Upload folder using huggingface_hub

global_step428/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31eb677de80948991321bfe446bcdaf4165d3f8c9efa96cfadccfaf3a07e6432
+size 5435499827

global_step428/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09e4495311047938552b63e0739ae4a551323c70509aed4b806aae92544537a7
+size 5435499827

global_step428/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e1e42c3abfc5e8f3bf1af140fef44ff8f8441a9256840b045d424c01026ce1e
+size 5435499827

global_step428/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:398bfa2896865cf30cef24ec6b68f47976d7f16cb3df673efceb7804d4b93241
+size 5435499827

global_step428/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a62bcb45fde6ec6327735c7a280ac3ce1f417f8924c8774a8c82a17cfbf4c295
+size 5435499827

global_step428/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb1d3d30b3e766144ee852f023bd74030024ab52c9f32771cdef96fb92ea360a
+size 5435499827

global_step428/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:874cb020cb85f973bcb6aabf19b131ea5f4119dfb72f80578d61c90b95b95b8f
+size 5435499827

global_step428/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68f1d3311e2ce3605022c06ddc3064dd9e31f60d59b8cef8e11e6d4ffb6c19c7
+size 5435499827

global_step428/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57e8c2eb8c8299c2c958127742fbf228e53106a6568c4971b8afabbd0109ff90
+size 153829

global_step428/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e282c8dfa4e2bc4ca41f74c0855b60c0b95c8b57349ed0f767b0ac93c6d90e4
+size 153829

global_step428/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a05bcb052dcc69a98918a82611f0e1e96657c85e106f480eda0074811399d0ca
+size 153829

global_step428/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55815a63565a06e023966b6a11030d06cfdc86e64fac8c811f57b00ab0e1e847
+size 153829

global_step428/zero_pp_rank_4_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43ea8ade0f653c47a1ac50680612d8e5218345ab1b86cb35384d76d563dd75cc
+size 153829

global_step428/zero_pp_rank_5_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9650144687908dce67d9631417bd0ee2e129d6fa67ed87825da1ef4167112639
+size 153829

global_step428/zero_pp_rank_6_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fb9e75a82b054c216c566b9165d02270f6da8be51ccc5c3ec334b0c2004d830
+size 153829

global_step428/zero_pp_rank_7_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47f5259391522659ec37fb679a3619fb26f04f6a1908753f5d9f85e2dda46e3f
+size 153829

latest

@@ -0,0 +1 @@
+global_step428

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1bec598899f9d59e70c1b4705ce420a1e0a670957b6c8153a589880068ae5a4
+size 15984

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c60d2348aae518f4c44693db9c9b4b3a3299c556e7f0a86c188b2e4c3e364a7c
+size 15984

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffe5a79d3bcb4ce033de360bc765e616316e3562aba25887cd85c4adbb935abf
+size 15984

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9a9d1f6e22677721841890e6a27855857e6840137650d609eb8e4ac13b71d29
+size 15984

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcac4ff84388a6a4fe3bcae6207c68b2ee5528fb3b6de8cc3588fe1975462aa5
+size 15984

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33fce3cdf5c1b8a8a291e0c73b384e3ad5252640e21e942b44b26b8b0928ffa9
+size 15984

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:919e675f3bcaf4f3c8ba35cd8debf85aec3bbc3c8e5019b74431e0a314e4d37a
+size 15984

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bf6479ce82b88efc6a72a8ee512162b3d0ecab972817296d38ab9c448bb8d96
+size 15984

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee8c731be5df723d260a08657a32a14cf7657c00593a1a4bb80bc7c00b297026
+size 1064

