Upload folder using huggingface_hub

config.json

@@ -0,0 +1,24 @@
+{
+  "_name_or_path": "/scratch/ace14856qn/finllm/weights/vicuna-7b",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 2048,
+  "max_sequence_length": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "pad_token_id": -1,
+  "rms_norm_eps": 1e-06,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.29.2",
+  "use_cache": false,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "pad_token_id": 0,
+  "transformers_version": "4.29.2"
+}

global_step3639/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00b03898cdd9f6dfd5a25172d6b1a59216396c5cadd8df564b402f191a02e312
+size 168022

global_step3639/zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03df8e22a4c413dc584f686b04542fe2a60f136c24ed945e29d79d3017704504
+size 10107627001

global_step3639/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d94cf7316f596e4de4c49cf79a5ac7642c5b220d7664f54ad0beace3fe97fc8
+size 168022

global_step3639/zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcc119a3938eb92d256d95c08d408c8e8556fcf10a8f771aa567ac42b56c4a9a
+size 10107627001

global_step3639/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fbfc3a858a6c1d6da9c68767928220efa9f7806cf357eb56d277a55650d179ba
+size 168022

global_step3639/zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b010db939a59b1eb80d553539d581589c9dd3066b1799410f75774bb877751a1
+size 10107627001

global_step3639/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7144ab90a39d68984bd0afc781c1aed3971815f744acf46bb76e23ca1c8e6e0
+size 168022

global_step3639/zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e3826a7991563a7a3fd5406075a47695c48840c3119ad95ea2a4d1b43b9dd0f
+size 10107627001

global_step3639/zero_pp_rank_4_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b483ace6acbaf9afae274369ea879ebbff802b037ca6f91fffaed30900b0960b
+size 168022

global_step3639/zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c73df25ad3243114efaee856ac7334a6bf66413c6da50141d4f7dc15d5ad148b
+size 10107627001

global_step3639/zero_pp_rank_5_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8d548c30c216cd78db737ba16d8df715a43bae2477163f11007a184fdf9d0eb
+size 168022

global_step3639/zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bce9cdebe47cd06b7065fd17c8b8ba6b79cefffc41471d94a3e11c2f125441d4
+size 10107627001

global_step3639/zero_pp_rank_6_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:724292452affe5ff8d98db96ebb9f356bd72324f32e2c281e23b608ddc3f245f
+size 168022

global_step3639/zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ed58c2e3cf103ce762ac8dba241d0946025d71b96874710e43af964df801f7a
+size 10107627001

global_step3639/zero_pp_rank_7_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd7630f212570e2624b64ae629c046edc7747508e61846faff7da6c3c22dd3ce
+size 168022

global_step3639/zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ead37ef4a3a2c87d53960d04995431d06eae49d3fa01838b42dd2954f5217320
+size 10107627001

latest

@@ -0,0 +1 @@
+global_step3639

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d7087bfca003c9e2f7295a6a1550795848770f88af5c034ae6876ce06be9216
+size 26953778121

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5539dd7b259c1dd70b24a442cd0db654de172851dc9ad592c00bd0873776a675
+size 21687

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae401ae645379434caffc0185c738039f26685db09ba0aba7a187a8ff97810a9
+size 21687

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c89a43885840bc17c1afdbf25def2d1f36021795ec85893c0d9b1fe1c5b8540d
+size 21687

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47edb967b5d8d4551ccb3bda2bf617d308d0981d37333b4a8839054f68577695
+size 21687

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fef2af776d8df77a340254ef61a2b8d0fcf40ee1ddfaeca625988e0165237db3
+size 21687

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1d3372a72c3e55207954229dad35b30e29127caffc546802c219fbbecc37247
+size 21687

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5903aacbc3201df025352bcb75c398360d0239d5813369122cc0785e94aa68dc
+size 21687

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0102f09cbe2f6f8f4be258e9801fe1495308691ad96ede2d8c503f2d48b5d4e
+size 21687

