoch": 2.6369426751592355,
+      "grad_norm": 0.30563025473541944,
+      "learning_rate": 2.1341552887820048e-05,
+      "loss": 0.2394,
+      "step": 207
+    },
+    {
+      "epoch": 2.6496815286624202,
+      "grad_norm": 0.2869377241233248,
+      "learning_rate": 2.1162896578209517e-05,
+      "loss": 0.2409,
+      "step": 208
+    },
+    {
+      "epoch": 2.662420382165605,
+      "grad_norm": 0.30715126599292863,
+      "learning_rate": 2.0984147109837564e-05,
+      "loss": 0.2423,
+      "step": 209
+    },
+    {
+      "epoch": 2.6751592356687897,
+      "grad_norm": 0.2894859890668604,
+      "learning_rate": 2.0805318802188307e-05,
+      "loss": 0.264,
+      "step": 210
+    },
+    {
+      "epoch": 2.6878980891719744,
+      "grad_norm": 0.28673853671855953,
+      "learning_rate": 2.0626425981061608e-05,
+      "loss": 0.2632,
+      "step": 211
+    },
+    {
+      "epoch": 2.700636942675159,
+      "grad_norm": 0.26547959114709113,
+      "learning_rate": 2.0447482977425465e-05,
+      "loss": 0.2165,
+      "step": 212
+    },
+    {
+      "epoch": 2.713375796178344,
+      "grad_norm": 0.29689259712489163,
+      "learning_rate": 2.0268504126267952e-05,
+      "loss": 0.254,
+      "step": 213
+    },
+    {
+      "epoch": 2.7261146496815285,
+      "grad_norm": 0.30794580549232115,
+      "learning_rate": 2.008950376544887e-05,
+      "loss": 0.234,
+      "step": 214
+    },
+    {
+      "epoch": 2.738853503184713,
+      "grad_norm": 0.29029164891637815,
+      "learning_rate": 1.9910496234551132e-05,
+      "loss": 0.2423,
+      "step": 215
+    },
+    {
+      "epoch": 2.7515923566878984,
+      "grad_norm": 0.28445043266611614,
+      "learning_rate": 1.9731495873732055e-05,
+      "loss": 0.2399,
+      "step": 216
+    },
+    {
+      "epoch": 2.7643312101910826,
+      "grad_norm": 0.28463685628918173,
+      "learning_rate": 1.9552517022574542e-05,
+      "loss": 0.2246,
+      "step": 217
+    },
+    {
+      "epoch": 2.777070063694268,
+      "grad_norm": 0.29242221813409025,
+      "learning_rate": 1.93735740189384e-05,
+      "loss": 0.2276,
+      "step": 218
+    },
+    {
+      "epoch": 2.789808917197452,
+      "grad_norm": 0.309031276466736,
+      "learning_rate": 1.9194681197811703e-05,
+      "loss": 0.2522,
+      "step": 219
+    },
+    {
+      "epoch": 2.802547770700637,
+      "grad_norm": 0.2769431867934569,
+      "learning_rate": 1.901585289016244e-05,
+      "loss": 0.2247,
+      "step": 220
+    },
+    {
+      "epoch": 2.8152866242038215,
+      "grad_norm": 0.27968185341237434,
+      "learning_rate": 1.8837103421790486e-05,
+      "loss": 0.2227,
+      "step": 221
+    },
+    {
+      "epoch": 2.8280254777070066,
+      "grad_norm": 0.2888083036658392,
+      "learning_rate": 1.8658447112179952e-05,
+      "loss": 0.2374,
+      "step": 222
+    },
+    {
+      "epoch": 2.840764331210191,
+      "grad_norm": 0.28582364336230653,
+      "learning_rate": 1.8479898273352084e-05,
+      "loss": 0.2536,
+      "step": 223
+    },
+    {
+      "epoch": 2.853503184713376,
+      "grad_norm": 0.2935452815809231,
+      "learning_rate": 1.83014712087187e-05,
+      "loss": 0.2181,
+      "step": 224
+    },
+    {
+      "epoch": 2.8662420382165603,
+      "grad_norm": 0.30167338921175085,
+      "learning_rate": 1.8123180211936417e-05,
+      "loss": 0.2279,
+      "step": 225
+    },
+    {
+      "epoch": 2.8789808917197455,
+      "grad_norm": 0.28313988031191023,
+      "learning_rate": 1.794503956576152e-05,
+      "loss": 0.2136,
+      "step": 226
+    },
+    {
+      "epoch": 2.8917197452229297,
+      "grad_norm": 0.3036257204617132,
+      "learning_rate": 1.776706354090585e-05,
+      "loss": 0.2431,
+      "step": 227
+    },
+    {
+      "epoch": 2.904458598726115,
+      "grad_norm": 0.28848979346617437,
+      "learning_rate": 1.758926639489354e-05,
+      "loss": 0.2423,
+      "step": 228
+    },
+    {
+      "epoch": 2.917197452229299,
+      "grad_norm": 0.272764590445302,
+      "learning_rate": 1.7411662370918893e-05,
+      "loss": 0.2301,
+      "step": 229
+    },
+    {
+      "epoch": 2.9299363057324843,
+      "grad_norm": 0.3047366111218703,
