Training in progress, epoch 1, checkpoint

last-checkpoint/README.md

@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.5.0

last-checkpoint/adapter_config.json

@@ -0,0 +1,21 @@
+{
+  "auto_mapping": null,
+  "base_model_name_or_path": "google/flan-t5-xxl",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "revision": null,
+  "target_modules": [
+    "q",
+    "v"
+  ],
+  "task_type": "SEQ_2_SEQ_LM"
+}

last-checkpoint/adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7eaa51169fb5f5f33b328f26090dcd19a47bde0a9efe64e78856dbbe04a07e7a
+size 888

last-checkpoint/global_step1581/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:ccc367dbb9740751e85fd2958b4867e548bfad92bc52db0c411c1d02943344cf
+size 56626640

last-checkpoint/global_step1581/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:f9f583059aebf5c348baa213b71780e3beb0ed5bea73b3eaaae551a3ee1fa3b6
+size 56626640

last-checkpoint/global_step1581/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5d9a7306e11004f5b07be5adcf241a253323781fe1b13c03e09f6e9c51b1a4f
+size 11136132566

last-checkpoint/global_step1581/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a71dd45f808f455f2743152cc5c2c5389f3f01dc02239d02a8986dbc29229ccf
+size 11136132374

last-checkpoint/latest

@@ -0,0 +1 @@
+global_step1581

last-checkpoint/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f232d8d672d6ff9e3cc5ac4ba64234b9f002d5e144a43ee183dc01871e2f10e5
+size 14512

last-checkpoint/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e25f573f6bcb03f0bbfe93da192a1ab49ba692dcab33095de44fc91e5e4216a5
+size 14512