trainer_state.json

@@ -0,0 +1,2218 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.8001484046500122,
+  "global_step": 3639,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.253431117230021e-06,
+      "loss": 3.9812,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.7916371566987085e-06,
+      "loss": 0.5473,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.506862234460042e-06,
+      "loss": 0.3243,
+      "step": 30
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.977968949561282e-06,
+      "loss": 0.3058,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.3298431961673955e-06,
+      "loss": 0.2495,
+      "step": 50
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 5.610809355829126e-06,
+      "loss": 0.3345,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 5.844707706136422e-06,
+      "loss": 0.2784,
+      "step": 70
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.04506827392873e-06,
+      "loss": 0.2421,
+      "step": 80
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 6.220313630636092e-06,
+      "loss": 0.2661,
+      "step": 90
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 6.376044588167711e-06,
+      "loss": 0.2831,
+      "step": 100
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 6.516174989029968e-06,
+      "loss": 0.2793,
+      "step": 110
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 6.643547802988659e-06,
+      "loss": 0.2488,
+      "step": 120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 6.7602933516900655e-06,
+      "loss": 0.2379,
+      "step": 130
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 6.868049235636083e-06,
+      "loss": 0.2388,
+      "step": 140
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 6.968101466168447e-06,
+      "loss": 0.2273,
+      "step": 150
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.061478425531149e-06,
+      "loss": 0.2571,
+      "step": 160
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 7.149015395297812e-06,
+      "loss": 0.2466,
+      "step": 170
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.231400066791303e-06,
+      "loss": 0.2315,
+      "step": 180
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 7.309205386775784e-06,
+      "loss": 0.2101,
+      "step": 190
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.375715089769526e-06,
+      "loss": 0.2425,
+      "step": 200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 7.44608846772297e-06,
+      "loss": 0.243,
+      "step": 210
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.513093380702437e-06,
+      "loss": 0.2521,
+      "step": 220
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 7.57703761945179e-06,
+      "loss": 0.2401,
+      "step": 230
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.638188625267883e-06,
+      "loss": 0.2342,
+      "step": 240
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 7.696780248446552e-06,
+      "loss": 0.2439,
+      "step": 250
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.753018148386997e-06,
+      "loss": 0.2553,
+      "step": 260
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 7.80708415034011e-06,
+      "loss": 0.2547,
+      "step": 270
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.859139791732239e-06,
+      "loss": 0.2316,
+      "step": 280
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 7.909329232496527e-06,
+      "loss": 0.2227,
+      "step": 290
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 7.957781661555314e-06,
+      "loss": 0.21,
+      "step": 300
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8e-06,
+      "loss": 0.2145,
+      "step": 310
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.98611111111111e-06,
+      "loss": 0.2101,
+      "step": 320
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 7.972222222222223e-06,
+      "loss": 0.1999,
+      "step": 330
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.958333333333333e-06,
+      "loss": 0.2254,
+      "step": 340
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 7.944444444444444e-06,
+      "loss": 0.2597,
+      "step": 350
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.930555555555554e-06,
+      "loss": 0.1814,
+      "step": 360
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 7.916666666666667e-06,
+      "loss": 0.2173,
+      "step": 370
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.902777777777777e-06,
+      "loss": 0.2136,
+      "step": 380
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.88888888888889e-06,
+      "loss": 0.2015,
+      "step": 390
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.875e-06,
+      "loss": 0.2059,
+      "step": 400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.86111111111111e-06,
+      "loss": 0.23,
+      "step": 410
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.847222222222221e-06,
+      "loss": 0.166,
+      "step": 420
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.833333333333333e-06,
+      "loss": 0.2403,
+      "step": 430
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.819444444444444e-06,
+      "loss": 0.2111,
+      "step": 440
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.805555555555555e-06,
+      "loss": 0.2309,
+      "step": 450
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.791666666666667e-06,
+      "loss": 0.2117,
+      "step": 460
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.777777777777777e-06,
+      "loss": 0.2085,
+      "step": 470
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.76388888888889e-06,
+      "loss": 0.1904,
+      "step": 480
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.75e-06,
+      "loss": 0.1954,
+      "step": 490
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.736111111111111e-06,
+      "loss": 0.2056,
+      "step": 500
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.722222222222222e-06,
+      "loss": 0.169,
+      "step": 510
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.708333333333332e-06,
+      "loss": 0.1814,
+      "step": 520
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.694444444444444e-06,
+      "loss": 0.21,
+      "step": 530
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.680555555555555e-06,
+      "loss": 0.2285,
