Duplicate from banghua/n_rm

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

global_step1400/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:964dfeff5337bbeaf204f228250b1f0c0a9a34b53cac198d9eb922b694fee53e
+size 10107635127

global_step1400/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02703e0fb94a05ed844c3f5fb0fdd4dff63a33d42a5899717cf27360074ba597
+size 10107635511

global_step1400/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a1d87dc115f836f3b418f2c0629cccdf8faa4821b7f0d75f550266236414339
+size 10107635511

global_step1400/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:f5ca02c175d6db6d53531a8384558640f08b9feef50a5fe267dd43caa13b084c
+size 10107635703

global_step1400/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:f530314f97b676a3a2074902239ae721333eadf3909a5ab9b552590935863623
+size 10107635575

global_step1400/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7803dbd98dfd3cb2e46e9ade255baf1151c332d79849d8c26b4f36f3ce9aac27
+size 10107635639

global_step1400/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:e1acdf3066c608dd50252cc6c12a6b17d13033e0a62f5bad1192e679b6f73f6b
+size 10107635575

global_step1400/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19680cdfcf02f32d10f4c744adff0dedd1405b7e9672d677cacb85039200ab64
+size 10107635255

global_step1400/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b2d03e9f8d039c27856dee853487a996b9033e94ae8a820571657e9b9f4d74c
+size 13477033283

latest

@@ -0,0 +1 @@
+global_step1400

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e822738b1730aee4bcd4695d25836907dd3b98dff1ac112260d89c2085c0a743
+size 26691724373

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4077b34b03f79b052bd53a09b269b2df2b9b4edbba886d14e19bc0ff6508ab00
+size 21687

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d6fc2e3e4688f0af35b81181a28d78078f10a4e63237915ef2e25612318a5b3
+size 21687

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b48f9dd20406c2f7ee61d289c703091bfb05aca0d3d4bc461fec41b66d43bfa5
+size 21687

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a513659dd182753b05daade475bbf0a51cfafbb0119721a6e1b8d60c45dacdb1
+size 21687

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:130f08688b3d017d9a1c2ac8ef50fa6d9637aa718b7ae19c54fd23cfd35490c2
+size 21687

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22929ee481c4bff4217495b52b918cafcda752ff40c793ba1081d43d57f7fa58
+size 21687

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98553316d85e76503cf36b5f7bf067dd5c1d3db5fda7842498a035a54c847a32
+size 21687

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6d3a685adbcf6a9447697719aec403e5cd0262aa6decadfb656356ede6df4e8
+size 21687