last-checkpoint/trainer_state.json

@@ -1,1250 +1,56 @@
 {
-  "best_metric": 1.5654487609863281,
-  "best_model_checkpoint": "dq158/pingusPongus/checkpoint-6323",
-  "epoch": 13.0,
+  "best_metric": 2.2135026454925537,
+  "best_model_checkpoint": "dq158/pingusPongus/checkpoint-1581",
+  "epoch": 1.0,
   "eval_steps": 500,
-  "global_step": 82199,
+  "global_step": 1581,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
-    {
-      "epoch": 0.08,
-      "learning_rate": 5e-06,
-      "loss": 1.8585,
-      "step": 500
-    },
-    {
-      "epoch": 0.16,
-      "learning_rate": 4.999805607800008e-06,
-      "loss": 1.823,
-      "step": 1000
-    },
-    {
-      "epoch": 0.24,
-      "learning_rate": 4.999222461430692e-06,
-      "loss": 1.8388,
-      "step": 1500
-    },
     {
       "epoch": 0.32,
-      "learning_rate": 4.998250651579336e-06,
-      "loss": 1.8372,
-      "step": 2000
-    },
-    {
-      "epoch": 0.4,
-      "learning_rate": 4.996890329375747e-06,
-      "loss": 1.8066,
-      "step": 2500
-    },
-    {
-      "epoch": 0.47,
-      "learning_rate": 4.995141706368742e-06,
-      "loss": 1.8485,
-      "step": 3000
-    },
-    {
-      "epoch": 0.55,
-      "learning_rate": 4.993005054493262e-06,
-      "loss": 1.8243,
-      "step": 3500
+      "learning_rate": 0.0008497772211805019,
+      "loss": 2.4966,
+      "step": 500
     },
     {
       "epoch": 0.63,
-      "learning_rate": 4.990480706028073e-06,
-      "loss": 1.8278,
-      "step": 4000
-    },
-    {
-      "epoch": 0.71,
-      "learning_rate": 4.987569053544098e-06,
-      "loss": 1.8126,
-      "step": 4500
-    },
-    {
-      "epoch": 0.79,
-      "learning_rate": 4.98427054984336e-06,
-      "loss": 1.8277,
-      "step": 5000
-    },
-    {
-      "epoch": 0.87,
-      "learning_rate": 4.980585707888573e-06,
-      "loss": 1.8475,
-      "step": 5500
+      "learning_rate": 0.0009445572420019074,
+      "loss": 2.3574,
+      "step": 1000
     },
     {
       "epoch": 0.95,
-      "learning_rate": 4.976515100723365e-06,
-      "loss": 1.8441,
-      "step": 6000
+      "learning_rate": 0.0009999999999999998,
+      "loss": 2.3094,
+      "step": 1500
     },
     {
       "epoch": 1.0,
       "eval_bleu": 1.0,
       "eval_brevity_penalty": 1.0,
       "eval_length_ratio": 1.0,
-      "eval_loss": 1.5654487609863281,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52412,
-      "eval_runtime": 683.1649,
-      "eval_samples_per_second": 4.115,
-      "eval_steps_per_second": 1.029,
-      "eval_translation_length": 52412,
-      "step": 6323
-    },
-    {
-      "epoch": 1.03,
-      "learning_rate": 4.972059361383162e-06,
-      "loss": 1.8281,
-      "step": 6500
-    },
-    {
-      "epoch": 1.11,
-      "learning_rate": 4.9672191827967395e-06,
-      "loss": 1.8431,
-      "step": 7000
-    },
-    {
-      "epoch": 1.19,
-      "learning_rate": 4.961995317678472e-06,
-      "loss": 1.8261,
-      "step": 7500
-    },
-    {
-      "epoch": 1.27,
-      "learning_rate": 4.9563885784112645e-06,
-      "loss": 1.8253,
-      "step": 8000
-    },
-    {
-      "epoch": 1.34,
-      "learning_rate": 4.950399836920221e-06,
-      "loss": 1.847,
-      "step": 8500
-    },
-    {
-      "epoch": 1.42,
-      "learning_rate": 4.944030024537049e-06,
-      "loss": 1.8209,
-      "step": 9000
-    },
-    {
-      "epoch": 1.5,
-      "learning_rate": 4.937280131855223e-06,
-      "loss": 1.8153,
-      "step": 9500
-    },
-    {
-      "epoch": 1.58,
-      "learning_rate": 4.930151208575933e-06,
-      "loss": 1.8591,
-      "step": 10000
-    },
-    {
-      "epoch": 1.66,
-      "learning_rate": 4.9226443633448426e-06,
-      "loss": 1.7892,
-      "step": 10500
-    },
-    {
-      "epoch": 1.74,
-      "learning_rate": 4.91476076357968e-06,
-      "loss": 1.8601,
-      "step": 11000
-    },
-    {