+      "step": 540
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.666666666666667e-06,
+      "loss": 0.2074,
+      "step": 550
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.652777777777778e-06,
+      "loss": 0.173,
+      "step": 560
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.638888888888888e-06,
+      "loss": 0.187,
+      "step": 570
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.625e-06,
+      "loss": 0.2002,
+      "step": 580
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.6111111111111104e-06,
+      "loss": 0.2257,
+      "step": 590
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.597222222222222e-06,
+      "loss": 0.2068,
+      "step": 600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.5833333333333324e-06,
+      "loss": 0.1968,
+      "step": 610
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 7.569444444444445e-06,
+      "loss": 0.2122,
+      "step": 620
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 7.555555555555555e-06,
+      "loss": 0.2306,
+      "step": 630
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 7.541666666666667e-06,
+      "loss": 0.2108,
+      "step": 640
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 7.527777777777777e-06,
+      "loss": 0.1923,
+      "step": 650
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 7.513888888888889e-06,
+      "loss": 0.1989,
+      "step": 660
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 7.499999999999999e-06,
+      "loss": 0.2179,
+      "step": 670
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 7.486111111111111e-06,
+      "loss": 0.1915,
+      "step": 680
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 7.472222222222222e-06,
+      "loss": 0.1965,
+      "step": 690
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 7.458333333333333e-06,
+      "loss": 0.2425,
+      "step": 700
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 7.444444444444444e-06,
+      "loss": 0.1933,
+      "step": 710
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 7.430555555555555e-06,
+      "loss": 0.2314,
+      "step": 720
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 7.416666666666666e-06,
+      "loss": 0.1901,
+      "step": 730
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 7.402777777777778e-06,
+      "loss": 0.1852,
+      "step": 740
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 7.388888888888889e-06,
+      "loss": 0.201,
+      "step": 750
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 7.375e-06,
+      "loss": 0.1636,
+      "step": 760
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 7.36111111111111e-06,
+      "loss": 0.184,
+      "step": 770
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 7.347222222222222e-06,
+      "loss": 0.1618,
+      "step": 780
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 7.333333333333332e-06,
+      "loss": 0.1867,
+      "step": 790
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 7.3194444444444446e-06,
+      "loss": 0.1877,
+      "step": 800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 7.305555555555555e-06,
+      "loss": 0.1569,
+      "step": 810
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 7.291666666666667e-06,
+      "loss": 0.1651,
+      "step": 820
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 7.277777777777777e-06,
+      "loss": 0.215,
+      "step": 830
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 7.263888888888889e-06,
+      "loss": 0.1609,
+      "step": 840
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 7.25e-06,
+      "loss": 0.1935,
+      "step": 850
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 7.236111111111111e-06,
+      "loss": 0.1688,
+      "step": 860
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 7.222222222222222e-06,
+      "loss": 0.1645,
+      "step": 870
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 7.208333333333333e-06,
+      "loss": 0.162,
+      "step": 880
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 7.194444444444444e-06,
+      "loss": 0.1467,
+      "step": 890
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 7.180555555555555e-06,
+      "loss": 0.1579,
+      "step": 900
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 7.166666666666667e-06,
+      "loss": 0.1774,
+      "step": 910
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 7.1527777777777775e-06,
+      "loss": 0.1931,
+      "step": 920
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 7.138888888888889e-06,
+      "loss": 0.1505,
+      "step": 930
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 7.1249999999999995e-06,
+      "loss": 0.1632,
+      "step": 940
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 7.11111111111111e-06,
+      "loss": 0.169,
+      "step": 950
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 7.097222222222222e-06,
+      "loss": 0.1762,
+      "step": 960
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 7.083333333333333e-06,
+      "loss": 0.1838,
+      "step": 970
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 7.0694444444444444e-06,
+      "loss": 0.1622,
+      "step": 980
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 7.055555555555555e-06,
+      "loss": 0.1562,
+      "step": 990
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 7.0416666666666664e-06,
+      "loss": 0.1475,
+      "step": 1000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 7.027777777777777e-06,
+      "loss": 0.1557,
+      "step": 1010
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 7.013888888888889e-06,
+      "loss": 0.1851,
+      "step": 1020
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 7e-06,
+      "loss": 0.1607,
+      "step": 1030
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 6.9861111111111105e-06,
+      "loss": 0.1541,
+      "step": 1040
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 6.972222222222222e-06,
+      "loss": 0.1509,
+      "step": 1050
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 6.9583333333333325e-06,
+      "loss": 0.1754,
+      "step": 1060
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 6.944444444444444e-06,
+      "loss": 0.205,
+      "step": 1070
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 6.930555555555555e-06,