trainer_state.json

@@ -0,0 +1,2629 @@
+{
+  "best_metric": 0.5059249997138977,
+  "best_model_checkpoint": "./Einstein-v3-model/checkpoint-428",
+  "epoch": 0.9994162288382954,
+  "eval_steps": 107,
+  "global_step": 428,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 5.000000000000001e-07,
+      "loss": 1.038,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "eval_loss": 1.125040888786316,
+      "eval_runtime": 969.9573,
+      "eval_samples_per_second": 1.161,
+      "eval_steps_per_second": 0.145,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 1.0692,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.5e-06,
+      "loss": 1.0161,
+      "step": 3
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.9363,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.5e-06,
+      "loss": 0.8647,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3e-06,
+      "loss": 0.9888,
+      "step": 6
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.5e-06,
+      "loss": 0.8645,
+      "step": 7
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.763,
+      "step": 8
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.5e-06,
+      "loss": 0.8058,
+      "step": 9
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5e-06,
+      "loss": 0.789,
+      "step": 10
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.9999293917983325e-06,
+      "loss": 0.798,
+      "step": 11
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.999717571181742e-06,
+      "loss": 0.7411,
+      "step": 12
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.9993645501152485e-06,
+      "loss": 0.748,
+      "step": 13
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.998870348539797e-06,
+      "loss": 0.7312,
+      "step": 14
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.998234994371135e-06,
+      "loss": 0.7179,
+      "step": 15
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.997458523498236e-06,
+      "loss": 0.7158,
+      "step": 16
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.996540979781269e-06,
+      "loss": 0.6975,
+      "step": 17
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.995482415049123e-06,
+      "loss": 0.6809,
+      "step": 18
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.99428288909648e-06,
+      "loss": 0.6709,
+      "step": 19
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.992942469680437e-06,
+      "loss": 0.6778,
+      "step": 20
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.991461232516675e-06,
+      "loss": 0.69,
+      "step": 21
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.989839261275191e-06,
+      "loss": 0.6518,
+      "step": 22
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.988076647575562e-06,
+      "loss": 0.6712,
+      "step": 23
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.986173490981773e-06,
+      "loss": 0.6429,
+      "step": 24
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.984129898996599e-06,
+      "loss": 0.682,
+      "step": 25
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.981945987055521e-06,
+      "loss": 0.6661,
+      "step": 26
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.979621878520217e-06,
+      "loss": 0.6287,
+      "step": 27
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.977157704671585e-06,
+      "loss": 0.6642,
+      "step": 28
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.974553604702332e-06,
+      "loss": 0.6644,
+      "step": 29
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.971809725709112e-06,
+      "loss": 0.6453,
+      "step": 30
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.968926222684213e-06,
+      "loss": 0.6199,
+      "step": 31
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.965903258506806e-06,
+      "loss": 0.6038,
+      "step": 32
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.9627410039337426e-06,
+      "loss": 0.6493,
+      "step": 33
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.959439637589909e-06,
+      "loss": 0.6304,
+      "step": 34
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.9559993459581375e-06,
+      "loss": 0.6692,
+      "step": 35
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.952420323368673e-06,
+      "loss": 0.5978,
+      "step": 36
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.948702771988195e-06,
+      "loss": 0.6318,
+      "step": 37
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.944846901808397e-06,
+      "loss": 0.5838,
+      "step": 38
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.940852930634126e-06,
+      "loss": 0.5884,
+      "step": 39
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.936721084071079e-06,
+      "loss": 0.5955,
+      "step": 40
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.932451595513063e-06,
+      "loss": 0.5704,
+      "step": 41
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.928044706128803e-06,
+      "loss": 0.5711,
+      "step": 42
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.923500664848327e-06,
+      "loss": 0.5886,
+      "step": 43
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.918819728348901e-06,
+      "loss": 0.5802,
+      "step": 44
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.9140021610405335e-06,
+      "loss": 0.6221,
+      "step": 45
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.909048235051033e-06,
+      "loss": 0.5859,
+      "step": 46
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.903958230210647e-06,
+      "loss": 0.5906,
+      "step": 47
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.8987324340362445e-06,
+      "loss": 0.586,
+      "step": 48
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.89337114171508e-06,
+      "loss": 0.5782,
+      "step": 49
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.887874656088124e-06,
+      "loss": 0.6341,
+      "step": 50
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.882243287632947e-06,
+      "loss": 0.5753,
+      "step": 51
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.8764773544461895e-06,
+      "loss": 0.5896,
+      "step": 52
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.8705771822255895e-06,
+      "loss": 0.567,
+      "step": 53
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.864543104251587e-06,
+      "loss": 0.5943,
+      "step": 54
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.858375461368499e-06,
+      "loss": 0.5942,
+      "step": 55
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.852074601965261e-06,
+      "loss": 0.5639,
+      "step": 56
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.845640881955757e-06,
+      "loss": 0.6058,
+      "step": 57
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.839074664758705e-06,
+      "loss": 0.5695,
+      "step": 58
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.832376321277136e-06,
+      "loss": 0.6233,
+      "step": 59
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.825546229877439e-06,
+      "loss": 0.5617,
+      "step": 60
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.818584776367992e-06,
+      "loss": 0.586,
+      "step": 61
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.811492353977366e-06,
+      "loss": 0.5609,
+      "step": 62
