first commit

all_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 400.0,
+    "eval_loss": 4.94140625,
+    "eval_runtime": 3.7873,
+    "eval_samples": 258,
+    "eval_samples_per_second": 68.123,
+    "eval_steps_per_second": 8.713,
+    "perplexity": 139.96693973881514,
+    "train_loss": 0.4328666927939967,
+    "train_runtime": 22235.2418,
+    "train_samples": 603,
+    "train_samples_per_second": 10.848,
+    "train_steps_per_second": 0.684
+}

config.json

@@ -0,0 +1,39 @@
+{
+  "_name_or_path": "gpt2",
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50256,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 768,
+  "n_head": 12,
+  "n_inner": null,
+  "n_layer": 12,
+  "n_positions": 1024,
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "text-generation": {
+      "do_sample": true,
+      "max_length": 50
+    }
+  },
+  "torch_dtype": "float16",
+  "transformers_version": "4.13.0",
+  "use_cache": false,
+  "vocab_size": 50257
+}

eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 400.0,
+    "eval_loss": 4.94140625,
+    "eval_runtime": 3.7873,
+    "eval_samples": 258,
+    "eval_samples_per_second": 68.123,
+    "eval_steps_per_second": 8.713,
+    "perplexity": 139.96693973881514
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bd85b148ef6d5f2ea91cf65ea6ac8daaa7d88454616d84e7ba068352ca9c48b
+size 261495703

runs/Dec18_17-45-49_louis-gpu/1639845963.6867917/events.out.tfevents.1639845963.louis-gpu.3745350.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d948efad1252c97027797802079c99099782cae89c1630fcb9004ec4f17b291
+size 4679

runs/Dec18_17-45-49_louis-gpu/events.out.tfevents.1639845963.louis-gpu.3745350.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ca8ac72d52221b7613f16a8a090f22d5822ba7f69420f5a31134d69cb07eb5e
+size 29150

runs/Dec18_17-45-49_louis-gpu/events.out.tfevents.1639868204.louis-gpu.3745350.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7036de84b16ccde7834779e0940820e8d4fa71b7f79ad177f33011d5a1307c75
+size 311

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<|endoftext|>", "bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "add_prefix_space": false, "model_max_length": 1024, "special_tokens_map_file": null, "name_or_path": "gpt2", "tokenizer_class": "GPT2Tokenizer"}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 400.0,
+    "train_loss": 0.4328666927939967,
+    "train_runtime": 22235.2418,
+    "train_samples": 603,
+    "train_samples_per_second": 10.848,
+    "train_steps_per_second": 0.684
+}

trainer_state.json

@@ -0,0 +1,813 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 400.0,
+  "global_step": 15200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 5.26,
+      "eval_loss": 1.7080078125,
+      "eval_runtime": 3.7763,
+      "eval_samples_per_second": 68.321,
+      "eval_steps_per_second": 8.739,
+      "step": 200
+    },
+    {
+      "epoch": 10.53,
+      "eval_loss": 1.7001953125,
+      "eval_runtime": 3.7858,
+      "eval_samples_per_second": 68.149,
+      "eval_steps_per_second": 8.717,
+      "step": 400
+    },
+    {
+      "epoch": 13.16,
+      "learning_rate": 5e-05,
+      "loss": 1.671,
+      "step": 500
+    },
+    {
+      "epoch": 15.79,
+      "eval_loss": 1.7138671875,
+      "eval_runtime": 3.7865,
+      "eval_samples_per_second": 68.137,
+      "eval_steps_per_second": 8.715,
+      "step": 600
+    },
+    {