+      "loss": 0.2037,
+      "step": 1080
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 6.916666666666667e-06,
+      "loss": 0.1424,
+      "step": 1090
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 6.902777777777777e-06,
+      "loss": 0.1594,
+      "step": 1100
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 6.888888888888889e-06,
+      "loss": 0.1838,
+      "step": 1110
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 6.874999999999999e-06,
+      "loss": 0.1626,
+      "step": 1120
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 6.86111111111111e-06,
+      "loss": 0.1689,
+      "step": 1130
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 6.847222222222222e-06,
+      "loss": 0.1548,
+      "step": 1140
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 6.833333333333333e-06,
+      "loss": 0.1435,
+      "step": 1150
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 6.819444444444444e-06,
+      "loss": 0.1939,
+      "step": 1160
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 6.805555555555555e-06,
+      "loss": 0.1802,
+      "step": 1170
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 6.791666666666666e-06,
+      "loss": 0.1865,
+      "step": 1180
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 6.777777777777778e-06,
+      "loss": 0.18,
+      "step": 1190
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 6.763888888888889e-06,
+      "loss": 0.1862,
+      "step": 1200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 6.75e-06,
+      "loss": 0.183,
+      "step": 1210
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.234619140625,
+      "eval_runtime": 601.4014,
+      "eval_samples_per_second": 39.832,
+      "eval_steps_per_second": 4.98,
+      "step": 1213
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 6.73611111111111e-06,
+      "loss": 0.1954,
+      "step": 1220
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 6.722222222222222e-06,
+      "loss": 0.1433,
+      "step": 1230
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 6.708333333333332e-06,
+      "loss": 0.1732,
+      "step": 1240
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 6.694444444444445e-06,
+      "loss": 0.1985,
+      "step": 1250
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 6.680555555555555e-06,
+      "loss": 0.1786,
+      "step": 1260
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.167,
+      "step": 1270
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 6.652777777777777e-06,
+      "loss": 0.1548,
+      "step": 1280
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 6.638888888888889e-06,
+      "loss": 0.1559,
+      "step": 1290
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 6.625e-06,
+      "loss": 0.176,
+      "step": 1300
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 6.611111111111111e-06,
+      "loss": 0.1687,
+      "step": 1310
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 6.597222222222222e-06,
+      "loss": 0.1497,
+      "step": 1320
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 6.583333333333333e-06,
+      "loss": 0.1891,
+      "step": 1330
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 6.569444444444444e-06,
+      "loss": 0.1783,
+      "step": 1340
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 6.555555555555555e-06,
+      "loss": 0.168,
+      "step": 1350
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 6.541666666666667e-06,
+      "loss": 0.1871,
+      "step": 1360
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 6.527777777777778e-06,
+      "loss": 0.1601,
+      "step": 1370
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 6.513888888888889e-06,
+      "loss": 0.1466,
+      "step": 1380
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 6.5e-06,
+      "loss": 0.1799,
+      "step": 1390
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 6.48611111111111e-06,
+      "loss": 0.1448,
+      "step": 1400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 6.472222222222222e-06,
+      "loss": 0.1459,
+      "step": 1410
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 6.458333333333333e-06,
+      "loss": 0.1677,
+      "step": 1420
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 6.4444444444444445e-06,
+      "loss": 0.1878,
+      "step": 1430
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 6.430555555555555e-06,
+      "loss": 0.1494,
+      "step": 1440
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 6.4166666666666665e-06,
+      "loss": 0.1406,
+      "step": 1450
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 6.402777777777777e-06,
+      "loss": 0.1538,
+      "step": 1460
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 6.390277777777778e-06,
+      "loss": 0.169,
+      "step": 1470
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 6.376388888888889e-06,
+      "loss": 0.1444,
+      "step": 1480
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 6.3625e-06,
+      "loss": 0.1561,
+      "step": 1490
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 6.348611111111111e-06,
+      "loss": 0.1493,
+      "step": 1500
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 6.334722222222222e-06,
+      "loss": 0.1547,
+      "step": 1510
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 6.320833333333333e-06,
+      "loss": 0.1749,
+      "step": 1520
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 6.3069444444444445e-06,
+      "loss": 0.1903,
+      "step": 1530
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 6.293055555555555e-06,
+      "loss": 0.18,
+      "step": 1540
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.2791666666666665e-06,
+      "loss": 0.1491,
+      "step": 1550
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 6.265277777777777e-06,
+      "loss": 0.1777,
+      "step": 1560
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.2513888888888886e-06,
+      "loss": 0.1226,
+      "step": 1570
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.237499999999999e-06,
+      "loss": 0.1822,
+      "step": 1580
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 6.2236111111111114e-06,
+      "loss": 0.143,
+      "step": 1590
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 6.209722222222222e-06,