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.804269363332112e-06,
+      "loss": 0.5371,
+      "step": 63
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.7969162124341354e-06,
+      "loss": 0.5585,
+      "step": 64
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.789433316637644e-06,
+      "loss": 0.5548,
+      "step": 65
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.781821098625691e-06,
+      "loss": 0.5613,
+      "step": 66
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.7740799883862966e-06,
+      "loss": 0.5483,
+      "step": 67
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.766210423188158e-06,
+      "loss": 0.5892,
+      "step": 68
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.758212847555953e-06,
+      "loss": 0.5813,
+      "step": 69
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.750087713245227e-06,
+      "loss": 0.552,
+      "step": 70
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.74183547921688e-06,
+      "loss": 0.5478,
+      "step": 71
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.733456611611233e-06,
+      "loss": 0.5648,
+      "step": 72
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.7249515837217075e-06,
+      "loss": 0.5717,
+      "step": 73
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.716320875968081e-06,
+      "loss": 0.5916,
+      "step": 74
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.707564975869357e-06,
+      "loss": 0.5562,
+      "step": 75
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.698684378016223e-06,
+      "loss": 0.5347,
+      "step": 76
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.6896795840431155e-06,
+      "loss": 0.5595,
+      "step": 77
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.680551102599881e-06,
+      "loss": 0.564,
+      "step": 78
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.671299449323045e-06,
+      "loss": 0.5646,
+      "step": 79
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.66192514680669e-06,
+      "loss": 0.5667,
+      "step": 80
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.652428724572929e-06,
+      "loss": 0.5726,
+      "step": 81
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.642810719041999e-06,
+      "loss": 0.544,
+      "step": 82
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.63307167350196e-06,
+      "loss": 0.5673,
+      "step": 83
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.623212138078004e-06,
+      "loss": 0.5743,
+      "step": 84
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.613232669701384e-06,
+      "loss": 0.5605,
+      "step": 85
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.603133832077953e-06,
+      "loss": 0.5754,
+      "step": 86
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.592916195656322e-06,
+      "loss": 0.5578,
+      "step": 87
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.582580337595636e-06,
+      "loss": 0.5261,
+      "step": 88
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.572126841732977e-06,
+      "loss": 0.5864,
+      "step": 89
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.561556298550379e-06,
+      "loss": 0.5388,
+      "step": 90
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.550869305141478e-06,
+      "loss": 0.5662,
+      "step": 91
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.5400664651777835e-06,
+      "loss": 0.5359,
+      "step": 92
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.529148388874577e-06,
+      "loss": 0.5485,
+      "step": 93
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.518115692956445e-06,
+      "loss": 0.5715,
+      "step": 94
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.506969000622443e-06,
+      "loss": 0.5412,
+      "step": 95
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.49570894151089e-06,
+      "loss": 0.5475,
+      "step": 96
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.484336151663807e-06,
+      "loss": 0.5443,
+      "step": 97
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.472851273490985e-06,
+      "loss": 0.5519,
+      "step": 98
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.4612549557336975e-06,
+      "loss": 0.5293,
+      "step": 99
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.449547853428061e-06,
+      "loss": 0.5545,
+      "step": 100
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.437730627868028e-06,
+      "loss": 0.5361,
+      "step": 101
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.425803946568033e-06,
+      "loss": 0.547,
+      "step": 102
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.413768483225292e-06,
+      "loss": 0.532,
+      "step": 103
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.401624917681743e-06,
+      "loss": 0.5515,
+      "step": 104
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.3893739358856465e-06,
+      "loss": 0.5424,
+      "step": 105
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.377016229852836e-06,
+      "loss": 0.5217,
+      "step": 106
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.364552497627632e-06,
+      "loss": 0.5254,
+      "step": 107
+    },
+    {
+      "epoch": 0.25,
+      "eval_loss": 0.5753679871559143,
+      "eval_runtime": 972.8304,
+      "eval_samples_per_second": 1.157,
+      "eval_steps_per_second": 0.145,
+      "step": 107
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.3519834432434095e-06,
+      "loss": 0.5599,
+      "step": 108
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.33930977668283e-06,
+      "loss": 0.5479,
+      "step": 109
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.326532213837735e-06,
+      "loss": 0.582,
+      "step": 110
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.3136514764687155e-06,
+      "loss": 0.5545,
+      "step": 111
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.300668292164329e-06,
+      "loss": 0.5487,
+      "step": 112
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.287583394300016e-06,
+      "loss": 0.5644,
+      "step": 113
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.274397521996658e-06,
+      "loss": 0.5358,
+      "step": 114
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.261111420078844e-06,
+      "loss": 0.5425,
+      "step": 115
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.247725839032781e-06,
+      "loss": 0.5211,
+      "step": 116
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.234241534963916e-06,
+      "loss": 0.5663,
+      "step": 117
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.220659269554217e-06,
+      "loss": 0.5382,
+      "step": 118
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.206979810019153e-06,
+      "loss": 0.554,
+      "step": 119
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.1932039290643534e-06,
+      "loss": 0.528,
+      "step": 120
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.179332404841963e-06,
+      "loss": 0.5146,
+      "step": 121
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.1653660209066835e-06,
+      "loss": 0.5608,
+      "step": 122
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.151305566171521e-06,
+      "loss": 0.5573,
+      "step": 123
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.137151834863213e-06,