+      "epoch": 21.05,
+      "eval_loss": 1.744140625,
+      "eval_runtime": 3.7879,
+      "eval_samples_per_second": 68.112,
+      "eval_steps_per_second": 8.712,
+      "step": 800
+    },
+    {
+      "epoch": 26.32,
+      "learning_rate": 5e-05,
+      "loss": 1.4438,
+      "step": 1000
+    },
+    {
+      "epoch": 26.32,
+      "eval_loss": 1.794921875,
+      "eval_runtime": 3.7869,
+      "eval_samples_per_second": 68.13,
+      "eval_steps_per_second": 8.714,
+      "step": 1000
+    },
+    {
+      "epoch": 31.58,
+      "eval_loss": 1.84375,
+      "eval_runtime": 3.7879,
+      "eval_samples_per_second": 68.112,
+      "eval_steps_per_second": 8.712,
+      "step": 1200
+    },
+    {
+      "epoch": 36.84,
+      "eval_loss": 1.896484375,
+      "eval_runtime": 3.7905,
+      "eval_samples_per_second": 68.066,
+      "eval_steps_per_second": 8.706,
+      "step": 1400
+    },
+    {
+      "epoch": 39.47,
+      "learning_rate": 5e-05,
+      "loss": 1.2806,
+      "step": 1500
+    },
+    {
+      "epoch": 42.11,
+      "eval_loss": 1.9619140625,
+      "eval_runtime": 3.7916,
+      "eval_samples_per_second": 68.044,
+      "eval_steps_per_second": 8.703,
+      "step": 1600
+    },
+    {
+      "epoch": 47.37,
+      "eval_loss": 2.01953125,
+      "eval_runtime": 3.7897,
+      "eval_samples_per_second": 68.08,
+      "eval_steps_per_second": 8.708,
+      "step": 1800
+    },
+    {
+      "epoch": 52.63,
+      "learning_rate": 5e-05,
+      "loss": 1.1433,
+      "step": 2000
+    },
+    {
+      "epoch": 52.63,
+      "eval_loss": 2.068359375,
+      "eval_runtime": 3.7891,
+      "eval_samples_per_second": 68.091,
+      "eval_steps_per_second": 8.709,
+      "step": 2000
+    },
+    {
+      "epoch": 57.89,
+      "eval_loss": 2.1171875,
+      "eval_runtime": 3.7902,
+      "eval_samples_per_second": 68.069,
+      "eval_steps_per_second": 8.707,
+      "step": 2200
+    },
+    {
+      "epoch": 63.16,
+      "eval_loss": 2.1953125,
+      "eval_runtime": 3.7898,
+      "eval_samples_per_second": 68.077,
+      "eval_steps_per_second": 8.708,
+      "step": 2400
+    },
+    {
+      "epoch": 65.79,
+      "learning_rate": 5e-05,
+      "loss": 1.027,
+      "step": 2500
+    },
+    {
+      "epoch": 68.42,
+      "eval_loss": 2.25,
+      "eval_runtime": 3.7881,
+      "eval_samples_per_second": 68.108,
+      "eval_steps_per_second": 8.711,
+      "step": 2600
+    },
+    {
+      "epoch": 73.68,
+      "eval_loss": 2.291015625,
+      "eval_runtime": 3.7876,
+      "eval_samples_per_second": 68.118,
+      "eval_steps_per_second": 8.713,
+      "step": 2800
+    },
+    {
+      "epoch": 78.95,
+      "learning_rate": 5e-05,
+      "loss": 0.9216,
+      "step": 3000
+    },
+    {
+      "epoch": 78.95,
+      "eval_loss": 2.34765625,
+      "eval_runtime": 3.7885,
+      "eval_samples_per_second": 68.101,
+      "eval_steps_per_second": 8.711,
+      "step": 3000
+    },
+    {
+      "epoch": 84.21,
+      "eval_loss": 2.423828125,
+      "eval_runtime": 3.7907,
+      "eval_samples_per_second": 68.062,
+      "eval_steps_per_second": 8.706,
+      "step": 3200
+    },
+    {
+      "epoch": 89.47,
+      "eval_loss": 2.482421875,
+      "eval_runtime": 3.7903,
+      "eval_samples_per_second": 68.068,
+      "eval_steps_per_second": 8.706,
+      "step": 3400
+    },
+    {
+      "epoch": 92.11,