+      "learning_rate": 1.7234265696705344e-05,
+      "loss": 0.2548,
+      "step": 230
+    },
+    {
+      "epoch": 2.9426751592356686,
+      "grad_norm": 0.29141657758684564,
+      "learning_rate": 1.7057090583365678e-05,
+      "loss": 0.2363,
+      "step": 231
+    },
+    {
+      "epoch": 2.9554140127388537,
+      "grad_norm": 0.2905701392836334,
+      "learning_rate": 1.6880151224263646e-05,
+      "loss": 0.2429,
+      "step": 232
+    },
+    {
+      "epoch": 2.968152866242038,
+      "grad_norm": 0.3055726799006245,
+      "learning_rate": 1.6703461793876876e-05,
+      "loss": 0.241,
+      "step": 233
+    },
+    {
+      "epoch": 2.980891719745223,
+      "grad_norm": 0.27702453651096914,
+      "learning_rate": 1.6527036446661396e-05,
+      "loss": 0.2785,
+      "step": 234
+    },
+    {
+      "epoch": 2.9936305732484074,
+      "grad_norm": 0.2958171896040813,
+      "learning_rate": 1.635088931591775e-05,
+      "loss": 0.2494,
+      "step": 235
+    },
+    {
+      "epoch": 3.0063694267515926,
+      "grad_norm": 0.3339252283259285,
+      "learning_rate": 1.6175034512658753e-05,
+      "loss": 0.1814,
+      "step": 236
+    },
+    {
+      "epoch": 3.0191082802547773,
+      "grad_norm": 0.33323829415209344,
+      "learning_rate": 1.5999486124479115e-05,
+      "loss": 0.1519,
+      "step": 237
+    },
+    {
+      "epoch": 3.031847133757962,
+      "grad_norm": 0.263725808295815,
+      "learning_rate": 1.5824258214426833e-05,
+      "loss": 0.1756,
+      "step": 238
+    },
+    {
+      "epoch": 3.0445859872611467,
+      "grad_norm": 0.3263063861580799,
+      "learning_rate": 1.5649364819876655e-05,
+      "loss": 0.1681,
+      "step": 239
+    },
+    {
+      "epoch": 3.0573248407643314,
+      "grad_norm": 0.3332704699662475,
+      "learning_rate": 1.547481995140556e-05,
+      "loss": 0.1533,
+      "step": 240
+    },
+    {
+      "epoch": 3.070063694267516,
+      "grad_norm": 0.3290607170386027,
+      "learning_rate": 1.5300637591670357e-05,
+      "loss": 0.1491,
+      "step": 241
+    },
+    {
+      "epoch": 3.082802547770701,
+      "grad_norm": 0.27626642274366475,
+      "learning_rate": 1.5126831694287564e-05,
+      "loss": 0.1386,
+      "step": 242
+    },
+    {
+      "epoch": 3.0955414012738856,
+      "grad_norm": 0.24383948766661087,
+      "learning_rate": 1.4953416182715566e-05,
+      "loss": 0.1454,
+      "step": 243
+    },
+    {
+      "epoch": 3.1082802547770703,
+      "grad_norm": 0.2565144051732309,
+      "learning_rate": 1.478040494913926e-05,
+      "loss": 0.1297,
+      "step": 244
+    },
+    {
+      "epoch": 3.121019108280255,
+      "grad_norm": 0.2854715023231579,
+      "learning_rate": 1.460781185335713e-05,
+      "loss": 0.1611,
+      "step": 245
+    },
+    {
+      "epoch": 3.1337579617834397,
+      "grad_norm": 0.23688489896659726,
+      "learning_rate": 1.443565072167095e-05,
+      "loss": 0.1537,
+      "step": 246
+    },
+    {
+      "epoch": 3.1464968152866244,
+      "grad_norm": 0.23498017729946113,
+      "learning_rate": 1.4263935345778202e-05,
+      "loss": 0.128,
+      "step": 247
+    },
+    {
+      "epoch": 3.159235668789809,
+      "grad_norm": 0.25210838232322463,
+      "learning_rate": 1.409267948166718e-05,
+      "loss": 0.1529,
+      "step": 248
+    },
+    {
+      "epoch": 3.171974522292994,
+      "grad_norm": 0.23817238492456902,
+      "learning_rate": 1.3921896848515064e-05,
+      "loss": 0.1431,
+      "step": 249
+    },
+    {
+      "epoch": 3.1847133757961785,
+      "grad_norm": 0.23878818674795252,
+      "learning_rate": 1.3751601127588849e-05,
+      "loss": 0.1511,
+      "step": 250
+    },
+    {
+      "epoch": 3.1974522292993632,
+      "grad_norm": 0.2253169671801177,
+      "learning_rate": 1.3581805961149371e-05,
+      "loss": 0.171,
+      "step": 251
+    },
+    {
+      "epoch": 3.210191082802548,
+      "grad_norm": 0.2399433284877313,
+      "learning_rate": 1.341252495135841e-05,
+      "loss": 0.125,
+      "step": 252
+    },
+    {
+      "epoch": 3.2229299363057327,
+      "grad_norm": 0.2482037509566062,
+      "learning_rate": 1.324377165918906e-05,
+      "loss": 0.1332,