+      "loss": 0.5345,
+      "step": 124
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.122905626477371e-06,
+      "loss": 0.5434,
+      "step": 125
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.108567745733318e-06,
+      "loss": 0.5685,
+      "step": 126
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.094139002528635e-06,
+      "loss": 0.5442,
+      "step": 127
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.07962021189341e-06,
+      "loss": 0.4957,
+      "step": 128
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.065012193944201e-06,
+      "loss": 0.5833,
+      "step": 129
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.050315773837708e-06,
+      "loss": 0.5178,
+      "step": 130
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.0355317817241705e-06,
+      "loss": 0.553,
+      "step": 131
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.020661052700462e-06,
+      "loss": 0.5466,
+      "step": 132
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.00570442676293e-06,
+      "loss": 0.5412,
+      "step": 133
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 3.990662748759946e-06,
+      "loss": 0.5573,
+      "step": 134
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.975536868344174e-06,
+      "loss": 0.5593,
+      "step": 135
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.9603276399245864e-06,
+      "loss": 0.5251,
+      "step": 136
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.945035922618198e-06,
+      "loss": 0.5341,
+      "step": 137
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.929662580201536e-06,
+      "loss": 0.5267,
+      "step": 138
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.91420848106185e-06,
+      "loss": 0.5161,
+      "step": 139
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 3.898674498148058e-06,
+      "loss": 0.5403,
+      "step": 140
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 3.883061508921439e-06,
+      "loss": 0.536,
+      "step": 141
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 3.8673703953060685e-06,
+      "loss": 0.5374,
+      "step": 142
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 3.8516020436389945e-06,
+      "loss": 0.4947,
+      "step": 143
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 3.835757344620183e-06,
+      "loss": 0.5328,
+      "step": 144
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 3.819837193262197e-06,
+      "loss": 0.5306,
+      "step": 145
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 3.803842488839642e-06,
+      "loss": 0.5334,
+      "step": 146
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 3.7877741348383703e-06,
+      "loss": 0.5238,
+      "step": 147
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.7716330389044463e-06,
+      "loss": 0.5396,
+      "step": 148
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.7554201127928747e-06,
+      "loss": 0.5121,
+      "step": 149
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.739136272316102e-06,
+      "loss": 0.5662,
+      "step": 150
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.72278243729228e-06,
+      "loss": 0.4901,
+      "step": 151
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 3.706359531493316e-06,
+      "loss": 0.4966,
+      "step": 152
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 3.6898684825926845e-06,
+      "loss": 0.5597,
+      "step": 153
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 3.6733102221130303e-06,
+      "loss": 0.5164,
+      "step": 154
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 3.656685685373552e-06,
+      "loss": 0.5555,
+      "step": 155
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 3.6399958114371597e-06,
+      "loss": 0.5309,
+      "step": 156
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 3.623241543057445e-06,
+      "loss": 0.5468,
+      "step": 157
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 3.606423826625414e-06,
+      "loss": 0.5183,
+      "step": 158
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 3.5895436121160388e-06,
+      "loss": 0.5361,
+      "step": 159
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 3.5726018530345913e-06,
+      "loss": 0.5343,
+      "step": 160
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.5555995063627842e-06,
+      "loss": 0.5548,
+      "step": 161
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.5385375325047167e-06,
+      "loss": 0.5728,
+      "step": 162
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.5214168952326205e-06,
+      "loss": 0.5234,
+      "step": 163
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.5042385616324243e-06,
+      "loss": 0.5047,
+      "step": 164
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 3.4870035020491216e-06,
+      "loss": 0.5179,
+      "step": 165
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 3.469712690031962e-06,
+      "loss": 0.5013,
+      "step": 166
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 3.4523671022794612e-06,
+      "loss": 0.5113,
+      "step": 167
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 3.4349677185842246e-06,
+      "loss": 0.5694,
+      "step": 168
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 3.4175155217776057e-06,
+      "loss": 0.4965,
+      "step": 169
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 3.4000114976741905e-06,
+      "loss": 0.5475,
+      "step": 170
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 3.38245663501611e-06,
+      "loss": 0.5014,
+      "step": 171
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 3.3648519254171906e-06,
+      "loss": 0.5079,
+      "step": 172
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 3.3471983633069414e-06,
+      "loss": 0.5222,
+      "step": 173
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 3.32949694587438e-06,
+      "loss": 0.5039,
+      "step": 174
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 3.3117486730117092e-06,
+      "loss": 0.5067,
+      "step": 175
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 3.2939545472578314e-06,
+      "loss": 0.5575,
+      "step": 176
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 3.276115573741724e-06,
+      "loss": 0.5155,
+      "step": 177
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 3.2582327601256567e-06,
+      "loss": 0.4915,
+      "step": 178
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 3.240307116548279e-06,
+      "loss": 0.5415,
+      "step": 179
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 3.222339655567556e-06,
+      "loss": 0.5326,
+      "step": 180
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 3.2043313921035747e-06,
+      "loss": 0.5921,
+      "step": 181
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 3.1862833433812137e-06,
+      "loss": 0.5318,
+      "step": 182
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 3.1681965288726825e-06,
+      "loss": 0.5151,
+      "step": 183
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 3.1500719702399406e-06,
+      "loss": 0.5281,
+      "step": 184
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 3.1319106912769797e-06,
+      "loss": 0.5046,
+      "step": 185
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 3.1137137178519983e-06,