+      "learning_rate": 5e-05,
+      "loss": 0.8209,
+      "step": 3500
+    },
+    {
+      "epoch": 94.74,
+      "eval_loss": 2.529296875,
+      "eval_runtime": 3.7863,
+      "eval_samples_per_second": 68.14,
+      "eval_steps_per_second": 8.716,
+      "step": 3600
+    },
+    {
+      "epoch": 100.0,
+      "eval_loss": 2.5859375,
+      "eval_runtime": 3.785,
+      "eval_samples_per_second": 68.164,
+      "eval_steps_per_second": 8.719,
+      "step": 3800
+    },
+    {
+      "epoch": 105.26,
+      "learning_rate": 5e-05,
+      "loss": 0.7231,
+      "step": 4000
+    },
+    {
+      "epoch": 105.26,
+      "eval_loss": 2.6640625,
+      "eval_runtime": 3.7856,
+      "eval_samples_per_second": 68.153,
+      "eval_steps_per_second": 8.717,
+      "step": 4000
+    },
+    {
+      "epoch": 110.53,
+      "eval_loss": 2.703125,
+      "eval_runtime": 3.7862,
+      "eval_samples_per_second": 68.142,
+      "eval_steps_per_second": 8.716,
+      "step": 4200
+    },
+    {
+      "epoch": 115.79,
+      "eval_loss": 2.78515625,
+      "eval_runtime": 3.7894,
+      "eval_samples_per_second": 68.084,
+      "eval_steps_per_second": 8.708,
+      "step": 4400
+    },
+    {
+      "epoch": 118.42,
+      "learning_rate": 5e-05,
+      "loss": 0.6281,
+      "step": 4500
+    },
+    {
+      "epoch": 121.05,
+      "eval_loss": 2.84375,
+      "eval_runtime": 3.7883,
+      "eval_samples_per_second": 68.105,
+      "eval_steps_per_second": 8.711,
+      "step": 4600
+    },
+    {
+      "epoch": 126.32,
+      "eval_loss": 2.921875,
+      "eval_runtime": 3.79,
+      "eval_samples_per_second": 68.074,
+      "eval_steps_per_second": 8.707,
+      "step": 4800
+    },
+    {
+      "epoch": 131.58,
+      "learning_rate": 5e-05,
+      "loss": 0.5384,
+      "step": 5000
+    },
+    {
+      "epoch": 131.58,
+      "eval_loss": 2.994140625,
+      "eval_runtime": 3.7895,
+      "eval_samples_per_second": 68.082,
+      "eval_steps_per_second": 8.708,
+      "step": 5000
+    },
+    {
+      "epoch": 136.84,
+      "eval_loss": 3.048828125,
+      "eval_runtime": 3.7912,
+      "eval_samples_per_second": 68.053,
+      "eval_steps_per_second": 8.704,
+      "step": 5200
+    },
+    {
+      "epoch": 142.11,
+      "eval_loss": 3.107421875,
+      "eval_runtime": 3.7872,
+      "eval_samples_per_second": 68.123,
+      "eval_steps_per_second": 8.713,
+      "step": 5400
+    },
+    {
+      "epoch": 144.74,
+      "learning_rate": 5e-05,
+      "loss": 0.4574,
+      "step": 5500
+    },
+    {
+      "epoch": 147.37,
+      "eval_loss": 3.169921875,
+      "eval_runtime": 3.7886,
+      "eval_samples_per_second": 68.1,
+      "eval_steps_per_second": 8.71,
+      "step": 5600
+    },
+    {
+      "epoch": 152.63,
+      "eval_loss": 3.2265625,
+      "eval_runtime": 3.7924,
+      "eval_samples_per_second": 68.03,
+      "eval_steps_per_second": 8.702,
+      "step": 5800
+    },
+    {
+      "epoch": 157.89,
+      "learning_rate": 5e-05,
+      "loss": 0.3848,
+      "step": 6000
+    },
+    {
+      "epoch": 157.89,
+      "eval_loss": 3.291015625,
+      "eval_runtime": 3.7859,
+      "eval_samples_per_second": 68.148,
+      "eval_steps_per_second": 8.717,
+      "step": 6000
+    },
+    {
+      "epoch": 163.16,
+      "eval_loss": 3.376953125,
+      "eval_runtime": 3.7886,
+      "eval_samples_per_second": 68.099,