-      "epoch": 1.82,
-      "learning_rate": 4.906501635288687e-06,
-      "loss": 1.8231,
-      "step": 11500
-    },
-    {
-      "epoch": 1.9,
-      "learning_rate": 4.8978682628799575e-06,
-      "loss": 1.7805,
-      "step": 12000
-    },
-    {
-      "epoch": 1.98,
-      "learning_rate": 4.888861988961698e-06,
-      "loss": 1.8429,
-      "step": 12500
-    },
-    {
-      "epoch": 2.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5659914016723633,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52542,
-      "eval_runtime": 676.0746,
-      "eval_samples_per_second": 4.158,
-      "eval_steps_per_second": 1.04,
-      "eval_translation_length": 52542,
-      "step": 12646
-    },
-    {
-      "epoch": 2.06,
-      "learning_rate": 4.879484214133427e-06,
-      "loss": 1.838,
-      "step": 13000
-    },
-    {
-      "epoch": 2.14,
-      "learning_rate": 4.8697363967681696e-06,
-      "loss": 1.8539,
-      "step": 13500
-    },
-    {
-      "epoch": 2.21,
-      "learning_rate": 4.8596200527856564e-06,
-      "loss": 1.8081,
-      "step": 14000
-    },
-    {
-      "epoch": 2.29,
-      "learning_rate": 4.849136755416576e-06,
-      "loss": 1.8135,
-      "step": 14500
-    },
-    {
-      "epoch": 2.37,
-      "learning_rate": 4.838288134957921e-06,
-      "loss": 1.8273,
-      "step": 15000
-    },
-    {
-      "epoch": 2.45,
-      "learning_rate": 4.827075878519448e-06,
-      "loss": 1.8316,
-      "step": 15500
-    },
-    {
-      "epoch": 2.53,
-      "learning_rate": 4.815501729761316e-06,
-      "loss": 1.8101,
-      "step": 16000
-    },
-    {
-      "epoch": 2.61,
-      "learning_rate": 4.803567488622915e-06,
-      "loss": 1.8257,
-      "step": 16500
-    },
-    {
-      "epoch": 2.69,
-      "learning_rate": 4.791275011042958e-06,
-      "loss": 1.8036,
-      "step": 17000
-    },
-    {
-      "epoch": 2.77,
-      "learning_rate": 4.778626208670853e-06,
-      "loss": 1.8235,
-      "step": 17500
-    },
-    {
-      "epoch": 2.85,
-      "learning_rate": 4.765623048569417e-06,
-      "loss": 1.8133,
-      "step": 18000
-    },
-    {
-      "epoch": 2.93,
-      "learning_rate": 4.752267552908968e-06,
-      "loss": 1.8316,
-      "step": 18500
-    },
-    {
-      "epoch": 3.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5685368776321411,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52485,
-      "eval_runtime": 677.2148,
-      "eval_samples_per_second": 4.151,
-      "eval_steps_per_second": 1.038,
-      "eval_translation_length": 52485,
-      "step": 18969
-    },
-    {
-      "epoch": 3.0,
-      "learning_rate": 4.738561798652854e-06,
-      "loss": 1.7842,
-      "step": 19000
-    },
-    {
-      "epoch": 3.08,
-      "learning_rate": 4.724507917234451e-06,
-      "loss": 1.8069,
-      "step": 19500
-    },
-    {
-      "epoch": 3.16,
-      "learning_rate": 4.710108094225704e-06,
-      "loss": 1.7776,
-      "step": 20000
-    },
-    {
-      "epoch": 3.24,
-      "learning_rate": 4.695364568997228e-06,
-      "loss": 1.8232,
-      "step": 20500
-    },
-    {
-      "epoch": 3.32,
-      "learning_rate": 4.680279634370071e-06,
-      "loss": 1.8125,
-      "step": 21000
-    },
-    {
-      "epoch": 3.4,
-      "learning_rate": 4.664855636259134e-06,
-      "loss": 1.841,
-      "step": 21500
-    },
-    {
-      "epoch": 3.48,
-      "learning_rate": 4.649094973308358e-06,
-      "loss": 1.8519,
-      "step": 22000
-    },
-    {
-      "epoch": 3.56,
-      "learning_rate": 4.633000096517698e-06,
-      "loss": 1.8293,
-      "step": 22500
-    },
-    {
-      "epoch": 3.64,
-      "learning_rate": 4.61657350886196e-06,
-      "loss": 1.8121,
-      "step": 23000
-    },
-    {
-      "epoch": 3.72,
-      "learning_rate": 4.5998177649015565e-06,
-      "loss": 1.7916,
-      "step": 23500
-    },
-    {
-      "epoch": 3.8,