+      "loss": 0.5022,
+      "step": 186
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 3.0954820778494516e-06,
+      "loss": 0.5835,
+      "step": 187
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 3.0772168011119894e-06,
+      "loss": 0.514,
+      "step": 188
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 3.0589189193822894e-06,
+      "loss": 0.5291,
+      "step": 189
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 3.0405894662447682e-06,
+      "loss": 0.5186,
+      "step": 190
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 3.0222294770672054e-06,
+      "loss": 0.5483,
+      "step": 191
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 3.0038399889422553e-06,
+      "loss": 0.5561,
+      "step": 192
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.985422040628867e-06,
+      "loss": 0.5307,
+      "step": 193
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.9669766724936074e-06,
+      "loss": 0.5137,
+      "step": 194
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.948504926451896e-06,
+      "loss": 0.5222,
+      "step": 195
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.930007845909146e-06,
+      "loss": 0.5254,
+      "step": 196
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.911486475701835e-06,
+      "loss": 0.5369,
+      "step": 197
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.892941862038475e-06,
+      "loss": 0.5371,
+      "step": 198
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.8743750524405254e-06,
+      "loss": 0.5285,
+      "step": 199
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.8557870956832135e-06,
+      "loss": 0.5319,
+      "step": 200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.837179041736299e-06,
+      "loss": 0.4983,
+      "step": 201
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.8185519417047624e-06,
+      "loss": 0.4962,
+      "step": 202
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.799906847769433e-06,
+      "loss": 0.5055,
+      "step": 203
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.781244813127552e-06,
+      "loss": 0.4918,
+      "step": 204
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.762566891933285e-06,
+      "loss": 0.5191,
+      "step": 205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.743874139238171e-06,
+      "loss": 0.5509,
+      "step": 206
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.725167610931534e-06,
+      "loss": 0.5296,
+      "step": 207
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.7064483636808314e-06,
+      "loss": 0.5335,
+      "step": 208
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.687717454871971e-06,
+      "loss": 0.4982,
+      "step": 209
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.6689759425495833e-06,
+      "loss": 0.4864,
+      "step": 210
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.650224885357251e-06,
+      "loss": 0.5334,
+      "step": 211
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.6314653424777194e-06,
+      "loss": 0.5309,
+      "step": 212
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.612698373573056e-06,
+      "loss": 0.5346,
+      "step": 213
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.593925038724802e-06,
+      "loss": 0.5144,
+      "step": 214
+    },
+    {
+      "epoch": 0.5,
+      "eval_loss": 0.5360019207000732,
+      "eval_runtime": 973.5712,
+      "eval_samples_per_second": 1.157,
+      "eval_steps_per_second": 0.145,
+      "step": 214
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.575146398374087e-06,
+      "loss": 0.5263,
+      "step": 215
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.5563635132617305e-06,
+      "loss": 0.5135,
+      "step": 216
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.5375774443683263e-06,
+      "loss": 0.5003,
+      "step": 217
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.518789252854305e-06,
+      "loss": 0.5005,
+      "step": 218
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.5e-06,
+      "loss": 0.5006,
+      "step": 219
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.4812107471456958e-06,
+      "loss": 0.5052,
+      "step": 220
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.4624225556316745e-06,
+      "loss": 0.4966,
+      "step": 221
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.44363648673827e-06,
+      "loss": 0.4997,
+      "step": 222
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.4248536016259137e-06,
+      "loss": 0.535,
+      "step": 223
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.4060749612751987e-06,
+      "loss": 0.5156,
+      "step": 224
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.3873016264269446e-06,
+      "loss": 0.5505,
+      "step": 225
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.368534657522281e-06,
+      "loss": 0.4973,
+      "step": 226
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.3497751146427494e-06,
+      "loss": 0.5186,
+      "step": 227
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.3310240574504184e-06,
+      "loss": 0.5199,
+      "step": 228
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.3122825451280294e-06,
+      "loss": 0.546,
+      "step": 229
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.2935516363191695e-06,
+      "loss": 0.4939,
+      "step": 230
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.2748323890684664e-06,
+      "loss": 0.5012,
+      "step": 231
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.2561258607618296e-06,
+      "loss": 0.4843,
+      "step": 232
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.2374331080667168e-06,
+      "loss": 0.5267,
+      "step": 233
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.2187551868724487e-06,
+      "loss": 0.4927,
+      "step": 234
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.200093152230568e-06,
+      "loss": 0.4968,
+      "step": 235
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.1814480582952376e-06,
+      "loss": 0.4787,
+      "step": 236
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.1628209582637024e-06,
+      "loss": 0.4645,
+      "step": 237
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 2.1442129043167877e-06,
+      "loss": 0.5348,
+      "step": 238
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 2.125624947559475e-06,
+      "loss": 0.5095,
+      "step": 239
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 2.1070581379615253e-06,
+      "loss": 0.5282,
+      "step": 240
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 2.088513524298165e-06,
+      "loss": 0.5046,
+      "step": 241
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.0699921540908542e-06,
+      "loss": 0.5325,
+      "step": 242
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.0514950735481053e-06,
+      "loss": 0.5104,
+      "step": 243
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.033023327506393e-06,
+      "loss": 0.4881,
+      "step": 244
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.014577959371134e-06,
+      "loss": 0.4938,
+      "step": 245
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.996160011057746e-06,
+      "loss": 0.4977,
+      "step": 246