+      "eval_steps_per_second": 8.71,
+      "step": 6200
+    },
+    {
+      "epoch": 168.42,
+      "eval_loss": 3.408203125,
+      "eval_runtime": 3.7885,
+      "eval_samples_per_second": 68.1,
+      "eval_steps_per_second": 8.71,
+      "step": 6400
+    },
+    {
+      "epoch": 171.05,
+      "learning_rate": 5e-05,
+      "loss": 0.3224,
+      "step": 6500
+    },
+    {
+      "epoch": 173.68,
+      "eval_loss": 3.4765625,
+      "eval_runtime": 3.7922,
+      "eval_samples_per_second": 68.034,
+      "eval_steps_per_second": 8.702,
+      "step": 6600
+    },
+    {
+      "epoch": 178.95,
+      "eval_loss": 3.529296875,
+      "eval_runtime": 3.7898,
+      "eval_samples_per_second": 68.077,
+      "eval_steps_per_second": 8.708,
+      "step": 6800
+    },
+    {
+      "epoch": 184.21,
+      "learning_rate": 5e-05,
+      "loss": 0.2697,
+      "step": 7000
+    },
+    {
+      "epoch": 184.21,
+      "eval_loss": 3.591796875,
+      "eval_runtime": 3.7854,
+      "eval_samples_per_second": 68.157,
+      "eval_steps_per_second": 8.718,
+      "step": 7000
+    },
+    {
+      "epoch": 189.47,
+      "eval_loss": 3.634765625,
+      "eval_runtime": 3.7918,
+      "eval_samples_per_second": 68.041,
+      "eval_steps_per_second": 8.703,
+      "step": 7200
+    },
+    {
+      "epoch": 194.74,
+      "eval_loss": 3.68359375,
+      "eval_runtime": 3.7891,
+      "eval_samples_per_second": 68.09,
+      "eval_steps_per_second": 8.709,
+      "step": 7400
+    },
+    {
+      "epoch": 197.37,
+      "learning_rate": 5e-05,
+      "loss": 0.2258,
+      "step": 7500
+    },
+    {
+      "epoch": 200.0,
+      "eval_loss": 3.7265625,
+      "eval_runtime": 3.7895,
+      "eval_samples_per_second": 68.083,
+      "eval_steps_per_second": 8.708,
+      "step": 7600
+    },
+    {
+      "epoch": 205.26,
+      "eval_loss": 3.79296875,
+      "eval_runtime": 3.7901,
+      "eval_samples_per_second": 68.073,
+      "eval_steps_per_second": 8.707,
+      "step": 7800
+    },
+    {
+      "epoch": 210.53,
+      "learning_rate": 5e-05,
+      "loss": 0.1893,
+      "step": 8000
+    },
+    {
+      "epoch": 210.53,
+      "eval_loss": 3.828125,
+      "eval_runtime": 3.7891,
+      "eval_samples_per_second": 68.09,
+      "eval_steps_per_second": 8.709,
+      "step": 8000
+    },
+    {
+      "epoch": 215.79,
+      "eval_loss": 3.880859375,
+      "eval_runtime": 3.7907,
+      "eval_samples_per_second": 68.062,
+      "eval_steps_per_second": 8.706,
+      "step": 8200
+    },
+    {
+      "epoch": 221.05,
+      "eval_loss": 3.923828125,
+      "eval_runtime": 3.7895,
+      "eval_samples_per_second": 68.082,
+      "eval_steps_per_second": 8.708,
+      "step": 8400
+    },
+    {
+      "epoch": 223.68,
+      "learning_rate": 5e-05,
+      "loss": 0.1602,
+      "step": 8500
+    },
+    {
+      "epoch": 226.32,
+      "eval_loss": 3.974609375,
+      "eval_runtime": 3.7894,
+      "eval_samples_per_second": 68.084,
+      "eval_steps_per_second": 8.708,
+      "step": 8600
+    },
+    {
+      "epoch": 231.58,
+      "eval_loss": 4.00390625,
+      "eval_runtime": 3.7923,
+      "eval_samples_per_second": 68.032,
+      "eval_steps_per_second": 8.702,
+      "step": 8800
+    },
+    {
+      "epoch": 236.84,
+      "learning_rate": 5e-05,
+      "loss": 0.137,
+      "step": 9000
+    },
+    {
+      "epoch": 236.84,