trainer_state.json

@@ -0,0 +1,985 @@
+{
+  "best_metric": 0.2997502386569977,
+  "best_model_checkpoint": "rm_ckptreward-p100-w0.3-s0/checkpoint-1400",
+  "epoch": 0.8934269304403318,
+  "eval_steps": 200,
+  "global_step": 1400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9999999999999996e-06,
+      "loss": 0.3897,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.505149978319905e-06,
+      "loss": 0.3441,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 7.385606273598311e-06,
+      "loss": 0.326,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 8.010299956639811e-06,
+      "loss": 0.3131,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 8.494850021680093e-06,
+      "loss": 0.3187,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 8.890756251918216e-06,
+      "loss": 0.3105,
+      "step": 60
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.225490200071284e-06,
+      "loss": 0.3137,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.515449934959717e-06,
+      "loss": 0.3115,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.771212547196623e-06,
+      "loss": 0.309,
+      "step": 90
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 0.3073,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1e-05,
+      "loss": 0.3152,
+      "step": 110
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1e-05,
+      "loss": 0.3064,
+      "step": 120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1e-05,
+      "loss": 0.3094,
+      "step": 130
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1e-05,
+      "loss": 0.3009,
+      "step": 140
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1e-05,
+      "loss": 0.3041,
+      "step": 150
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1e-05,
+      "loss": 0.305,
+      "step": 160
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1e-05,
+      "loss": 0.304,
+      "step": 170
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1e-05,
+      "loss": 0.2999,
+      "step": 180
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1e-05,
+      "loss": 0.3095,
+      "step": 190
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1e-05,
+      "loss": 0.3042,
+      "step": 200
+    },
+    {
+      "epoch": 0.13,
+      "eval_val_accuracy": 896.5238095238095,
+      "eval_val_loss": 0.30677446722984314,
+      "eval_val_runtime": 2250.4996,
+      "eval_val_samples_per_second": 4.443,
+      "eval_val_steps_per_second": 0.555,
+      "step": 200
+    },
+    {
+      "epoch": 0.13,
+      "eval_test_accuracy": 898.6904761904761,
+      "eval_test_loss": 0.30444496870040894,
+      "eval_test_runtime": 2251.863,
+      "eval_test_samples_per_second": 4.441,
+      "eval_test_steps_per_second": 0.555,
+      "step": 200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1e-05,
+      "loss": 0.3018,
+      "step": 210
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1e-05,
+      "loss": 0.3078,
+      "step": 220
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1e-05,
+      "loss": 0.3028,
+      "step": 230
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1e-05,
+      "loss": 0.3043,
+      "step": 240
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1e-05,
+      "loss": 0.3081,
+      "step": 250
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1e-05,
+      "loss": 0.3056,
+      "step": 260
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1e-05,
+      "loss": 0.3083,
+      "step": 270
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1e-05,
+      "loss": 0.3,
+      "step": 280
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1e-05,
+      "loss": 0.3036,
+      "step": 290
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1e-05,
+      "loss": 0.307,
+      "step": 300
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1e-05,
+      "loss": 0.3029,
+      "step": 310
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1e-05,
+      "loss": 0.3035,
+      "step": 320
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1e-05,
+      "loss": 0.3052,
+      "step": 330
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1e-05,
+      "loss": 0.3034,
+      "step": 340
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1e-05,
+      "loss": 0.3012,
+      "step": 350
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1e-05,
+      "loss": 0.3023,
+      "step": 360
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1e-05,
+      "loss": 0.305,
+      "step": 370
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1e-05,
+      "loss": 0.3075,
+      "step": 380
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1e-05,
+      "loss": 0.3043,
+      "step": 390
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1e-05,
+      "loss": 0.302,
+      "step": 400
+    },
+    {
+      "epoch": 0.26,
+      "eval_val_accuracy": 901.1428571428571,
+      "eval_val_loss": 0.3049960434436798,
+      "eval_val_runtime": 2251.3044,
+      "eval_val_samples_per_second": 4.442,
+      "eval_val_steps_per_second": 0.555,
+      "step": 400
+    },
+    {
+      "epoch": 0.26,
+      "eval_test_accuracy": 902.547619047619,
+      "eval_test_loss": 0.3042500615119934,
+      "eval_test_runtime": 2251.0717,
+      "eval_test_samples_per_second": 4.442,
+      "eval_test_steps_per_second": 0.555,
+      "step": 400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1e-05,
+      "loss": 0.2984,
+      "step": 410
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1e-05,
+      "loss": 0.3001,
+      "step": 420
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1e-05,
+      "loss": 0.3034,
+      "step": 430
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1e-05,
+      "loss": 0.3005,
+      "step": 440
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1e-05,
+      "loss": 0.308,
+      "step": 450
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1e-05,
+      "loss": 0.3011,
+      "step": 460
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1e-05,
+      "loss": 0.3038,
+      "step": 470
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1e-05,
+      "loss": 0.2988,
+      "step": 480
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1e-05,
+      "loss": 0.3057,
+      "step": 490