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.9777705229327954e-06,
+      "loss": 0.4936,
+      "step": 247
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.959410533755232e-06,
+      "loss": 0.5176,
+      "step": 248
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.9410810806177105e-06,
+      "loss": 0.5057,
+      "step": 249
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.922783198888011e-06,
+      "loss": 0.4982,
+      "step": 250
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.9045179221505497e-06,
+      "loss": 0.5128,
+      "step": 251
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8862862821480023e-06,
+      "loss": 0.5186,
+      "step": 252
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8680893087230207e-06,
+      "loss": 0.5112,
+      "step": 253
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8499280297600594e-06,
+      "loss": 0.4928,
+      "step": 254
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.8318034711273181e-06,
+      "loss": 0.5209,
+      "step": 255
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.813716656618788e-06,
+      "loss": 0.5073,
+      "step": 256
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.7956686078964257e-06,
+      "loss": 0.4724,
+      "step": 257
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.7776603444324445e-06,
+      "loss": 0.51,
+      "step": 258
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.759692883451721e-06,
+      "loss": 0.5193,
+      "step": 259
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.741767239874344e-06,
+      "loss": 0.519,
+      "step": 260
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.723884426258277e-06,
+      "loss": 0.5073,
+      "step": 261
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.7060454527421688e-06,
+      "loss": 0.5178,
+      "step": 262
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.6882513269882916e-06,
+      "loss": 0.5006,
+      "step": 263
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.6705030541256211e-06,
+      "loss": 0.4955,
+      "step": 264
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.6528016366930594e-06,
+      "loss": 0.5231,
+      "step": 265
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.6351480745828098e-06,
+      "loss": 0.4647,
+      "step": 266
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.6175433649838901e-06,
+      "loss": 0.493,
+      "step": 267
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.5999885023258099e-06,
+      "loss": 0.4896,
+      "step": 268
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.5824844782223956e-06,
+      "loss": 0.516,
+      "step": 269
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.5650322814157764e-06,
+      "loss": 0.5059,
+      "step": 270
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.5476328977205396e-06,
+      "loss": 0.5017,
+      "step": 271
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.5302873099680378e-06,
+      "loss": 0.4947,
+      "step": 272
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.5129964979508792e-06,
+      "loss": 0.4764,
+      "step": 273
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.495761438367577e-06,
+      "loss": 0.5008,
+      "step": 274
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.47858310476738e-06,
+      "loss": 0.5402,
+      "step": 275
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4614624674952843e-06,
+      "loss": 0.4882,
+      "step": 276
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4444004936372166e-06,
+      "loss": 0.4946,
+      "step": 277
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4273981469654093e-06,
+      "loss": 0.5305,
+      "step": 278
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4104563878839623e-06,
+      "loss": 0.4951,
+      "step": 279
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3935761733745865e-06,
+      "loss": 0.5188,
+      "step": 280
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3767584569425562e-06,
+      "loss": 0.4943,
+      "step": 281
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.360004188562841e-06,
+      "loss": 0.485,
+      "step": 282
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3433143146264494e-06,
+      "loss": 0.5002,
+      "step": 283
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3266897778869704e-06,
+      "loss": 0.5005,
+      "step": 284
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3101315174073162e-06,
+      "loss": 0.513,
+      "step": 285
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.2936404685066852e-06,
+      "loss": 0.5159,
+      "step": 286
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.2772175627077204e-06,
+      "loss": 0.5532,
+      "step": 287
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.2608637276838987e-06,
+      "loss": 0.4815,
+      "step": 288
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.244579887207126e-06,
+      "loss": 0.4783,
+      "step": 289
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2283669610955543e-06,
+      "loss": 0.4875,
+      "step": 290
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2122258651616305e-06,
+      "loss": 0.5021,
+      "step": 291
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.1961575111603588e-06,
+      "loss": 0.4948,
+      "step": 292
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.1801628067378033e-06,
+      "loss": 0.4622,
+      "step": 293
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.1642426553798175e-06,
+      "loss": 0.5352,
+      "step": 294
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.148397956361007e-06,
+      "loss": 0.6051,
+      "step": 295
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.1326296046939334e-06,
+      "loss": 0.506,
+      "step": 296
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.1169384910785613e-06,
+      "loss": 0.5109,
+      "step": 297
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1013255018519426e-06,
+      "loss": 0.5157,
+      "step": 298
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.0857915189381512e-06,
+      "loss": 0.5131,
+      "step": 299
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.0703374197984654e-06,
+      "loss": 0.5295,
+      "step": 300
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.054964077381803e-06,
+      "loss": 0.4669,
+      "step": 301
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.0396723600754144e-06,
+      "loss": 0.4831,
+      "step": 302
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.0244631316558268e-06,
+      "loss": 0.5104,
+      "step": 303
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.009337251240055e-06,
+      "loss": 0.4822,
+      "step": 304
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 9.942955732370706e-07,
+      "loss": 0.4747,
+      "step": 305
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 9.793389472995393e-07,
+      "loss": 0.4954,
+      "step": 306
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.644682182758305e-07,
+      "loss": 0.4992,
+      "step": 307
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.496842261622921e-07,
+      "loss": 0.4863,
+      "step": 308
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.349878060557998e-07,