+      "eval_loss": 4.046875,
+      "eval_runtime": 3.7922,
+      "eval_samples_per_second": 68.034,
+      "eval_steps_per_second": 8.702,
+      "step": 9000
+    },
+    {
+      "epoch": 242.11,
+      "eval_loss": 4.07421875,
+      "eval_runtime": 3.7901,
+      "eval_samples_per_second": 68.072,
+      "eval_steps_per_second": 8.707,
+      "step": 9200
+    },
+    {
+      "epoch": 247.37,
+      "eval_loss": 4.12109375,
+      "eval_runtime": 3.7896,
+      "eval_samples_per_second": 68.08,
+      "eval_steps_per_second": 8.708,
+      "step": 9400
+    },
+    {
+      "epoch": 250.0,
+      "learning_rate": 5e-05,
+      "loss": 0.1179,
+      "step": 9500
+    },
+    {
+      "epoch": 252.63,
+      "eval_loss": 4.15625,
+      "eval_runtime": 3.7912,
+      "eval_samples_per_second": 68.053,
+      "eval_steps_per_second": 8.704,
+      "step": 9600
+    },
+    {
+      "epoch": 257.89,
+      "eval_loss": 4.203125,
+      "eval_runtime": 3.7923,
+      "eval_samples_per_second": 68.032,
+      "eval_steps_per_second": 8.702,
+      "step": 9800
+    },
+    {
+      "epoch": 263.16,
+      "learning_rate": 5e-05,
+      "loss": 0.1024,
+      "step": 10000
+    },
+    {
+      "epoch": 263.16,
+      "eval_loss": 4.234375,
+      "eval_runtime": 3.7852,
+      "eval_samples_per_second": 68.159,
+      "eval_steps_per_second": 8.718,
+      "step": 10000
+    },
+    {
+      "epoch": 268.42,
+      "eval_loss": 4.2734375,
+      "eval_runtime": 3.7869,
+      "eval_samples_per_second": 68.129,
+      "eval_steps_per_second": 8.714,
+      "step": 10200
+    },
+    {
+      "epoch": 273.68,
+      "eval_loss": 4.3046875,
+      "eval_runtime": 3.7892,
+      "eval_samples_per_second": 68.088,
+      "eval_steps_per_second": 8.709,
+      "step": 10400
+    },
+    {
+      "epoch": 276.32,
+      "learning_rate": 5e-05,
+      "loss": 0.0901,
+      "step": 10500
+    },
+    {
+      "epoch": 278.95,
+      "eval_loss": 4.3125,
+      "eval_runtime": 3.7869,
+      "eval_samples_per_second": 68.129,
+      "eval_steps_per_second": 8.714,
+      "step": 10600
+    },
+    {
+      "epoch": 284.21,
+      "eval_loss": 4.375,
+      "eval_runtime": 3.7872,
+      "eval_samples_per_second": 68.125,
+      "eval_steps_per_second": 8.714,
+      "step": 10800
+    },
+    {
+      "epoch": 289.47,
+      "learning_rate": 5e-05,
+      "loss": 0.0796,
+      "step": 11000
+    },
+    {
+      "epoch": 289.47,
+      "eval_loss": 4.390625,
+      "eval_runtime": 3.7843,
+      "eval_samples_per_second": 68.177,
+      "eval_steps_per_second": 8.72,
+      "step": 11000
+    },
+    {
+      "epoch": 294.74,
+      "eval_loss": 4.4375,
+      "eval_runtime": 3.7881,
+      "eval_samples_per_second": 68.107,
+      "eval_steps_per_second": 8.711,
+      "step": 11200
+    },
+    {
+      "epoch": 300.0,
+      "eval_loss": 4.453125,
+      "eval_runtime": 3.7869,
+      "eval_samples_per_second": 68.129,
+      "eval_steps_per_second": 8.714,
+      "step": 11400
+    },
+    {
+      "epoch": 302.63,
+      "learning_rate": 5e-05,
+      "loss": 0.0706,
+      "step": 11500
+    },
+    {
+      "epoch": 305.26,
+      "eval_loss": 4.5078125,
+      "eval_runtime": 3.7854,
+      "eval_samples_per_second": 68.156,
+      "eval_steps_per_second": 8.718,
+      "step": 11600
+    },
+    {
+      "epoch": 310.53,