+      "step": 253
+    },
+    {
+      "epoch": 3.2356687898089174,
+      "grad_norm": 0.21581244708702438,
+      "learning_rate": 1.3075559603339354e-05,
+      "loss": 0.1479,
+      "step": 254
+    },
+    {
+      "epoch": 3.248407643312102,
+      "grad_norm": 0.2075414978318476,
+      "learning_rate": 1.2907902259149287e-05,
+      "loss": 0.1289,
+      "step": 255
+    },
+    {
+      "epoch": 3.261146496815287,
+      "grad_norm": 0.24464034758512393,
+      "learning_rate": 1.274081305752135e-05,
+      "loss": 0.1505,
+      "step": 256
+    },
+    {
+      "epoch": 3.2738853503184715,
+      "grad_norm": 0.23054729756063358,
+      "learning_rate": 1.2574305383844528e-05,
+      "loss": 0.1534,
+      "step": 257
+    },
+    {
+      "epoch": 3.286624203821656,
+      "grad_norm": 0.2020043798081996,
+      "learning_rate": 1.2408392576922075e-05,
+      "loss": 0.1381,
+      "step": 258
+    },
+    {
+      "epoch": 3.299363057324841,
+      "grad_norm": 0.22799906894946498,
+      "learning_rate": 1.2243087927902905e-05,
+      "loss": 0.1719,
+      "step": 259
+    },
+    {
+      "epoch": 3.3121019108280256,
+      "grad_norm": 0.24533082757636485,
+      "learning_rate": 1.2078404679216864e-05,
+      "loss": 0.1609,
+      "step": 260
+    },
+    {
+      "epoch": 3.3248407643312103,
+      "grad_norm": 0.20714305709401984,
+      "learning_rate": 1.1914356023513904e-05,
+      "loss": 0.1179,
+      "step": 261
+    },
+    {
+      "epoch": 3.337579617834395,
+      "grad_norm": 0.21668069968026532,
+      "learning_rate": 1.1750955102607193e-05,
+      "loss": 0.1647,
+      "step": 262
+    },
+    {
+      "epoch": 3.3503184713375798,
+      "grad_norm": 0.2184496195596042,
+      "learning_rate": 1.1588215006420374e-05,
+      "loss": 0.1553,
+      "step": 263
+    },
+    {
+      "epoch": 3.3630573248407645,
+      "grad_norm": 0.20337547245152646,
+      "learning_rate": 1.1426148771938915e-05,
+      "loss": 0.1315,
+      "step": 264
+    },
+    {
+      "epoch": 3.375796178343949,
+      "grad_norm": 0.2037566957772752,
+      "learning_rate": 1.1264769382165748e-05,
+      "loss": 0.1539,
+      "step": 265
+    },
+    {
+      "epoch": 3.388535031847134,
+      "grad_norm": 0.20032474440683493,
+      "learning_rate": 1.110408976508118e-05,
+      "loss": 0.1215,
+      "step": 266
+    },
+    {
+      "epoch": 3.4012738853503186,
+      "grad_norm": 0.2136164552846967,
+      "learning_rate": 1.094412279260726e-05,
+      "loss": 0.1428,
+      "step": 267
+    },
+    {
+      "epoch": 3.4140127388535033,
+      "grad_norm": 0.21177053858403921,
+      "learning_rate": 1.0784881279576635e-05,
+      "loss": 0.1595,
+      "step": 268
+    },
+    {
+      "epoch": 3.426751592356688,
+      "grad_norm": 0.2029892437448442,
+      "learning_rate": 1.0626377982705929e-05,
+      "loss": 0.1383,
+      "step": 269
+    },
+    {
+      "epoch": 3.4394904458598727,
+      "grad_norm": 0.20919896892050535,
+      "learning_rate": 1.0468625599573842e-05,
+      "loss": 0.1424,
+      "step": 270
+    },
+    {
+      "epoch": 3.4522292993630574,
+      "grad_norm": 0.2070233782186885,
+      "learning_rate": 1.0311636767603952e-05,
+      "loss": 0.1467,
+      "step": 271
+    },
+    {
+      "epoch": 3.464968152866242,
+      "grad_norm": 0.19702950998811397,
+      "learning_rate": 1.0155424063052306e-05,
+      "loss": 0.1341,
+      "step": 272
+    },
+    {
+      "epoch": 3.477707006369427,
+      "grad_norm": 0.2063307034689398,
+      "learning_rate": 1.0000000000000006e-05,
+      "loss": 0.1333,
+      "step": 273
+    },
+    {
+      "epoch": 3.4904458598726116,
+      "grad_norm": 0.20719963685656465,
+      "learning_rate": 9.84537702935065e-06,
+      "loss": 0.1388,
+      "step": 274
+    },
+    {
+      "epoch": 3.5031847133757963,
+      "grad_norm": 0.215076053264252,
+      "learning_rate": 9.691567537832964e-06,
+      "loss": 0.1382,
+      "step": 275
+    },
+    {
+      "epoch": 3.515923566878981,
+      "grad_norm": 0.2088643606927074,
+      "learning_rate": 9.538583847008452e-06,
+      "loss": 0.138,
+      "step": 276
+    },
+    {
+      "epoch": 3.5286624203821657,