+      "loss": 0.4711,
+      "step": 309
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.203797881065907e-07,
+      "loss": 0.54,
+      "step": 310
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 9.058609974713655e-07,
+      "loss": 0.5112,
+      "step": 311
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 8.914322542666822e-07,
+      "loss": 0.4862,
+      "step": 312
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 8.770943735226303e-07,
+      "loss": 0.4967,
+      "step": 313
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 8.628481651367876e-07,
+      "loss": 0.4836,
+      "step": 314
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 8.486944338284797e-07,
+      "loss": 0.4816,
+      "step": 315
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 8.346339790933167e-07,
+      "loss": 0.4775,
+      "step": 316
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 8.206675951580382e-07,
+      "loss": 0.4966,
+      "step": 317
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 8.067960709356479e-07,
+      "loss": 0.4697,
+      "step": 318
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 7.930201899808476e-07,
+      "loss": 0.4939,
+      "step": 319
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 7.793407304457836e-07,
+      "loss": 0.4889,
+      "step": 320
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 7.657584650360847e-07,
+      "loss": 0.483,
+      "step": 321
+    },
+    {
+      "epoch": 0.75,
+      "eval_loss": 0.5117549896240234,
+      "eval_runtime": 974.2092,
+      "eval_samples_per_second": 1.156,
+      "eval_steps_per_second": 0.145,
+      "step": 321
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 7.522741609672194e-07,
+      "loss": 0.4917,
+      "step": 322
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 7.388885799211573e-07,
+      "loss": 0.5108,
+      "step": 323
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 7.256024780033418e-07,
+      "loss": 0.4503,
+      "step": 324
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 7.124166056999854e-07,
+      "loss": 0.4906,
+      "step": 325
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 6.993317078356709e-07,
+      "loss": 0.492,
+      "step": 326
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 6.863485235312853e-07,
+      "loss": 0.4756,
+      "step": 327
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.734677861622652e-07,
+      "loss": 0.5299,
+      "step": 328
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.60690223317171e-07,
+      "loss": 0.4963,
+      "step": 329
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.480165567565913e-07,
+      "loss": 0.5079,
+      "step": 330
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.354475023723685e-07,
+      "loss": 0.4878,
+      "step": 331
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.229837701471645e-07,
+      "loss": 0.4631,
+      "step": 332
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.106260641143547e-07,
+      "loss": 0.4811,
+      "step": 333
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5.983750823182574e-07,
+      "loss": 0.4797,
+      "step": 334
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5.86231516774709e-07,
+      "loss": 0.5034,
+      "step": 335
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5.741960534319677e-07,
+      "loss": 0.5047,
+      "step": 336
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5.622693721319728e-07,
+      "loss": 0.508,
+      "step": 337
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5.504521465719392e-07,
+      "loss": 0.4828,
+      "step": 338
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5.387450442663026e-07,
+      "loss": 0.4878,
+      "step": 339
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5.271487265090163e-07,
+      "loss": 0.4994,
+      "step": 340
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5.156638483361933e-07,
+      "loss": 0.5069,
+      "step": 341
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5.0429105848911e-07,
+      "loss": 0.4739,
+      "step": 342
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 4.930309993775578e-07,
+      "loss": 0.4773,
+      "step": 343
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 4.818843070435561e-07,
+      "loss": 0.4791,
+      "step": 344
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.708516111254238e-07,
+      "loss": 0.4662,
+      "step": 345
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.5993353482221697e-07,
+      "loss": 0.4834,
+      "step": 346
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.4913069485852197e-07,
+      "loss": 0.5012,
+      "step": 347
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.3844370144962153e-07,
+      "loss": 0.492,
+      "step": 348
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.2787315826702396e-07,
+      "loss": 0.468,
+      "step": 349
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 4.1741966240436446e-07,
+      "loss": 0.485,
+      "step": 350
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 4.070838043436787e-07,
+      "loss": 0.5009,
+      "step": 351
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 3.9686616792204677e-07,
+      "loss": 0.4994,
+      "step": 352
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 3.867673302986161e-07,
+      "loss": 0.4665,
+      "step": 353
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 3.7678786192199695e-07,
+      "loss": 0.482,
+      "step": 354
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 3.6692832649804085e-07,
+      "loss": 0.4914,
+      "step": 355
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 3.571892809580013e-07,
+      "loss": 0.5156,
+      "step": 356
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 3.475712754270716e-07,
+      "loss": 0.5109,
+      "step": 357
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 3.3807485319331037e-07,
+      "loss": 0.4865,
+      "step": 358
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 3.2870055067695557e-07,
+      "loss": 0.4672,
+      "step": 359
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 3.194488974001203e-07,
+      "loss": 0.5019,
+      "step": 360
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 3.1032041595688514e-07,
+      "loss": 0.4891,
+      "step": 361
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 3.0131562198377763e-07,
+      "loss": 0.4944,
+      "step": 362
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 2.9243502413064365e-07,
+      "loss": 0.4971,
+      "step": 363
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 2.8367912403191976e-07,
+      "loss": 0.4814,
+      "step": 364
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 2.7504841627829293e-07,
+      "loss": 0.4853,
+      "step": 365
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 2.6654338838876664e-07,
+      "loss": 0.4849,
+      "step": 366
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.581645207831204e-07,
+      "loss": 0.4495,
+      "step": 367
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.4991228675477293e-07,
+      "loss": 0.5105,
+      "step": 368
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.4178715244404796e-07,
+      "loss": 0.4874,
+      "step": 369