+      "eval_loss": 4.515625,
+      "eval_runtime": 3.787,
+      "eval_samples_per_second": 68.128,
+      "eval_steps_per_second": 8.714,
+      "step": 11800
+    },
+    {
+      "epoch": 315.79,
+      "learning_rate": 5e-05,
+      "loss": 0.0631,
+      "step": 12000
+    },
+    {
+      "epoch": 315.79,
+      "eval_loss": 4.53515625,
+      "eval_runtime": 3.7837,
+      "eval_samples_per_second": 68.187,
+      "eval_steps_per_second": 8.722,
+      "step": 12000
+    },
+    {
+      "epoch": 321.05,
+      "eval_loss": 4.5859375,
+      "eval_runtime": 3.7869,
+      "eval_samples_per_second": 68.13,
+      "eval_steps_per_second": 8.714,
+      "step": 12200
+    },
+    {
+      "epoch": 326.32,
+      "eval_loss": 4.609375,
+      "eval_runtime": 3.788,
+      "eval_samples_per_second": 68.11,
+      "eval_steps_per_second": 8.712,
+      "step": 12400
+    },
+    {
+      "epoch": 328.95,
+      "learning_rate": 5e-05,
+      "loss": 0.0573,
+      "step": 12500
+    },
+    {
+      "epoch": 331.58,
+      "eval_loss": 4.63671875,
+      "eval_runtime": 3.7891,
+      "eval_samples_per_second": 68.09,
+      "eval_steps_per_second": 8.709,
+      "step": 12600
+    },
+    {
+      "epoch": 336.84,
+      "eval_loss": 4.63671875,
+      "eval_runtime": 3.7855,
+      "eval_samples_per_second": 68.154,
+      "eval_steps_per_second": 8.717,
+      "step": 12800
+    },
+    {
+      "epoch": 342.11,
+      "learning_rate": 5e-05,
+      "loss": 0.0521,
+      "step": 13000
+    },
+    {
+      "epoch": 342.11,
+      "eval_loss": 4.6640625,
+      "eval_runtime": 3.7838,
+      "eval_samples_per_second": 68.185,
+      "eval_steps_per_second": 8.721,
+      "step": 13000
+    },
+    {
+      "epoch": 347.37,
+      "eval_loss": 4.70703125,
+      "eval_runtime": 3.7834,
+      "eval_samples_per_second": 68.192,
+      "eval_steps_per_second": 8.722,
+      "step": 13200
+    },
+    {
+      "epoch": 352.63,
+      "eval_loss": 4.69921875,
+      "eval_runtime": 3.789,
+      "eval_samples_per_second": 68.092,
+      "eval_steps_per_second": 8.709,
+      "step": 13400
+    },
+    {
+      "epoch": 355.26,
+      "learning_rate": 5e-05,
+      "loss": 0.0475,
+      "step": 13500
+    },
+    {
+      "epoch": 357.89,
+      "eval_loss": 4.75390625,
+      "eval_runtime": 3.7901,
+      "eval_samples_per_second": 68.073,
+      "eval_steps_per_second": 8.707,
+      "step": 13600
+    },
+    {
+      "epoch": 363.16,
+      "eval_loss": 4.765625,
+      "eval_runtime": 3.7877,
+      "eval_samples_per_second": 68.116,
+      "eval_steps_per_second": 8.712,
+      "step": 13800
+    },
+    {
+      "epoch": 368.42,
+      "learning_rate": 5e-05,
+      "loss": 0.0437,
+      "step": 14000
+    },
+    {
+      "epoch": 368.42,
+      "eval_loss": 4.80078125,
+      "eval_runtime": 3.7858,
+      "eval_samples_per_second": 68.15,
+      "eval_steps_per_second": 8.717,
+      "step": 14000
+    },
+    {
+      "epoch": 373.68,
+      "eval_loss": 4.83203125,
+      "eval_runtime": 3.7888,
+      "eval_samples_per_second": 68.095,
+      "eval_steps_per_second": 8.71,
+      "step": 14200
+    },
+    {
+      "epoch": 378.95,
+      "eval_loss": 4.8515625,
+      "eval_runtime": 3.7901,
+      "eval_samples_per_second": 68.073,
+      "eval_steps_per_second": 8.707,
+      "step": 14400
+    },
+    {
+      "epoch": 381.58,