+      "grad_norm": 0.19464590100491783,
+      "learning_rate": 9.386438212284372e-06,
+      "loss": 0.1602,
+      "step": 277
+    },
+    {
+      "epoch": 3.5414012738853504,
+      "grad_norm": 0.19748474073166788,
+      "learning_rate": 9.235142821931928e-06,
+      "loss": 0.1205,
+      "step": 278
+    },
+    {
+      "epoch": 3.554140127388535,
+      "grad_norm": 0.20462175920146233,
+      "learning_rate": 9.084709796109907e-06,
+      "loss": 0.1243,
+      "step": 279
+    },
+    {
+      "epoch": 3.56687898089172,
+      "grad_norm": 0.19781750645119578,
+      "learning_rate": 8.93515118589373e-06,
+      "loss": 0.1184,
+      "step": 280
+    },
+    {
+      "epoch": 3.5796178343949046,
+      "grad_norm": 0.1932570190163277,
+      "learning_rate": 8.786478972310023e-06,
+      "loss": 0.1402,
+      "step": 281
+    },
+    {
+      "epoch": 3.5923566878980893,
+      "grad_norm": 0.2130457950220486,
+      "learning_rate": 8.638705065376887e-06,
+      "loss": 0.1572,
+      "step": 282
+    },
+    {
+      "epoch": 3.605095541401274,
+      "grad_norm": 0.21977855143398284,
+      "learning_rate": 8.491841303149728e-06,
+      "loss": 0.1666,
+      "step": 283
+    },
+    {
+      "epoch": 3.6178343949044587,
+      "grad_norm": 0.19809617527946968,
+      "learning_rate": 8.345899450772975e-06,
+      "loss": 0.1423,
+      "step": 284
+    },
+    {
+      "epoch": 3.6305732484076434,
+      "grad_norm": 0.18690090576844234,
+      "learning_rate": 8.200891199537549e-06,
+      "loss": 0.1364,
+      "step": 285
+    },
+    {
+      "epoch": 3.643312101910828,
+      "grad_norm": 0.18446815748564052,
+      "learning_rate": 8.056828165944282e-06,
+      "loss": 0.1276,
+      "step": 286
+    },
+    {
+      "epoch": 3.656050955414013,
+      "grad_norm": 0.19914110063591683,
+      "learning_rate": 7.913721890773354e-06,
+      "loss": 0.1452,
+      "step": 287
+    },
+    {
+      "epoch": 3.6687898089171975,
+      "grad_norm": 0.19784197594445632,
+      "learning_rate": 7.771583838159756e-06,
+      "loss": 0.1393,
+      "step": 288
+    },
+    {
+      "epoch": 3.6815286624203822,
+      "grad_norm": 0.18779371172725068,
+      "learning_rate": 7.630425394674903e-06,
+      "loss": 0.1486,
+      "step": 289
+    },
+    {
+      "epoch": 3.694267515923567,
+      "grad_norm": 0.19091202103101299,
+      "learning_rate": 7.49025786841445e-06,
+      "loss": 0.1532,
+      "step": 290
+    },
+    {
+      "epoch": 3.7070063694267517,
+      "grad_norm": 0.19572576915772363,
+      "learning_rate": 7.3510924880924575e-06,
+      "loss": 0.1498,
+      "step": 291
+    },
+    {
+      "epoch": 3.7197452229299364,
+      "grad_norm": 0.19383886127171826,
+      "learning_rate": 7.212940402141808e-06,
+      "loss": 0.1243,
+      "step": 292
+    },
+    {
+      "epoch": 3.732484076433121,
+      "grad_norm": 0.18951075996073852,
+      "learning_rate": 7.075812677821145e-06,
+      "loss": 0.1408,
+      "step": 293
+    },
+    {
+      "epoch": 3.745222929936306,
+      "grad_norm": 0.18725185654705204,
+      "learning_rate": 6.939720300328303e-06,
+      "loss": 0.1445,
+      "step": 294
+    },
+    {
+      "epoch": 3.7579617834394905,
+      "grad_norm": 0.1923854581483771,
+      "learning_rate": 6.8046741719202385e-06,
+      "loss": 0.1322,
+      "step": 295
+    },
+    {
+      "epoch": 3.770700636942675,
+      "grad_norm": 0.19857273274849133,
+      "learning_rate": 6.67068511103971e-06,
+      "loss": 0.1472,
+      "step": 296
+    },
+    {
+      "epoch": 3.78343949044586,
+      "grad_norm": 0.18958112392058588,
+      "learning_rate": 6.537763851448593e-06,
+      "loss": 0.1188,
+      "step": 297
+    },
+    {
+      "epoch": 3.7961783439490446,
+      "grad_norm": 0.1754562981627312,
+      "learning_rate": 6.4059210413680175e-06,
+      "loss": 0.1352,
+      "step": 298
+    },
+    {
+      "epoch": 3.8089171974522293,
+      "grad_norm": 0.1827948964704896,
+      "learning_rate": 6.275167242625331e-06,
+      "loss": 0.1343,
+      "step": 299
+    },
+    {
+      "epoch": 3.821656050955414,
+      "grad_norm": 0.17640473509943577,
+      "learning_rate": 6.145512929808013e-06,