-      "learning_rate": 4.582735470385229e-06,
-      "loss": 1.8352,
-      "step": 24000
-    },
-    {
-      "epoch": 3.87,
-      "learning_rate": 4.56532928184483e-06,
-      "loss": 1.7778,
-      "step": 24500
-    },
-    {
-      "epoch": 3.95,
-      "learning_rate": 4.547601906182184e-06,
-      "loss": 1.815,
-      "step": 25000
-    },
-    {
-      "epoch": 4.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5691736936569214,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52469,
-      "eval_runtime": 676.8094,
-      "eval_samples_per_second": 4.153,
-      "eval_steps_per_second": 1.039,
-      "eval_translation_length": 52469,
-      "step": 25292
-    },
-    {
-      "epoch": 4.03,
-      "learning_rate": 4.529556100248137e-06,
-      "loss": 1.8259,
-      "step": 25500
-    },
-    {
-      "epoch": 4.11,
-      "learning_rate": 4.511194670413822e-06,
-      "loss": 1.8127,
-      "step": 26000
-    },
-    {
-      "epoch": 4.19,
-      "learning_rate": 4.49252047213423e-06,
-      "loss": 1.7847,
-      "step": 26500
-    },
-    {
-      "epoch": 4.27,
-      "learning_rate": 4.473536409504151e-06,
-      "loss": 1.8137,
-      "step": 27000
-    },
-    {
-      "epoch": 4.35,
-      "learning_rate": 4.454245434806545e-06,
-      "loss": 1.8443,
-      "step": 27500
-    },
-    {
-      "epoch": 4.43,
-      "learning_rate": 4.4346505480534205e-06,
-      "loss": 1.7972,
-      "step": 28000
-    },
-    {
-      "epoch": 4.51,
-      "learning_rate": 4.4147547965192934e-06,
-      "loss": 1.8035,
-      "step": 28500
-    },
-    {
-      "epoch": 4.59,
-      "learning_rate": 4.394561274267293e-06,
-      "loss": 1.7983,
-      "step": 29000
-    },
-    {
-      "epoch": 4.67,
-      "learning_rate": 4.374073121667992e-06,
-      "loss": 1.8227,
-      "step": 29500
-    },
-    {
-      "epoch": 4.74,
-      "learning_rate": 4.3532935249110366e-06,
-      "loss": 1.7775,
-      "step": 30000
-    },
-    {
-      "epoch": 4.82,
-      "learning_rate": 4.3322257155096496e-06,
-      "loss": 1.8181,
-      "step": 30500
-    },
-    {
-      "epoch": 4.9,
-      "learning_rate": 4.310872969798085e-06,
-      "loss": 1.8333,
-      "step": 31000
-    },
-    {
-      "epoch": 4.98,
-      "learning_rate": 4.289238608422115e-06,
-      "loss": 1.8452,
-      "step": 31500
-    },
-    {
-      "epoch": 5.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.56978178024292,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52449,
-      "eval_runtime": 676.5149,
-      "eval_samples_per_second": 4.155,
-      "eval_steps_per_second": 1.039,
-      "eval_translation_length": 52449,
-      "step": 31615
-    },
-    {
-      "epoch": 5.06,
-      "learning_rate": 4.267325995822624e-06,
-      "loss": 1.8142,
-      "step": 32000
-    },
-    {
-      "epoch": 5.14,
-      "learning_rate": 4.2451385397123864e-06,
-      "loss": 1.8047,
-      "step": 32500
-    },
-    {
-      "epoch": 5.22,
-      "learning_rate": 4.222679690546128e-06,
-      "loss": 1.8006,
-      "step": 33000
-    },
-    {
-      "epoch": 5.3,
-      "learning_rate": 4.199952940983926e-06,
-      "loss": 1.7971,
-      "step": 33500
-    },
-    {
-      "epoch": 5.38,
-      "learning_rate": 4.176961825348059e-06,
-      "loss": 1.825,
-      "step": 34000
-    },
-    {
-      "epoch": 5.46,
-      "learning_rate": 4.1537099190733656e-06,
-      "loss": 1.8121,
-      "step": 34500
-    },
-    {
-      "epoch": 5.54,
-      "learning_rate": 4.130200838151217e-06,
-      "loss": 1.8179,
-      "step": 35000
-    },
-    {
-      "epoch": 5.61,
-      "learning_rate": 4.106438238567183e-06,
-      "loss": 1.8005,
-      "step": 35500
-    },
-    {
-      "epoch": 5.69,
-      "learning_rate": 4.08242581573247e-06,
-      "loss": 1.8367,
-      "step": 36000
-    },
-    {
-      "epoch": 5.77,