+      "loss": 0.1411,
+      "step": 1600
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 6.1958333333333334e-06,
+      "loss": 0.1452,
+      "step": 1610
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 6.181944444444444e-06,
+      "loss": 0.1673,
+      "step": 1620
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 6.169444444444444e-06,
+      "loss": 0.1706,
+      "step": 1630
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 6.155555555555556e-06,
+      "loss": 0.1755,
+      "step": 1640
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 6.141666666666667e-06,
+      "loss": 0.1831,
+      "step": 1650
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 6.127777777777778e-06,
+      "loss": 0.1402,
+      "step": 1660
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 6.113888888888889e-06,
+      "loss": 0.1434,
+      "step": 1670
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 6.099999999999999e-06,
+      "loss": 0.1347,
+      "step": 1680
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 6.086111111111111e-06,
+      "loss": 0.1614,
+      "step": 1690
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 6.072222222222222e-06,
+      "loss": 0.1533,
+      "step": 1700
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 6.0583333333333335e-06,
+      "loss": 0.1666,
+      "step": 1710
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 6.044444444444444e-06,
+      "loss": 0.1725,
+      "step": 1720
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.0305555555555555e-06,
+      "loss": 0.1819,
+      "step": 1730
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 6.016666666666666e-06,
+      "loss": 0.1739,
+      "step": 1740
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.0027777777777775e-06,
+      "loss": 0.1573,
+      "step": 1750
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5.988888888888889e-06,
+      "loss": 0.1458,
+      "step": 1760
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.9749999999999995e-06,
+      "loss": 0.1243,
+      "step": 1770
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5.961111111111111e-06,
+      "loss": 0.1657,
+      "step": 1780
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.9472222222222216e-06,
+      "loss": 0.1871,
+      "step": 1790
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5.933333333333333e-06,
+      "loss": 0.1365,
+      "step": 1800
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.9194444444444444e-06,
+      "loss": 0.0987,
+      "step": 1810
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5.905555555555556e-06,
+      "loss": 0.1396,
+      "step": 1820
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5.8916666666666664e-06,
+      "loss": 0.1693,
+      "step": 1830
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5.877777777777778e-06,
+      "loss": 0.1391,
+      "step": 1840
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5.8638888888888885e-06,
+      "loss": 0.1439,
+      "step": 1850
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5.849999999999999e-06,
+      "loss": 0.1503,
+      "step": 1860
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.836111111111111e-06,
+      "loss": 0.1359,
+      "step": 1870
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.822222222222222e-06,
+      "loss": 0.1423,
+      "step": 1880
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5.808333333333333e-06,
+      "loss": 0.1317,
+      "step": 1890
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.794444444444444e-06,
+      "loss": 0.1431,
+      "step": 1900
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.780555555555555e-06,
+      "loss": 0.1235,
+      "step": 1910
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5.766666666666666e-06,
+      "loss": 0.1397,
+      "step": 1920
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5.752777777777778e-06,
+      "loss": 0.1375,
+      "step": 1930
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5.738888888888889e-06,
+      "loss": 0.1341,
+      "step": 1940
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5.724999999999999e-06,
+      "loss": 0.1408,
+      "step": 1950
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5.711111111111111e-06,
+      "loss": 0.1766,
+      "step": 1960
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5.697222222222221e-06,
+      "loss": 0.177,
+      "step": 1970
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5.683333333333334e-06,
+      "loss": 0.1586,
+      "step": 1980
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5.669444444444444e-06,
+      "loss": 0.1447,
+      "step": 1990
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.655555555555556e-06,
+      "loss": 0.1452,
+      "step": 2000
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5.641666666666666e-06,
+      "loss": 0.1078,
+      "step": 2010
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.627777777777778e-06,
+      "loss": 0.1133,
+      "step": 2020
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.613888888888888e-06,
+      "loss": 0.1288,
+      "step": 2030
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5.6e-06,
+      "loss": 0.1128,
+      "step": 2040
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5.586111111111111e-06,
+      "loss": 0.1298,
+      "step": 2050
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5.572222222222222e-06,
+      "loss": 0.1189,
+      "step": 2060
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5.558333333333333e-06,
+      "loss": 0.1478,
+      "step": 2070
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5.544444444444444e-06,
+      "loss": 0.1179,
+      "step": 2080
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5.530555555555556e-06,
+      "loss": 0.1119,
+      "step": 2090
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5.516666666666667e-06,
+      "loss": 0.1355,
+      "step": 2100
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5.502777777777778e-06,
+      "loss": 0.1105,
+      "step": 2110
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5.488888888888889e-06,
+      "loss": 0.1289,
+      "step": 2120
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5.474999999999999e-06,