+      "loss": 0.1463,
+      "step": 300
+    },
+    {
+      "epoch": 3.8343949044585988,
+      "grad_norm": 0.19352956970296584,
+      "learning_rate": 6.016968489424572e-06,
+      "loss": 0.136,
+      "step": 301
+    },
+    {
+      "epoch": 3.8471337579617835,
+      "grad_norm": 0.17765227219483207,
+      "learning_rate": 5.889544219072465e-06,
+      "loss": 0.1371,
+      "step": 302
+    },
+    {
+      "epoch": 3.859872611464968,
+      "grad_norm": 0.17378510781197484,
+      "learning_rate": 5.7632503266131925e-06,
+      "loss": 0.1508,
+      "step": 303
+    },
+    {
+      "epoch": 3.872611464968153,
+      "grad_norm": 0.1754866346893611,
+      "learning_rate": 5.638096929354522e-06,
+      "loss": 0.1371,
+      "step": 304
+    },
+    {
+      "epoch": 3.8853503184713376,
+      "grad_norm": 0.17990543007311735,
+      "learning_rate": 5.514094053240035e-06,
+      "loss": 0.1504,
+      "step": 305
+    },
+    {
+      "epoch": 3.8980891719745223,
+      "grad_norm": 0.1767314352665911,
+      "learning_rate": 5.39125163204594e-06,
+      "loss": 0.1384,
+      "step": 306
+    },
+    {
+      "epoch": 3.910828025477707,
+      "grad_norm": 0.18388289790416432,
+      "learning_rate": 5.269579506585259e-06,
+      "loss": 0.125,
+      "step": 307
+    },
+    {
+      "epoch": 3.9235668789808917,
+      "grad_norm": 0.18709650841448822,
+      "learning_rate": 5.149087423919541e-06,
+      "loss": 0.1719,
+      "step": 308
+    },
+    {
+      "epoch": 3.9363057324840764,
+      "grad_norm": 0.1765027020500491,
+      "learning_rate": 5.029785036577976e-06,
+      "loss": 0.1365,
+      "step": 309
+    },
+    {
+      "epoch": 3.949044585987261,
+      "grad_norm": 0.1740407642246799,
+      "learning_rate": 4.911681901784198e-06,
+      "loss": 0.1455,
+      "step": 310
+    },
+    {
+      "epoch": 3.961783439490446,
+      "grad_norm": 0.17955395869877427,
+      "learning_rate": 4.794787480690597e-06,
+      "loss": 0.135,
+      "step": 311
+    },
+    {
+      "epoch": 3.9745222929936306,
+      "grad_norm": 0.18507428221903918,
+      "learning_rate": 4.679111137620442e-06,
+      "loss": 0.1575,
+      "step": 312
+    },
+    {
+      "epoch": 3.9872611464968153,
+      "grad_norm": 0.1805033770468359,
+      "learning_rate": 4.5646621393177e-06,
+      "loss": 0.1388,
+      "step": 313
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.16048437681749295,
+      "learning_rate": 4.451449654204685e-06,
+      "loss": 0.1153,
+      "step": 314
+    },
+    {
+      "epoch": 4.012738853503185,
+      "grad_norm": 0.24959599645461245,
+      "learning_rate": 4.339482751647557e-06,
+      "loss": 0.1103,
+      "step": 315
+    },
+    {
+      "epoch": 4.025477707006369,
+      "grad_norm": 0.23271125680775137,
+      "learning_rate": 4.228770401229824e-06,
+      "loss": 0.1186,
+      "step": 316
+    },
+    {
+      "epoch": 4.038216560509555,
+      "grad_norm": 0.2122069247359907,
+      "learning_rate": 4.119321472033779e-06,
+      "loss": 0.1008,
+      "step": 317
+    },
+    {
+      "epoch": 4.050955414012739,
+      "grad_norm": 0.15556832545642946,
+      "learning_rate": 4.011144731929981e-06,
+      "loss": 0.0879,
+      "step": 318
+    },
+    {
+      "epoch": 4.063694267515924,
+      "grad_norm": 0.13068949666043236,
+      "learning_rate": 3.904248846874894e-06,
+      "loss": 0.101,
+      "step": 319
+    },
+    {
+      "epoch": 4.076433121019108,
+      "grad_norm": 0.15988306601505795,
+      "learning_rate": 3.7986423802166705e-06,
+      "loss": 0.1042,
+      "step": 320
+    },
+    {
+      "epoch": 4.089171974522293,
+      "grad_norm": 0.2008005962776904,
+      "learning_rate": 3.694333792009115e-06,
+      "loss": 0.1186,
+      "step": 321
+    },
+    {
+      "epoch": 4.101910828025478,
+      "grad_norm": 0.22309601017298253,
+      "learning_rate": 3.5913314383339937e-06,
+      "loss": 0.1079,
+      "step": 322
+    },
+    {
+      "epoch": 4.114649681528663,
+      "grad_norm": 0.2171458668604706,
+      "learning_rate": 3.4896435706316e-06,
+      "loss": 0.0992,
+      "step": 323
+    },
+    {
+      "epoch": 4.127388535031847,