-      "learning_rate": 4.058167303909241e-06,
-      "loss": 1.8062,
-      "step": 36500
-    },
-    {
-      "epoch": 5.85,
-      "learning_rate": 4.033666475629881e-06,
-      "loss": 1.8092,
-      "step": 37000
-    },
-    {
-      "epoch": 5.93,
-      "learning_rate": 4.008927141110319e-06,
-      "loss": 1.7638,
-      "step": 37500
-    },
-    {
-      "epoch": 6.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5709949731826782,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52568,
-      "eval_runtime": 677.5535,
-      "eval_samples_per_second": 4.149,
-      "eval_steps_per_second": 1.038,
-      "eval_translation_length": 52568,
-      "step": 37938
-    },
-    {
-      "epoch": 6.01,
-      "learning_rate": 3.9839531476574855e-06,
-      "loss": 1.801,
-      "step": 38000
-    },
-    {
-      "epoch": 6.09,
-      "learning_rate": 3.958748379071004e-06,
-      "loss": 1.7813,
-      "step": 38500
-    },
-    {
-      "epoch": 6.17,
-      "learning_rate": 3.933316755039209e-06,
-      "loss": 1.7742,
-      "step": 39000
-    },
-    {
-      "epoch": 6.25,
-      "learning_rate": 3.9076622305295755e-06,
-      "loss": 1.7852,
-      "step": 39500
-    },
-    {
-      "epoch": 6.33,
-      "learning_rate": 3.88178879517367e-06,
-      "loss": 1.8312,
-      "step": 40000
-    },
-    {
-      "epoch": 6.41,
-      "learning_rate": 3.855700472646708e-06,
-      "loss": 1.8144,
-      "step": 40500
-    },
-    {
-      "epoch": 6.48,
-      "learning_rate": 3.82940132004182e-06,
-      "loss": 1.7856,
-      "step": 41000
-    },
-    {
-      "epoch": 6.56,
-      "learning_rate": 3.8028954272391116e-06,
-      "loss": 1.8139,
-      "step": 41500
-    },
-    {
-      "epoch": 6.64,
-      "learning_rate": 3.7761869162696334e-06,
-      "loss": 1.8018,
-      "step": 42000
-    },
-    {
-      "epoch": 6.72,
-      "learning_rate": 3.7492799406743512e-06,
-      "loss": 1.7771,
-      "step": 42500
-    },
-    {
-      "epoch": 6.8,
-      "learning_rate": 3.722178684858209e-06,
-      "loss": 1.8217,
-      "step": 43000
-    },
-    {
-      "epoch": 6.88,
-      "learning_rate": 3.6948873634394e-06,
-      "loss": 1.8276,
-      "step": 43500
-    },
-    {
-      "epoch": 6.96,
-      "learning_rate": 3.667410220593933e-06,
-      "loss": 1.8267,
-      "step": 44000
-    },
-    {
-      "epoch": 7.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.571208119392395,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52554,
-      "eval_runtime": 677.3503,
-      "eval_samples_per_second": 4.15,
-      "eval_steps_per_second": 1.038,
-      "eval_translation_length": 52554,
-      "step": 44261
-    },
-    {
-      "epoch": 7.04,
-      "learning_rate": 3.639751529395606e-06,
-      "loss": 1.7645,
-      "step": 44500
-    },
-    {
-      "epoch": 7.12,
-      "learning_rate": 3.611915591151483e-06,
-      "loss": 1.8167,
-      "step": 45000
-    },
-    {
-      "epoch": 7.2,
-      "learning_rate": 3.5839067347329844e-06,
-      "loss": 1.808,
-      "step": 45500
-    },
-    {
-      "epoch": 7.28,
-      "learning_rate": 3.5557293159026845e-06,
-      "loss": 1.7742,
-      "step": 46000
-    },
-    {
-      "epoch": 7.35,
-      "learning_rate": 3.5273877166369326e-06,
-      "loss": 1.784,
-      "step": 46500
-    },
-    {
-      "epoch": 7.43,
-      "learning_rate": 3.4988863444443942e-06,
-      "loss": 1.7732,
-      "step": 47000
-    },
-    {
-      "epoch": 7.51,
-      "learning_rate": 3.4702296316806243e-06,
-      "loss": 1.8029,
-      "step": 47500
-    },
-    {
-      "epoch": 7.59,
-      "learning_rate": 3.4414220348587744e-06,
-      "loss": 1.8167,
-      "step": 48000
-    },
-    {
-      "epoch": 7.67,
-      "learning_rate": 3.412468033956543e-06,
-      "loss": 1.8037,
-      "step": 48500
-    },
-    {
-      "epoch": 7.75,
-      "learning_rate": 3.3833721317194756e-06,