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1e-05,
+      "loss": 0.3077,
+      "step": 500
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1e-05,
+      "loss": 0.2962,
+      "step": 510
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1e-05,
+      "loss": 0.3074,
+      "step": 520
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1e-05,
+      "loss": 0.2979,
+      "step": 530
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1e-05,
+      "loss": 0.3027,
+      "step": 540
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1e-05,
+      "loss": 0.2993,
+      "step": 550
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1e-05,
+      "loss": 0.3006,
+      "step": 560
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1e-05,
+      "loss": 0.3018,
+      "step": 570
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1e-05,
+      "loss": 0.3009,
+      "step": 580
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1e-05,
+      "loss": 0.2983,
+      "step": 590
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1e-05,
+      "loss": 0.2968,
+      "step": 600
+    },
+    {
+      "epoch": 0.38,
+      "eval_val_accuracy": 902.4047619047618,
+      "eval_val_loss": 0.30277058482170105,
+      "eval_val_runtime": 2248.9082,
+      "eval_val_samples_per_second": 4.447,
+      "eval_val_steps_per_second": 0.556,
+      "step": 600
+    },
+    {
+      "epoch": 0.38,
+      "eval_test_accuracy": 904.8809523809524,
+      "eval_test_loss": 0.3016507923603058,
+      "eval_test_runtime": 2249.9795,
+      "eval_test_samples_per_second": 4.444,
+      "eval_test_steps_per_second": 0.556,
+      "step": 600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1e-05,
+      "loss": 0.3034,
+      "step": 610
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1e-05,
+      "loss": 0.304,
+      "step": 620
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1e-05,
+      "loss": 0.2947,
+      "step": 630
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1e-05,
+      "loss": 0.2953,
+      "step": 640
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1e-05,
+      "loss": 0.3028,
+      "step": 650
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1e-05,
+      "loss": 0.2979,
+      "step": 660
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1e-05,
+      "loss": 0.3039,
+      "step": 670
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1e-05,
+      "loss": 0.2969,
+      "step": 680
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1e-05,
+      "loss": 0.2994,
+      "step": 690
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1e-05,
+      "loss": 0.2961,
+      "step": 700
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1e-05,
+      "loss": 0.3031,
+      "step": 710
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1e-05,
+      "loss": 0.3037,
+      "step": 720
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1e-05,
+      "loss": 0.3008,
+      "step": 730
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1e-05,
+      "loss": 0.3089,
+      "step": 740
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1e-05,
+      "loss": 0.3061,
+      "step": 750
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1e-05,
+      "loss": 0.2987,
+      "step": 760
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1e-05,
+      "loss": 0.3041,
+      "step": 770
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1e-05,
+      "loss": 0.2993,
+      "step": 780
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1e-05,
+      "loss": 0.2936,
+      "step": 790
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1e-05,
+      "loss": 0.301,
+      "step": 800
+    },
+    {
+      "epoch": 0.51,
+      "eval_val_accuracy": 904.3333333333333,
+      "eval_val_loss": 0.30156368017196655,
+      "eval_val_runtime": 2246.2231,
+      "eval_val_samples_per_second": 4.452,
+      "eval_val_steps_per_second": 0.556,
+      "step": 800
+    },
+    {
+      "epoch": 0.51,
+      "eval_test_accuracy": 907.6904761904761,
+      "eval_test_loss": 0.3001132309436798,
+      "eval_test_runtime": 2247.6939,
+      "eval_test_samples_per_second": 4.449,
+      "eval_test_steps_per_second": 0.556,
+      "step": 800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1e-05,
+      "loss": 0.299,
+      "step": 810
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1e-05,
+      "loss": 0.304,
+      "step": 820
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1e-05,
+      "loss": 0.3013,
+      "step": 830
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1e-05,
+      "loss": 0.2989,
+      "step": 840
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1e-05,
+      "loss": 0.3027,
+      "step": 850
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1e-05,
+      "loss": 0.3019,
+      "step": 860
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1e-05,
+      "loss": 0.3033,
+      "step": 870
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1e-05,
+      "loss": 0.2984,
+      "step": 880
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1e-05,
+      "loss": 0.2967,
+      "step": 890
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1e-05,
+      "loss": 0.3,
+      "step": 900
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1e-05,
+      "loss": 0.3025,
+      "step": 910
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1e-05,
+      "loss": 0.3037,
+      "step": 920
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1e-05,
+      "loss": 0.297,
+      "step": 930
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1e-05,
+      "loss": 0.3019,
+      "step": 940
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1e-05,
+      "loss": 0.2966,
+      "step": 950
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1e-05,
+      "loss": 0.3023,
+      "step": 960
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1e-05,
+      "loss": 0.2961,
+      "step": 970
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1e-05,
+      "loss": 0.2987,
+      "step": 980
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1e-05,
+      "loss": 0.2995,
+      "step": 990
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1e-05,