+      "grad_norm": 0.1754223239205233,
+      "learning_rate": 3.3892783350397675e-06,
+      "loss": 0.0966,
+      "step": 324
+    },
+    {
+      "epoch": 4.140127388535032,
+      "grad_norm": 0.17536902204340468,
+      "learning_rate": 3.290243771741275e-06,
+      "loss": 0.1002,
+      "step": 325
+    },
+    {
+      "epoch": 4.1528662420382165,
+      "grad_norm": 0.1349302126131413,
+      "learning_rate": 3.1925478143197418e-06,
+      "loss": 0.0864,
+      "step": 326
+    },
+    {
+      "epoch": 4.165605095541402,
+      "grad_norm": 0.14115081640236687,
+      "learning_rate": 3.0961982891241083e-06,
+      "loss": 0.1042,
+      "step": 327
+    },
+    {
+      "epoch": 4.178343949044586,
+      "grad_norm": 0.14152556061179694,
+      "learning_rate": 3.001202914641628e-06,
+      "loss": 0.0832,
+      "step": 328
+    },
+    {
+      "epoch": 4.191082802547771,
+      "grad_norm": 0.14396584823928577,
+      "learning_rate": 2.907569300879596e-06,
+      "loss": 0.0863,
+      "step": 329
+    },
+    {
+      "epoch": 4.203821656050955,
+      "grad_norm": 0.13656039310896373,
+      "learning_rate": 2.815304948755664e-06,
+      "loss": 0.105,
+      "step": 330
+    },
+    {
+      "epoch": 4.2165605095541405,
+      "grad_norm": 0.14707618001361536,
+      "learning_rate": 2.7244172494969978e-06,
+      "loss": 0.1121,
+      "step": 331
+    },
+    {
+      "epoch": 4.229299363057325,
+      "grad_norm": 0.14952406400214932,
+      "learning_rate": 2.6349134840481294e-06,
+      "loss": 0.1039,
+      "step": 332
+    },
+    {
+      "epoch": 4.24203821656051,
+      "grad_norm": 0.45773988775828156,
+      "learning_rate": 2.546800822487714e-06,
+      "loss": 0.1006,
+      "step": 333
+    },
+    {
+      "epoch": 4.254777070063694,
+      "grad_norm": 0.14086482905242464,
+      "learning_rate": 2.4600863234541338e-06,
+      "loss": 0.103,
+      "step": 334
+    },
+    {
+      "epoch": 4.267515923566879,
+      "grad_norm": 0.13406472253081728,
+      "learning_rate": 2.374776933580025e-06,
+      "loss": 0.1102,
+      "step": 335
+    },
+    {
+      "epoch": 4.280254777070064,
+      "grad_norm": 0.14550219594207334,
+      "learning_rate": 2.2908794869358044e-06,
+      "loss": 0.1125,
+      "step": 336
+    },
+    {
+      "epoch": 4.292993630573249,
+      "grad_norm": 0.1347484077586462,
+      "learning_rate": 2.2084007044821764e-06,
+      "loss": 0.0835,
+      "step": 337
+    },
+    {
+      "epoch": 4.305732484076433,
+      "grad_norm": 0.1290805497722182,
+      "learning_rate": 2.127347193531757e-06,
+      "loss": 0.1041,
+      "step": 338
+    },
+    {
+      "epoch": 4.318471337579618,
+      "grad_norm": 0.13309082164534528,
+      "learning_rate": 2.0477254472197237e-06,
+      "loss": 0.0967,
+      "step": 339
+    },
+    {
+      "epoch": 4.3312101910828025,
+      "grad_norm": 0.13761067925193232,
+      "learning_rate": 1.96954184398368e-06,
+      "loss": 0.1113,
+      "step": 340
+    },
+    {
+      "epoch": 4.343949044585988,
+      "grad_norm": 0.13572266975580224,
+      "learning_rate": 1.8928026470526917e-06,
+      "loss": 0.1053,
+      "step": 341
+    },
+    {
+      "epoch": 4.356687898089172,
+      "grad_norm": 0.12800656352242193,
+      "learning_rate": 1.817514003945524e-06,
+      "loss": 0.0867,
+      "step": 342
+    },
+    {
+      "epoch": 4.369426751592357,
+      "grad_norm": 0.13496217012761974,
+      "learning_rate": 1.743681945978184e-06,
+      "loss": 0.11,
+      "step": 343
+    },
+    {
+      "epoch": 4.382165605095541,
+      "grad_norm": 0.13765332230964036,
+      "learning_rate": 1.6713123877807413e-06,
+      "loss": 0.1133,
+      "step": 344
+    },
+    {
+      "epoch": 4.3949044585987265,
+      "grad_norm": 0.13352458324587824,
+      "learning_rate": 1.6004111268235156e-06,
+      "loss": 0.1058,
+      "step": 345
+    },
+    {
+      "epoch": 4.407643312101911,
+      "grad_norm": 0.13191561231330995,
+      "learning_rate": 1.5309838429526714e-06,
+      "loss": 0.1051,
+      "step": 346
+    },
+    {
+      "epoch": 4.420382165605096,
+      "grad_norm": 0.12490688145220193,
+      "learning_rate": 1.4630360979351644e-06,