-      "loss": 1.7689,
-      "step": 49000
-    },
-    {
-      "epoch": 7.83,
-      "learning_rate": 3.3541388529607303e-06,
-      "loss": 1.8414,
-      "step": 49500
-    },
-    {
-      "epoch": 7.91,
-      "learning_rate": 3.324772743857404e-06,
-      "loss": 1.795,
-      "step": 50000
-    },
-    {
-      "epoch": 7.99,
-      "learning_rate": 3.2952783712435406e-06,
-      "loss": 1.8108,
-      "step": 50500
-    },
-    {
-      "epoch": 8.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5723803043365479,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52356,
-      "eval_runtime": 677.354,
-      "eval_samples_per_second": 4.15,
-      "eval_steps_per_second": 1.038,
-      "eval_translation_length": 52356,
-      "step": 50584
-    },
-    {
-      "epoch": 8.07,
-      "learning_rate": 3.265660321899923e-06,
-      "loss": 1.8002,
-      "step": 51000
-    },
-    {
-      "epoch": 8.14,
-      "learning_rate": 3.235923201840768e-06,
-      "loss": 1.7785,
-      "step": 51500
-    },
-    {
-      "epoch": 8.22,
-      "learning_rate": 3.2060716355974274e-06,
-      "loss": 1.7734,
-      "step": 52000
-    },
-    {
-      "epoch": 8.3,
-      "learning_rate": 3.1761102654992106e-06,
-      "loss": 1.8028,
-      "step": 52500
-    },
-    {
-      "epoch": 8.38,
-      "learning_rate": 3.1460437509514345e-06,
-      "loss": 1.7929,
-      "step": 53000
-    },
-    {
-      "epoch": 8.46,
-      "learning_rate": 3.115876767710828e-06,
-      "loss": 1.8039,
-      "step": 53500
-    },
-    {
-      "epoch": 8.54,
-      "learning_rate": 3.0856140071583806e-06,
-      "loss": 1.8066,
-      "step": 54000
-    },
-    {
-      "epoch": 8.62,
-      "learning_rate": 3.0552601755697765e-06,
-      "loss": 1.7612,
-      "step": 54500
-    },
-    {
-      "epoch": 8.7,
-      "learning_rate": 3.024819993383493e-06,
-      "loss": 1.8281,
-      "step": 55000
-    },
-    {
-      "epoch": 8.78,
-      "learning_rate": 2.9942981944667193e-06,
-      "loss": 1.7766,
-      "step": 55500
-    },
-    {
-      "epoch": 8.86,
-      "learning_rate": 2.963699525379166e-06,
-      "loss": 1.8176,
-      "step": 56000
-    },
-    {
-      "epoch": 8.94,
-      "learning_rate": 2.933028744634912e-06,
-      "loss": 1.79,
-      "step": 56500
-    },
-    {
-      "epoch": 9.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5721025466918945,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52640,
-      "eval_runtime": 678.5345,
-      "eval_samples_per_second": 4.143,
-      "eval_steps_per_second": 1.036,
-      "eval_translation_length": 52640,
-      "step": 56907
-    },
-    {
-      "epoch": 9.01,
-      "learning_rate": 2.9022906219623958e-06,
-      "loss": 1.8089,
-      "step": 57000
-    },
-    {
-      "epoch": 9.09,
-      "learning_rate": 2.871489937562647e-06,
-      "loss": 1.7695,
-      "step": 57500
-    },
-    {
-      "epoch": 9.17,
-      "learning_rate": 2.8406314813659073e-06,
-      "loss": 1.7845,
-      "step": 58000
-    },
-    {
-      "epoch": 9.25,
-      "learning_rate": 2.8097200522867294e-06,
-      "loss": 1.7954,
-      "step": 58500
-    },
-    {
-      "epoch": 9.33,
-      "learning_rate": 2.7787604574776745e-06,
-      "loss": 1.8204,
-      "step": 59000
-    },
-    {
-      "epoch": 9.41,
-      "learning_rate": 2.747757511581739e-06,
-      "loss": 1.7786,
-      "step": 59500
-    },
-    {
-      "epoch": 9.49,
-      "learning_rate": 2.716716035983611e-06,
-      "loss": 1.7995,
-      "step": 60000
-    },
-    {
-      "epoch": 9.57,
-      "learning_rate": 2.685640858059876e-06,
-      "loss": 1.8058,
-      "step": 60500
-    },
-    {
-      "epoch": 9.65,
-      "learning_rate": 2.6545368104282955e-06,
-      "loss": 1.7961,
-      "step": 61000
-    },
-    {
-      "epoch": 9.73,
-      "learning_rate": 2.623408730196268e-06,