+      "loss": 0.1127,
+      "step": 2130
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5.461111111111111e-06,
+      "loss": 0.1814,
+      "step": 2140
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5.447222222222222e-06,
+      "loss": 0.1554,
+      "step": 2150
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5.4333333333333335e-06,
+      "loss": 0.1038,
+      "step": 2160
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5.419444444444444e-06,
+      "loss": 0.1085,
+      "step": 2170
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5.4055555555555556e-06,
+      "loss": 0.1364,
+      "step": 2180
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5.391666666666666e-06,
+      "loss": 0.1089,
+      "step": 2190
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5.377777777777778e-06,
+      "loss": 0.1484,
+      "step": 2200
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5.363888888888889e-06,
+      "loss": 0.1254,
+      "step": 2210
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5.35e-06,
+      "loss": 0.1228,
+      "step": 2220
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5.336111111111111e-06,
+      "loss": 0.154,
+      "step": 2230
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5.322222222222222e-06,
+      "loss": 0.1745,
+      "step": 2240
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5.308333333333333e-06,
+      "loss": 0.1031,
+      "step": 2250
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5.294444444444444e-06,
+      "loss": 0.1359,
+      "step": 2260
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5.280555555555556e-06,
+      "loss": 0.1194,
+      "step": 2270
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5.2666666666666665e-06,
+      "loss": 0.1398,
+      "step": 2280
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5.252777777777778e-06,
+      "loss": 0.1277,
+      "step": 2290
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5.2388888888888885e-06,
+      "loss": 0.1121,
+      "step": 2300
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5.224999999999999e-06,
+      "loss": 0.1067,
+      "step": 2310
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5.211111111111111e-06,
+      "loss": 0.1101,
+      "step": 2320
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5.197222222222222e-06,
+      "loss": 0.1192,
+      "step": 2330
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5.183333333333333e-06,
+      "loss": 0.1055,
+      "step": 2340
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5.169444444444444e-06,
+      "loss": 0.1232,
+      "step": 2350
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5.155555555555555e-06,
+      "loss": 0.1096,
+      "step": 2360
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5.141666666666666e-06,
+      "loss": 0.0831,
+      "step": 2370
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5.127777777777778e-06,
+      "loss": 0.149,
+      "step": 2380
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5.113888888888889e-06,
+      "loss": 0.1328,
+      "step": 2390
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5.0999999999999995e-06,
+      "loss": 0.1193,
+      "step": 2400
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5.086111111111111e-06,
+      "loss": 0.1222,
+      "step": 2410
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5.0722222222222215e-06,
+      "loss": 0.1395,
+      "step": 2420
+    },
+    {
+      "epoch": 1.2,
+      "eval_loss": 0.26611328125,
+      "eval_runtime": 595.4698,
+      "eval_samples_per_second": 40.229,
+      "eval_steps_per_second": 5.03,
+      "step": 2426
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5.058333333333334e-06,
+      "loss": 0.1261,
+      "step": 2430
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5.044444444444444e-06,
+      "loss": 0.1416,
+      "step": 2440
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5.030555555555556e-06,
+      "loss": 0.1453,
+      "step": 2450
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5.016666666666666e-06,
+      "loss": 0.114,
+      "step": 2460
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5.002777777777778e-06,
+      "loss": 0.133,
+      "step": 2470
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.988888888888888e-06,
+      "loss": 0.1207,
+      "step": 2480
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.975e-06,
+      "loss": 0.1123,
+      "step": 2490
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 4.961111111111111e-06,
+      "loss": 0.1039,
+      "step": 2500
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 4.947222222222222e-06,
+      "loss": 0.091,
+      "step": 2510
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 4.933333333333333e-06,
+      "loss": 0.1588,
+      "step": 2520
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 4.919444444444444e-06,
+      "loss": 0.098,
+      "step": 2530
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.905555555555555e-06,
+      "loss": 0.1033,
+      "step": 2540
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.891666666666667e-06,
+      "loss": 0.1473,
+      "step": 2550
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 4.877777777777778e-06,
+      "loss": 0.1331,
+      "step": 2560
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 4.863888888888889e-06,
+      "loss": 0.1196,
+      "step": 2570
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 4.849999999999999e-06,
+      "loss": 0.1379,
+      "step": 2580
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 4.836111111111111e-06,
+      "loss": 0.124,
+      "step": 2590
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 4.822222222222222e-06,
+      "loss": 0.1058,
+      "step": 2600
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 4.808333333333334e-06,
+      "loss": 0.1132,
+      "step": 2610
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 4.794444444444444e-06,
+      "loss": 0.1077,
+      "step": 2620
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 4.780555555555556e-06,
+      "loss": 0.1266,
+      "step": 2630
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 4.766666666666666e-06,
+      "loss": 0.1292,
+      "step": 2640