+      "loss": 0.09,
+      "step": 347
+    },
+    {
+      "epoch": 4.43312101910828,
+      "grad_norm": 0.15947738459761718,
+      "learning_rate": 1.396573335013236e-06,
+      "loss": 0.0964,
+      "step": 348
+    },
+    {
+      "epoch": 4.445859872611465,
+      "grad_norm": 0.12096509710405506,
+      "learning_rate": 1.3316008784683265e-06,
+      "loss": 0.0907,
+      "step": 349
+    },
+    {
+      "epoch": 4.45859872611465,
+      "grad_norm": 0.12188980792759932,
+      "learning_rate": 1.2681239331945695e-06,
+      "loss": 0.1016,
+      "step": 350
+    },
+    {
+      "epoch": 4.471337579617835,
+      "grad_norm": 0.13559724068658183,
+      "learning_rate": 1.2061475842818337e-06,
+      "loss": 0.1072,
+      "step": 351
+    },
+    {
+      "epoch": 4.484076433121019,
+      "grad_norm": 0.12292648232545614,
+      "learning_rate": 1.1456767966083393e-06,
+      "loss": 0.1087,
+      "step": 352
+    },
+    {
+      "epoch": 4.496815286624204,
+      "grad_norm": 0.1334907116310344,
+      "learning_rate": 1.086716414442952e-06,
+      "loss": 0.1161,
+      "step": 353
+    },
+    {
+      "epoch": 4.509554140127388,
+      "grad_norm": 0.13271118963324863,
+      "learning_rate": 1.0292711610570904e-06,
+      "loss": 0.0958,
+      "step": 354
+    },
+    {
+      "epoch": 4.522292993630574,
+      "grad_norm": 0.13680838872833267,
+      "learning_rate": 9.733456383463658e-07,
+      "loss": 0.0943,
+      "step": 355
+    },
+    {
+      "epoch": 4.535031847133758,
+      "grad_norm": 0.13801489368931033,
+      "learning_rate": 9.189443264619102e-07,
+      "loss": 0.123,
+      "step": 356
+    },
+    {
+      "epoch": 4.547770700636943,
+      "grad_norm": 0.12529844924959202,
+      "learning_rate": 8.660715834514977e-07,
+      "loss": 0.0975,
+      "step": 357
+    },
+    {
+      "epoch": 4.560509554140127,
+      "grad_norm": 0.12938877893813736,
+      "learning_rate": 8.147316449103959e-07,
+      "loss": 0.0918,
+      "step": 358
+    },
+    {
+      "epoch": 4.573248407643312,
+      "grad_norm": 0.12115317481092167,
+      "learning_rate": 7.649286236420806e-07,
+      "loss": 0.1135,
+      "step": 359
+    },
+    {
+      "epoch": 4.585987261146497,
+      "grad_norm": 0.11833791739810552,
+      "learning_rate": 7.166665093287539e-07,
+      "loss": 0.0854,
+      "step": 360
+    },
+    {
+      "epoch": 4.598726114649682,
+      "grad_norm": 0.12705837653951255,
+      "learning_rate": 6.69949168211721e-07,
+      "loss": 0.1109,
+      "step": 361
+    },
+    {
+      "epoch": 4.611464968152866,
+      "grad_norm": 0.12508614962340267,
+      "learning_rate": 6.247803427816945e-07,
+      "loss": 0.0974,
+      "step": 362
+    },
+    {
+      "epoch": 4.624203821656051,
+      "grad_norm": 0.1452236579744733,
+      "learning_rate": 5.811636514789598e-07,
+      "loss": 0.1157,
+      "step": 363
+    },
+    {
+      "epoch": 4.6369426751592355,
+      "grad_norm": 0.1300660477914448,
+      "learning_rate": 5.391025884035239e-07,
+      "loss": 0.0915,
+      "step": 364
+    },
+    {
+      "epoch": 4.649681528662421,
+      "grad_norm": 0.1200540678678256,
+      "learning_rate": 4.986005230351954e-07,
+      "loss": 0.0872,
+      "step": 365
+    },
+    {
+      "epoch": 4.662420382165605,
+      "grad_norm": 0.11487997298035525,
+      "learning_rate": 4.5966069996365993e-07,
+      "loss": 0.1165,
+      "step": 366
+    },
+    {
+      "epoch": 4.67515923566879,
+      "grad_norm": 0.139543063544079,
+      "learning_rate": 4.22286238628562e-07,
+      "loss": 0.1053,
+      "step": 367
+    },
+    {
+      "epoch": 4.687898089171974,
+      "grad_norm": 0.12478098586523038,
+      "learning_rate": 3.8648013306960664e-07,
+      "loss": 0.0858,
+      "step": 368
+    },
+    {
+      "epoch": 4.7006369426751595,
+      "grad_norm": 0.1314853226678359,
+      "learning_rate": 3.522452516867048e-07,
+      "loss": 0.1179,
+      "step": 369
+    },
+    {
+      "epoch": 4.713375796178344,
+      "grad_norm": 0.1561600576660889,
+      "learning_rate": 3.1958433701019697e-07,
+      "loss": 0.0977,
+      "step": 370
+    },
+    {
+      "epoch": 4.726114649681529,