-      "loss": 1.7866,
-      "step": 61500
-    },
-    {
-      "epoch": 9.81,
-      "learning_rate": 2.592261458208591e-06,
-      "loss": 1.7643,
-      "step": 62000
-    },
-    {
-      "epoch": 9.88,
-      "learning_rate": 2.5610998382946463e-06,
-      "loss": 1.7679,
-      "step": 62500
-    },
-    {
-      "epoch": 9.96,
-      "learning_rate": 2.529928716515112e-06,
-      "loss": 1.8195,
-      "step": 63000
-    },
-    {
-      "epoch": 10.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5732102394104004,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52485,
-      "eval_runtime": 677.793,
-      "eval_samples_per_second": 4.147,
-      "eval_steps_per_second": 1.037,
-      "eval_translation_length": 52485,
-      "step": 63230
-    },
-    {
-      "epoch": 10.04,
-      "learning_rate": 2.498752940408342e-06,
-      "loss": 1.7938,
-      "step": 63500
-    },
-    {
-      "epoch": 10.12,
-      "learning_rate": 2.4675773582364977e-06,
-      "loss": 1.7688,
-      "step": 64000
-    },
-    {
-      "epoch": 10.2,
-      "learning_rate": 2.436406818231583e-06,
-      "loss": 1.7701,
-      "step": 64500
-    },
-    {
-      "epoch": 10.28,
-      "learning_rate": 2.4052461678414753e-06,
-      "loss": 1.7821,
-      "step": 65000
-    },
-    {
-      "epoch": 10.36,
-      "learning_rate": 2.37410025297608e-06,
-      "loss": 1.8251,
-      "step": 65500
-    },
-    {
-      "epoch": 10.44,
-      "learning_rate": 2.342973917253726e-06,
-      "loss": 1.7384,
-      "step": 66000
-    },
-    {
-      "epoch": 10.52,
-      "learning_rate": 2.3118720012479183e-06,
-      "loss": 1.8001,
-      "step": 66500
-    },
-    {
-      "epoch": 10.6,
-      "learning_rate": 2.280799341734556e-06,
-      "loss": 1.8386,
-      "step": 67000
-    },
-    {
-      "epoch": 10.68,
-      "learning_rate": 2.249760770939754e-06,
-      "loss": 1.8098,
-      "step": 67500
-    },
-    {
-      "epoch": 10.75,
-      "learning_rate": 2.218761115788362e-06,
-      "loss": 1.8059,
-      "step": 68000
-    },
-    {
-      "epoch": 10.83,
-      "learning_rate": 2.1878051971533093e-06,
-      "loss": 1.757,
-      "step": 68500
-    },
-    {
-      "epoch": 10.91,
-      "learning_rate": 2.156897829105898e-06,
-      "loss": 1.8037,
-      "step": 69000
-    },
-    {
-      "epoch": 10.99,
-      "learning_rate": 2.1260438181671446e-06,
-      "loss": 1.7714,
-      "step": 69500
-    },
-    {
-      "epoch": 11.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5735211372375488,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52469,
-      "eval_runtime": 678.026,
-      "eval_samples_per_second": 4.146,
-      "eval_steps_per_second": 1.037,
-      "eval_translation_length": 52469,
-      "step": 69553
-    },
-    {
-      "epoch": 11.07,
-      "learning_rate": 2.0952479625603017e-06,
-      "loss": 1.7783,
-      "step": 70000
-    },
-    {
-      "epoch": 11.15,
-      "learning_rate": 2.0645150514646657e-06,
-      "loss": 1.7443,
-      "step": 70500
-    },
-    {
-      "epoch": 11.23,
-      "learning_rate": 2.0338498642707977e-06,
-      "loss": 1.7678,
-      "step": 71000
-    },
-    {
-      "epoch": 11.31,
-      "learning_rate": 2.0032571698372577e-06,
-      "loss": 1.7786,
-      "step": 71500
-    },
-    {
-      "epoch": 11.39,
-      "learning_rate": 1.9727417257489874e-06,
-      "loss": 1.7768,
-      "step": 72000
-    },
-    {
-      "epoch": 11.47,
-      "learning_rate": 1.9423082775774337e-06,
-      "loss": 1.7953,
-      "step": 72500
-    },
-    {
-      "epoch": 11.55,
-      "learning_rate": 1.9119615581425524e-06,
-      "loss": 1.7715,
-      "step": 73000
-    },
-    {
-      "epoch": 11.62,
-      "learning_rate": 1.881706286776785e-06,
-      "loss": 1.8047,
-      "step": 73500
-    },
-    {
-      "epoch": 11.7,