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 4.752777777777778e-06,
+      "loss": 0.0969,
+      "step": 2650
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 4.738888888888889e-06,
+      "loss": 0.0946,
+      "step": 2660
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 4.725e-06,
+      "loss": 0.1479,
+      "step": 2670
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 4.711111111111111e-06,
+      "loss": 0.133,
+      "step": 2680
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 4.697222222222222e-06,
+      "loss": 0.1118,
+      "step": 2690
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 4.683333333333333e-06,
+      "loss": 0.1362,
+      "step": 2700
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 4.669444444444444e-06,
+      "loss": 0.1384,
+      "step": 2710
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 4.655555555555556e-06,
+      "loss": 0.1275,
+      "step": 2720
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 4.6416666666666666e-06,
+      "loss": 0.1416,
+      "step": 2730
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 4.627777777777778e-06,
+      "loss": 0.1119,
+      "step": 2740
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 4.6138888888888886e-06,
+      "loss": 0.1079,
+      "step": 2750
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.599999999999999e-06,
+      "loss": 0.1197,
+      "step": 2760
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.5861111111111114e-06,
+      "loss": 0.1115,
+      "step": 2770
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 4.572222222222222e-06,
+      "loss": 0.1023,
+      "step": 2780
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 4.5583333333333335e-06,
+      "loss": 0.1109,
+      "step": 2790
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.544444444444444e-06,
+      "loss": 0.1184,
+      "step": 2800
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.5305555555555555e-06,
+      "loss": 0.1103,
+      "step": 2810
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.516666666666666e-06,
+      "loss": 0.1074,
+      "step": 2820
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.502777777777778e-06,
+      "loss": 0.1139,
+      "step": 2830
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.488888888888889e-06,
+      "loss": 0.1107,
+      "step": 2840
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.4749999999999995e-06,
+      "loss": 0.0777,
+      "step": 2850
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.461111111111111e-06,
+      "loss": 0.1348,
+      "step": 2860
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 4.4472222222222215e-06,
+      "loss": 0.1362,
+      "step": 2870
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 4.433333333333333e-06,
+      "loss": 0.1374,
+      "step": 2880
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 4.419444444444444e-06,
+      "loss": 0.1135,
+      "step": 2890
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 4.405555555555556e-06,
+      "loss": 0.1166,
+      "step": 2900
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 4.391666666666666e-06,
+      "loss": 0.1202,
+      "step": 2910
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 4.377777777777778e-06,
+      "loss": 0.1455,
+      "step": 2920
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 4.3638888888888884e-06,
+      "loss": 0.0831,
+      "step": 2930
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 4.35e-06,
+      "loss": 0.1211,
+      "step": 2940
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 4.336111111111111e-06,
+      "loss": 0.1194,
+      "step": 2950
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 4.322222222222222e-06,
+      "loss": 0.1124,
+      "step": 2960
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 4.308333333333333e-06,
+      "loss": 0.0947,
+      "step": 2970
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 4.294444444444444e-06,
+      "loss": 0.1152,
+      "step": 2980
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 4.280555555555555e-06,
+      "loss": 0.0858,
+      "step": 2990
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 4.266666666666667e-06,
+      "loss": 0.087,
+      "step": 3000
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 4.252777777777778e-06,
+      "loss": 0.1237,
+      "step": 3010
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 4.238888888888889e-06,
+      "loss": 0.0809,
+      "step": 3020
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 4.224999999999999e-06,
+      "loss": 0.1022,
+      "step": 3030
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 4.211111111111111e-06,
+      "loss": 0.101,
+      "step": 3040
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 4.197222222222221e-06,
+      "loss": 0.1327,
+      "step": 3050
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 4.183333333333334e-06,
+      "loss": 0.1056,
+      "step": 3060
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 4.169444444444444e-06,
+      "loss": 0.1129,
+      "step": 3070
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 4.155555555555556e-06,
+      "loss": 0.1119,
+      "step": 3080
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 4.141666666666666e-06,
+      "loss": 0.1287,
+      "step": 3090
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 4.127777777777778e-06,
+      "loss": 0.1179,
+      "step": 3100
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 4.113888888888889e-06,
+      "loss": 0.1152,
+      "step": 3110
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 4.1e-06,
+      "loss": 0.0833,
+      "step": 3120
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 4.086111111111111e-06,
+      "loss": 0.1304,
+      "step": 3130
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 4.072222222222222e-06,
+      "loss": 0.0813,
+      "step": 3140
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 4.058333333333333e-06,
+      "loss": 0.1068,
+      "step": 3150
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 4.044444444444444e-06,
+      "loss": 0.0998,
+      "step": 3160
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 4.030555555555556e-06,
+      "loss": 0.1109,
+      "step": 3170
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 4.016666666666667e-06,