+      "grad_norm": 0.1275347994190016,
+      "learning_rate": 2.8850000548115155e-07,
+      "loss": 0.112,
+      "step": 371
+    },
+    {
+      "epoch": 4.738853503184713,
+      "grad_norm": 0.12188252873510248,
+      "learning_rate": 2.5899474724174313e-07,
+      "loss": 0.0862,
+      "step": 372
+    },
+    {
+      "epoch": 4.751592356687898,
+      "grad_norm": 0.1318771763835626,
+      "learning_rate": 2.3107092593579905e-07,
+      "loss": 0.1142,
+      "step": 373
+    },
+    {
+      "epoch": 4.764331210191083,
+      "grad_norm": 0.14220929835983984,
+      "learning_rate": 2.0473077851942858e-07,
+      "loss": 0.1118,
+      "step": 374
+    },
+    {
+      "epoch": 4.777070063694268,
+      "grad_norm": 0.1219858566908112,
+      "learning_rate": 1.799764150818306e-07,
+      "loss": 0.1084,
+      "step": 375
+    },
+    {
+      "epoch": 4.789808917197452,
+      "grad_norm": 0.12566174956929266,
+      "learning_rate": 1.5680981867625566e-07,
+      "loss": 0.1077,
+      "step": 376
+    },
+    {
+      "epoch": 4.802547770700637,
+      "grad_norm": 0.12253664048339573,
+      "learning_rate": 1.3523284516113955e-07,
+      "loss": 0.0899,
+      "step": 377
+    },
+    {
+      "epoch": 4.8152866242038215,
+      "grad_norm": 0.12142275297990943,
+      "learning_rate": 1.1524722305144231e-07,
+      "loss": 0.0956,
+      "step": 378
+    },
+    {
+      "epoch": 4.828025477707007,
+      "grad_norm": 0.1235160540744103,
+      "learning_rate": 9.685455338016347e-08,
+      "loss": 0.1077,
+      "step": 379
+    },
+    {
+      "epoch": 4.840764331210191,
+      "grad_norm": 0.12467941962401048,
+      "learning_rate": 8.005630957010014e-08,
+      "loss": 0.0793,
+      "step": 380
+    },
+    {
+      "epoch": 4.853503184713376,
+      "grad_norm": 0.20795298981735666,
+      "learning_rate": 6.485383731580142e-08,
+      "loss": 0.1045,
+      "step": 381
+    },
+    {
+      "epoch": 4.86624203821656,
+      "grad_norm": 0.12799721551039098,
+      "learning_rate": 5.1248354475768034e-08,
+      "loss": 0.101,
+      "step": 382
+    },
+    {
+      "epoch": 4.8789808917197455,
+      "grad_norm": 0.1258631802102174,
+      "learning_rate": 3.924095097489922e-08,
+      "loss": 0.0944,
+      "step": 383
+    },
+    {
+      "epoch": 4.89171974522293,
+      "grad_norm": 0.12812579528294077,
+      "learning_rate": 2.8832588717164766e-08,
+      "loss": 0.1053,
+      "step": 384
+    },
+    {
+      "epoch": 4.904458598726115,
+      "grad_norm": 0.12751773731121074,
+      "learning_rate": 2.0024101508555604e-08,
+      "loss": 0.0982,
+      "step": 385
+    },
+    {
+      "epoch": 4.917197452229299,
+      "grad_norm": 0.13129186435975,
+      "learning_rate": 1.281619499029274e-08,
+      "loss": 0.1091,
+      "step": 386
+    },
+    {
+      "epoch": 4.929936305732484,
+      "grad_norm": 0.12192936592060051,
+      "learning_rate": 7.209446582292501e-09,
+      "loss": 0.1065,
+      "step": 387
+    },
+    {
+      "epoch": 4.942675159235669,
+      "grad_norm": 0.13025711875394239,
+      "learning_rate": 3.2043054369057523e-09,
+      "loss": 0.1076,
+      "step": 388
+    },
+    {
+      "epoch": 4.955414012738854,
+      "grad_norm": 0.11678098849828318,
+      "learning_rate": 8.010924029533406e-10,
+      "loss": 0.1073,
+      "step": 389
+    },
+    {
+      "epoch": 4.968152866242038,
+      "grad_norm": 0.12263643856967697,
+      "learning_rate": 0.0,
+      "loss": 0.1206,
+      "step": 390
+    },
+    {
+      "epoch": 4.968152866242038,
+      "step": 390,
+      "total_flos": 4.6317288847048704e+17,
+      "train_loss": 0.30433785139750213,
+      "train_runtime": 26201.4476,
+      "train_samples_per_second": 1.908,
+      "train_steps_per_second": 0.015
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 390,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "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": 4.6317288847048704e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c54959564ce89463eee89264102be47c5db47e35f4fdf3151a0e247e22037b63
-size 7224
+oid sha256:94da09efce32ded52b07167a713c7e2d5d21f294e8b090ddb07077d42205eac3
+size 7352