-      "learning_rate": 1.8515471685911402e-06,
-      "loss": 1.7781,
-      "step": 74000
-    },
-    {
-      "epoch": 11.78,
-      "learning_rate": 1.821488893743488e-06,
-      "loss": 1.8197,
-      "step": 74500
-    },
-    {
-      "epoch": 11.86,
-      "learning_rate": 1.7915361367091677e-06,
-      "loss": 1.8159,
-      "step": 75000
-    },
-    {
-      "epoch": 11.94,
-      "learning_rate": 1.7616935555540475e-06,
-      "loss": 1.8004,
-      "step": 75500
-    },
-    {
-      "epoch": 12.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5739296674728394,
-      "eval_precisions": [
-        1.0,
-        1.0,
-        1.0,
-        1.0
-      ],
-      "eval_reference_length": 52457,
-      "eval_runtime": 676.9983,
-      "eval_samples_per_second": 4.152,
-      "eval_steps_per_second": 1.038,
-      "eval_translation_length": 52457,
-      "step": 75876
-    },
-    {
-      "epoch": 12.02,
-      "learning_rate": 1.7319657912101309e-06,
-      "loss": 1.7871,
-      "step": 76000
-    },
-    {
-      "epoch": 12.1,
-      "learning_rate": 1.7023574667538268e-06,
-      "loss": 1.7728,
-      "step": 76500
-    },
-    {
-      "epoch": 12.18,
-      "learning_rate": 1.6728731866869999e-06,
-      "loss": 1.792,
-      "step": 77000
-    },
-    {
-      "epoch": 12.26,
-      "learning_rate": 1.6435175362209033e-06,
-      "loss": 1.8009,
-      "step": 77500
-    },
-    {
-      "epoch": 12.34,
-      "learning_rate": 1.6142950805631178e-06,
-      "loss": 1.751,
-      "step": 78000
-    },
-    {
-      "epoch": 12.41,
-      "learning_rate": 1.5852103642075995e-06,
-      "loss": 1.7877,
-      "step": 78500
-    },
-    {
-      "epoch": 12.49,
-      "learning_rate": 1.5562679102279453e-06,
-      "loss": 1.7936,
-      "step": 79000
-    },
-    {
-      "epoch": 12.57,
-      "learning_rate": 1.5274722195740005e-06,
-      "loss": 1.7884,
-      "step": 79500
-    },
-    {
-      "epoch": 12.65,
-      "learning_rate": 1.4988277703718882e-06,
-      "loss": 1.7617,
-      "step": 80000
-    },
-    {
-      "epoch": 12.73,
-      "learning_rate": 1.4703390172276072e-06,
-      "loss": 1.7916,
-      "step": 80500
-    },
-    {
-      "epoch": 12.81,
-      "learning_rate": 1.4420103905342767e-06,
-      "loss": 1.7773,
-      "step": 81000
-    },
-    {
-      "epoch": 12.89,
-      "learning_rate": 1.4138462957831472e-06,
-      "loss": 1.7798,
-      "step": 81500
-    },
-    {
-      "epoch": 12.97,
-      "learning_rate": 1.3858511128784937e-06,
-      "loss": 1.7658,
-      "step": 82000
-    },
-    {
-      "epoch": 13.0,
-      "eval_bleu": 1.0,
-      "eval_brevity_penalty": 1.0,
-      "eval_length_ratio": 1.0,
-      "eval_loss": 1.5740926265716553,
+      "eval_loss": 2.2135026454925537,
       "eval_precisions": [
         1.0,
         1.0,
         1.0,
         1.0
       ],
-      "eval_reference_length": 52468,
-      "eval_runtime": 677.6429,
-      "eval_samples_per_second": 4.148,
-      "eval_steps_per_second": 1.037,
-      "eval_translation_length": 52468,
-      "step": 82199
+      "eval_reference_length": 53394,
+      "eval_runtime": 2937.3471,
+      "eval_samples_per_second": 0.957,
+      "eval_steps_per_second": 0.12,
+      "eval_translation_length": 53394,
+      "step": 1581
     }
   ],
   "logging_steps": 500,
-  "max_steps": 126460,
-  "num_train_epochs": 20,
+  "max_steps": 15810,
+  "num_train_epochs": 10,
   "save_steps": 500,
-  "total_flos": 2.251365766099108e+17,
+  "total_flos": 771945142419456.0,
   "trial_name": null,
   "trial_params": null
 }

last-checkpoint/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ce6df3970ff39f84beab7b635dd3c941539643a1d32dedf54263683eef40519
-size 4664
+oid sha256:da68e390908bbf17cdb4f041431915c87c2f3f6b8699cebf8dcb70aae5296d20
+size 6648

last-checkpoint/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)