+      "loss": 0.1193,
+      "step": 3180
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 4.002777777777778e-06,
+      "loss": 0.1234,
+      "step": 3190
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 3.988888888888889e-06,
+      "loss": 0.1049,
+      "step": 3200
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 3.975e-06,
+      "loss": 0.1017,
+      "step": 3210
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 3.9611111111111115e-06,
+      "loss": 0.1044,
+      "step": 3220
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 3.947222222222222e-06,
+      "loss": 0.1263,
+      "step": 3230
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 3.933333333333333e-06,
+      "loss": 0.1419,
+      "step": 3240
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 3.919444444444444e-06,
+      "loss": 0.1294,
+      "step": 3250
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 3.9055555555555555e-06,
+      "loss": 0.0939,
+      "step": 3260
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 3.891666666666666e-06,
+      "loss": 0.1374,
+      "step": 3270
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 3.8777777777777775e-06,
+      "loss": 0.0959,
+      "step": 3280
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 3.863888888888889e-06,
+      "loss": 0.1009,
+      "step": 3290
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 3.8499999999999996e-06,
+      "loss": 0.1305,
+      "step": 3300
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 3.836111111111111e-06,
+      "loss": 0.1303,
+      "step": 3310
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 3.8222222222222224e-06,
+      "loss": 0.1282,
+      "step": 3320
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 3.808333333333333e-06,
+      "loss": 0.1053,
+      "step": 3330
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 3.794444444444444e-06,
+      "loss": 0.1042,
+      "step": 3340
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 3.7805555555555555e-06,
+      "loss": 0.0849,
+      "step": 3350
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 3.7666666666666665e-06,
+      "loss": 0.1274,
+      "step": 3360
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 3.7527777777777775e-06,
+      "loss": 0.1228,
+      "step": 3370
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 3.738888888888889e-06,
+      "loss": 0.1129,
+      "step": 3380
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 3.725e-06,
+      "loss": 0.1128,
+      "step": 3390
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 3.711111111111111e-06,
+      "loss": 0.1317,
+      "step": 3400
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 3.6972222222222224e-06,
+      "loss": 0.1246,
+      "step": 3410
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 3.683333333333333e-06,
+      "loss": 0.0806,
+      "step": 3420
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 3.669444444444444e-06,
+      "loss": 0.119,
+      "step": 3430
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 3.6555555555555554e-06,
+      "loss": 0.0893,
+      "step": 3440
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 3.6416666666666664e-06,
+      "loss": 0.1058,
+      "step": 3450
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 3.6277777777777774e-06,
+      "loss": 0.0944,
+      "step": 3460
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 3.613888888888889e-06,
+      "loss": 0.126,
+      "step": 3470
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 3.6e-06,
+      "loss": 0.1104,
+      "step": 3480
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 3.5861111111111113e-06,
+      "loss": 0.1108,
+      "step": 3490
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 3.5722222222222223e-06,
+      "loss": 0.1197,
+      "step": 3500
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 3.558333333333333e-06,
+      "loss": 0.1182,
+      "step": 3510
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 3.5444444444444443e-06,
+      "loss": 0.0988,
+      "step": 3520
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 3.5305555555555553e-06,
+      "loss": 0.0949,
+      "step": 3530
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 3.5166666666666663e-06,
+      "loss": 0.1102,
+      "step": 3540
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 3.5027777777777777e-06,
+      "loss": 0.1203,
+      "step": 3550
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 3.4888888888888888e-06,
+      "loss": 0.1343,
+      "step": 3560
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 3.4749999999999998e-06,
+      "loss": 0.0851,
+      "step": 3570
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 3.461111111111111e-06,
+      "loss": 0.1181,
+      "step": 3580
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 3.447222222222222e-06,
+      "loss": 0.0931,
+      "step": 3590
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 3.433333333333333e-06,
+      "loss": 0.1068,
+      "step": 3600
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 3.4194444444444442e-06,
+      "loss": 0.0884,
+      "step": 3610
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 3.4055555555555552e-06,
+      "loss": 0.1326,
+      "step": 3620
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 3.3916666666666667e-06,
+      "loss": 0.1307,
+      "step": 3630
+    },
+    {
+      "epoch": 1.8,
+      "eval_loss": 0.249267578125,
+      "eval_runtime": 590.255,
+      "eval_samples_per_second": 40.584,
+      "eval_steps_per_second": 5.074,
+      "step": 3639
+    }
+  ],
+  "max_steps": 6063,
+  "num_train_epochs": 3,
+  "total_flos": 134441240494080.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c42f106339a0328c004e4952494046976bb21234b5589de38804b01f680386c0
+size 5115

zero_to_fp32.py

@@ -0,0 +1,578 @@
+#!/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 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_dicts.append(torch.load(f, map_location=device))
+
+    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 _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()
+            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("-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)