+      "loss": 0.2976,
+      "step": 1000
+    },
+    {
+      "epoch": 0.64,
+      "eval_val_accuracy": 906.5952380952382,
+      "eval_val_loss": 0.3006158769130707,
+      "eval_val_runtime": 2245.8387,
+      "eval_val_samples_per_second": 4.453,
+      "eval_val_steps_per_second": 0.557,
+      "step": 1000
+    },
+    {
+      "epoch": 0.64,
+      "eval_test_accuracy": 907.5952380952382,
+      "eval_test_loss": 0.29969537258148193,
+      "eval_test_runtime": 2246.9928,
+      "eval_test_samples_per_second": 4.45,
+      "eval_test_steps_per_second": 0.556,
+      "step": 1000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1e-05,
+      "loss": 0.2988,
+      "step": 1010
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1e-05,
+      "loss": 0.2999,
+      "step": 1020
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1e-05,
+      "loss": 0.2995,
+      "step": 1030
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1e-05,
+      "loss": 0.2991,
+      "step": 1040
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1e-05,
+      "loss": 0.2922,
+      "step": 1050
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1e-05,
+      "loss": 0.2957,
+      "step": 1060
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1e-05,
+      "loss": 0.305,
+      "step": 1070
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1e-05,
+      "loss": 0.2996,
+      "step": 1080
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1e-05,
+      "loss": 0.3032,
+      "step": 1090
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1e-05,
+      "loss": 0.3,
+      "step": 1100
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1e-05,
+      "loss": 0.2937,
+      "step": 1110
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1e-05,
+      "loss": 0.2976,
+      "step": 1120
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1e-05,
+      "loss": 0.2993,
+      "step": 1130
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1e-05,
+      "loss": 0.2994,
+      "step": 1140
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1e-05,
+      "loss": 0.3033,
+      "step": 1150
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1e-05,
+      "loss": 0.3029,
+      "step": 1160
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1e-05,
+      "loss": 0.3014,
+      "step": 1170
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1e-05,
+      "loss": 0.2976,
+      "step": 1180
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1e-05,
+      "loss": 0.2978,
+      "step": 1190
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1e-05,
+      "loss": 0.2977,
+      "step": 1200
+    },
+    {
+      "epoch": 0.77,
+      "eval_val_accuracy": 905.9285714285714,
+      "eval_val_loss": 0.30016693472862244,
+      "eval_val_runtime": 2246.3367,
+      "eval_val_samples_per_second": 4.452,
+      "eval_val_steps_per_second": 0.556,
+      "step": 1200
+    },
+    {
+      "epoch": 0.77,
+      "eval_test_accuracy": 907.4761904761905,
+      "eval_test_loss": 0.29916131496429443,
+      "eval_test_runtime": 2247.4238,
+      "eval_test_samples_per_second": 4.45,
+      "eval_test_steps_per_second": 0.556,
+      "step": 1200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1e-05,
+      "loss": 0.3029,
+      "step": 1210
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1e-05,
+      "loss": 0.2976,
+      "step": 1220
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1e-05,
+      "loss": 0.2965,
+      "step": 1230
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1e-05,
+      "loss": 0.2966,
+      "step": 1240
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1e-05,
+      "loss": 0.2993,
+      "step": 1250
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1e-05,
+      "loss": 0.2961,
+      "step": 1260
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1e-05,
+      "loss": 0.2959,
+      "step": 1270
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1e-05,
+      "loss": 0.2998,
+      "step": 1280
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1e-05,
+      "loss": 0.3024,
+      "step": 1290
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1e-05,
+      "loss": 0.2934,
+      "step": 1300
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1e-05,
+      "loss": 0.2966,
+      "step": 1310
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1e-05,
+      "loss": 0.3051,
+      "step": 1320
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1e-05,
+      "loss": 0.2988,
+      "step": 1330
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1e-05,
+      "loss": 0.3001,
+      "step": 1340
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1e-05,
+      "loss": 0.3032,
+      "step": 1350
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1e-05,
+      "loss": 0.2972,
+      "step": 1360
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1e-05,
+      "loss": 0.2996,
+      "step": 1370
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1e-05,
+      "loss": 0.298,
+      "step": 1380
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1e-05,
+      "loss": 0.2959,
+      "step": 1390
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1e-05,
+      "loss": 0.2951,
+      "step": 1400
+    },
+    {
+      "epoch": 0.89,
+      "eval_val_accuracy": 909.1904761904761,
+      "eval_val_loss": 0.2997502386569977,
+      "eval_val_runtime": 2245.4318,
+      "eval_val_samples_per_second": 4.453,
+      "eval_val_steps_per_second": 0.557,
+      "step": 1400
+    },
+    {
+      "epoch": 0.89,
+      "eval_test_accuracy": 907.5238095238095,
+      "eval_test_loss": 0.2989273965358734,
+      "eval_test_runtime": 2246.796,
+      "eval_test_samples_per_second": 4.451,
+      "eval_test_steps_per_second": 0.556,
+      "step": 1400
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 7835,
+  "num_train_epochs": 5,
+  "save_steps": 200,
+  "total_flos": 0.0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0970cf0f8a8b2c721397d2b59e62220a6e0b2228225ba1537abeaf5fd367de8
+size 5307

zero_to_fp32.py

@@ -0,0 +1,587 @@
+#!/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 _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("-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)