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.3378957681184283e-07,
+      "loss": 0.4764,
+      "step": 370
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.2592001161370392e-07,
+      "loss": 0.516,
+      "step": 371
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.1817890137430936e-07,
+      "loss": 0.5093,
+      "step": 372
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.1056668336235624e-07,
+      "loss": 0.484,
+      "step": 373
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.0308378756586562e-07,
+      "loss": 0.4886,
+      "step": 374
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.9573063666788878e-07,
+      "loss": 0.5288,
+      "step": 375
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.8850764602263428e-07,
+      "loss": 0.5072,
+      "step": 376
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.8141522363200797e-07,
+      "loss": 0.4875,
+      "step": 377
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.7445377012256127e-07,
+      "loss": 0.4842,
+      "step": 378
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.676236787228652e-07,
+      "loss": 0.4777,
+      "step": 379
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.6092533524129623e-07,
+      "loss": 0.4904,
+      "step": 380
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.543591180442436e-07,
+      "loss": 0.4872,
+      "step": 381
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.4792539803473921e-07,
+      "loss": 0.4778,
+      "step": 382
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.4162453863150183e-07,
+      "loss": 0.4944,
+      "step": 383
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.3545689574841341e-07,
+      "loss": 0.4711,
+      "step": 384
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.2942281777441168e-07,
+      "loss": 0.5078,
+      "step": 385
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.2352264555381134e-07,
+      "loss": 0.4752,
+      "step": 386
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.1775671236705366e-07,
+      "loss": 0.4983,
+      "step": 387
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.121253439118769e-07,
+      "loss": 0.4865,
+      "step": 388
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.0662885828492037e-07,
+      "loss": 0.4872,
+      "step": 389
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.0126756596375687e-07,
+      "loss": 0.4642,
+      "step": 390
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 9.604176978935342e-08,
+      "loss": 0.4933,
+      "step": 391
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 9.095176494896662e-08,
+      "loss": 0.5095,
+      "step": 392
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 8.599783895946762e-08,
+      "loss": 0.5184,
+      "step": 393
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 8.118027165109926e-08,
+      "loss": 0.5139,
+      "step": 394
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 7.649933515167407e-08,
+      "loss": 0.4938,
+      "step": 395
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 7.195529387119815e-08,
+      "loss": 0.4902,
+      "step": 396
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 6.75484044869379e-08,
+      "loss": 0.4729,
+      "step": 397
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 6.327891592892126e-08,
+      "loss": 0.4762,
+      "step": 398
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.914706936587494e-08,
+      "loss": 0.4657,
+      "step": 399
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.515309819160402e-08,
+      "loss": 0.4822,
+      "step": 400
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.129722801180542e-08,
+      "loss": 0.4863,
+      "step": 401
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.75796766313269e-08,
+      "loss": 0.4953,
+      "step": 402
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.4000654041862764e-08,
+      "loss": 0.4952,
+      "step": 403
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.05603624100917e-08,
+      "loss": 0.4605,
+      "step": 404
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 3.72589960662581e-08,
+      "loss": 0.4967,
+      "step": 405
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 3.4096741493194196e-08,
+      "loss": 0.4784,
+      "step": 406
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 3.107377731578709e-08,
+      "loss": 0.5004,
+      "step": 407
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.819027429088822e-08,
+      "loss": 0.4836,
+      "step": 408
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.544639529766829e-08,
+      "loss": 0.475,
+      "step": 409
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.284229532841603e-08,
+      "loss": 0.4853,
+      "step": 410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.0378121479783798e-08,
+      "loss": 0.5007,
+      "step": 411
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.8054012944479225e-08,
+      "loss": 0.4842,
+      "step": 412
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5870101003402083e-08,
+      "loss": 0.4887,
+      "step": 413
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.382650901822713e-08,
+      "loss": 0.492,
+      "step": 414
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.1923352424439149e-08,
+      "loss": 0.48,
+      "step": 415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.0160738724809549e-08,
+      "loss": 0.5003,
+      "step": 416
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 8.538767483325384e-09,
+      "loss": 0.5103,
+      "step": 417
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 7.05753031956441e-09,
+      "loss": 0.5304,
+      "step": 418
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5.717110903520617e-09,
+      "loss": 0.5166,
+      "step": 419
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 4.517584950877451e-09,
+      "loss": 0.4779,
+      "step": 420
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 3.4590202187315124e-09,
+      "loss": 0.5138,
+      "step": 421
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.5414765017642285e-09,
+      "loss": 0.5101,
+      "step": 422
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.765005628865113e-09,
+      "loss": 0.4836,
+      "step": 423
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.1296514602038289e-09,
+      "loss": 0.5055,
+      "step": 424
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.354498847521706e-10,
+      "loss": 0.4525,
+      "step": 425
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.8242881825846225e-10,
+      "loss": 0.4781,
+      "step": 426
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 7.060820166826521e-11,
+      "loss": 0.5002,
+      "step": 427
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0,
+      "loss": 0.4674,
+      "step": 428
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.5059249997138977,
+      "eval_runtime": 973.6949,
+      "eval_samples_per_second": 1.156,
+      "eval_steps_per_second": 0.145,
+      "step": 428
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 428,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 214,
+  "total_flos": 179124295434240.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)