+      "learning_rate": 5e-05,
+      "loss": 0.0399,
+      "step": 14500
+    },
+    {
+      "epoch": 384.21,
+      "eval_loss": 4.86328125,
+      "eval_runtime": 3.7938,
+      "eval_samples_per_second": 68.006,
+      "eval_steps_per_second": 8.698,
+      "step": 14600
+    },
+    {
+      "epoch": 389.47,
+      "eval_loss": 4.89453125,
+      "eval_runtime": 3.7887,
+      "eval_samples_per_second": 68.098,
+      "eval_steps_per_second": 8.71,
+      "step": 14800
+    },
+    {
+      "epoch": 394.74,
+      "learning_rate": 5e-05,
+      "loss": 0.0367,
+      "step": 15000
+    },
+    {
+      "epoch": 394.74,
+      "eval_loss": 4.90625,
+      "eval_runtime": 3.7864,
+      "eval_samples_per_second": 68.138,
+      "eval_steps_per_second": 8.715,
+      "step": 15000
+    },
+    {
+      "epoch": 400.0,
+      "eval_loss": 4.94140625,
+      "eval_runtime": 3.791,
+      "eval_samples_per_second": 68.057,
+      "eval_steps_per_second": 8.705,
+      "step": 15200
+    },
+    {
+      "epoch": 400.0,
+      "step": 15200,
+      "total_flos": 1.2604727427386573e+17,
+      "train_loss": 0.4328666927939967,
+      "train_runtime": 22235.2418,
+      "train_samples_per_second": 10.848,
+      "train_steps_per_second": 0.684
+    }
+  ],
+  "max_steps": 15200,
+  "num_train_epochs": 400,
+  "total_flos": 1.2604727427386573e+17,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a669d4b7a2fb08f2edacbf4cd42cf41502c9f9b962efa190cdc49ccce3537cef
+size 3823

zero_to_fp32.py

@@ -0,0 +1,453 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+from collections import OrderedDict
+
+# 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.
+import deepspeed
+from deepspeed.utils import logger
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+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_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir, "*_optim_states.pt")))
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    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
+    }
+    return buffers
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not "zero_stage" in state_dicts[0]['optimizer_state_dict']:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0]['optimizer_state_dict']["zero_stage"]
+    world_size = state_dicts[0]['optimizer_state_dict']["partition_count"]
+    param_shapes = state_dicts[0]["param_shapes"]
+    # 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, param_shapes, 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, param_shapes, 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_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers = parse_model_state(model_file)
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # 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.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # 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"
+    )
+
+    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 _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # 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
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # 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
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # 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"{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 fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)