Upload folder using huggingface_hub

checkpoint-2000/config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "togethercomputer/RedPajama-INCITE-Base-3B-v1",
+  "architectures": [
+    "GPTNeoXForCausalLM"
+  ],
+  "bos_token_id": 0,
+  "classifier_dropout": 0.1,
+  "eos_token_id": 0,
+  "hidden_act": "gelu",
+  "hidden_size": 2560,
+  "initializer_range": 0.02,
+  "intermediate_size": 10240,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 2048,
+  "model_type": "gpt_neox",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "rotary_emb_base": 10000,
+  "rotary_pct": 1.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.29.2",
+  "use_cache": true,
+  "use_parallel_residual": false,
+  "vocab_size": 50432
+}

checkpoint-2000/generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "transformers_version": "4.29.2"
+}

checkpoint-2000/global_step2000/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae0f6f03e2b9a042e7040d37571dda22bd0d3e950df03ed5e4680affaebe259b
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6469298c8ace90fbc7971259714e479bb9737cc6c51f64a35d18e328c12e9ad5
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b607792f96f06e258eec60ef22dcc98b30de0ad9bd562bc74a3a0f3e27dbba8
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:befae2d8bb07e538242bcaa6b5a2f05b78e65d2ce137fc40418c066a1c83c438
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f97bc04e27ce3ffe3ce67efabfa7e39013a771ebe34c9c0f1209c5ba79f9726
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fb121882454f4bcd5bc7d138491c07425a145392908fc4ebb0ffa8fbec0127e
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2148d28ad84e3ca2dbea7cdb4ea8ad87377dc902e70eb9dc8954f489a029d448
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:beff7de1f10add0f38eec5deb06b9b5b0be956388fb7c9b7ad25013b6beddca3
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_4_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b801eff4581db2df051b70bc0ad27541d48905c1a5a62087906cd1155a05f2
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:421b5a808ae76da386bf0d4b2d2f6e8f4b5249726379793dc781abba11cced76
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_5_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e37d12e876d126a72a887d165ebc0f7e0da7a0254a0a99441add1a427778d055
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b68830f940fb8d8571a2a3b8e53bc68672f2ec72b2fb64856d81f6d366acba4
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_6_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99dd15972281c642940b47f190323f2794d09fe4692091f47139f923f40bea9a
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa08c5fc751191064a192d03ca7b5e6964aff16f3a55b121e71c924e2b6155e0
+size 4163799934

checkpoint-2000/global_step2000/zero_pp_rank_7_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36d07c4e5ee7ef1707a189669fb0567221e81a6e3d6924efa5f21097e76c7e3b
+size 134451731

checkpoint-2000/global_step2000/zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f1e6a8b930a05f5e00f10db83dc5c6766abd1be24fffa44ff3db705684db599
+size 4163799934

checkpoint-2000/latest

@@ -0,0 +1 @@
+global_step2000

checkpoint-2000/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b41d1298850da0d466fd2e4c29f479ff3b63ee88a71380e65a10da57e5a1b86
+size 5686106713

checkpoint-2000/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f861394d4c199f79c4525d5c4e4a61b3b63c86a2b1d3dc24b3dc7a2cc4fa854
+size 21687

checkpoint-2000/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd08b19f43c113c5022aca7a2acf0240613bc8251cd7af1148f3c44559e35154
+size 21687

checkpoint-2000/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c00d88da7de57a836e00899e7c78e54c079c1a1fe1e7043fc0be33366a97fe0
+size 21687

checkpoint-2000/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6188a73b12856dc180d93425453d17da5bdd83832db408e0f7d2083eaa43ba5
+size 21687

checkpoint-2000/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d19b7bd12d313b6f367b34796e48df3b04d36d8495ca95d863ee4f2336c70e4e
+size 21687

checkpoint-2000/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6256ffefcfd855ee249d911b58b6816ea62a0d91948199d62ecf03d4c97de31
+size 21687

checkpoint-2000/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:547eb574539d57fcd695b847934f427c088fd30c5dbc54a5f3359143a7c4b460
+size 21687

checkpoint-2000/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:181de8aa271fc693ebecd232d432aae59ec9f7545015a370608182c9b36e0cfa
+size 21687

checkpoint-2000/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "pad_token": "[PAD]",
+  "unk_token": "<|endoftext|>"
+}

checkpoint-2000/tokenizer_config.json

@@ -0,0 +1,11 @@
+{
+  "add_prefix_space": false,
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "padding_side": "right",
+  "tokenizer_class": "GPTNeoXTokenizer",
+  "unk_token": "<|endoftext|>"
+}

checkpoint-2000/trainer_state.json

@@ -0,0 +1,1216 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1333.3333333333333,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 6.67,
+      "learning_rate": 9.46713625058711e-06,
+      "loss": 2.0455,
+      "step": 10
+    },
+    {
+      "epoch": 13.33,
+      "learning_rate": 1.3783995508828243e-05,
+      "loss": 1.425,
+      "step": 20
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 1.603472631319529e-05,
+      "loss": 0.5237,
+      "step": 30
+    },
+    {
+      "epoch": 26.67,
+      "learning_rate": 1.7567641489142956e-05,
+      "loss": 0.1184,
+      "step": 40
+    },
+    {
+      "epoch": 33.33,
+      "learning_rate": 1.8731528764550483e-05,
+      "loss": 0.0585,
+      "step": 50
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 1.9670033192067303e-05,
+      "loss": 0.0411,
+      "step": 60
+    },
+    {
+      "epoch": 46.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0321,
+      "step": 70
+    },
+    {
+      "epoch": 53.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0232,
+      "step": 80
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0182,
+      "step": 90
+    },
+    {
+      "epoch": 66.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0137,
+      "step": 100
+    },
+    {
+      "epoch": 73.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0111,
+      "step": 110
+    },
+    {
+      "epoch": 80.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0096,
+      "step": 120
+    },
+    {
+      "epoch": 86.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0085,
+      "step": 130
+    },
+    {
+      "epoch": 93.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0078,
+      "step": 140
+    },
+    {
+      "epoch": 100.0,
+      "learning_rate": 2e-05,
+      "loss": 0.007,
+      "step": 150
+    },
+    {
+      "epoch": 106.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0066,
+      "step": 160
+    },
+    {
+      "epoch": 113.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0061,
+      "step": 170
+    },
+    {
+      "epoch": 120.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0057,
+      "step": 180
+    },
+    {
+      "epoch": 126.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0054,
+      "step": 190
+    },
+    {
+      "epoch": 133.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0052,
+      "step": 200
+    },
+    {
+      "epoch": 140.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0049,
+      "step": 210
+    },
+    {
+      "epoch": 146.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0048,
+      "step": 220
+    },
+    {
+      "epoch": 153.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0048,
+      "step": 230
+    },
+    {
+      "epoch": 160.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0045,
+      "step": 240
+    },
+    {
+      "epoch": 166.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0046,
+      "step": 250
+    },
+    {
+      "epoch": 173.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0045,
+      "step": 260
+    },
+    {
+      "epoch": 180.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 270
+    },
+    {
+      "epoch": 186.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0044,
+      "step": 280
+    },
+    {
+      "epoch": 193.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 290
+    },
+    {
+      "epoch": 200.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 300
+    },
+    {
+      "epoch": 206.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 310
+    },
+    {
+      "epoch": 213.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 320
+    },
+    {
+      "epoch": 220.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 330
+    },
+    {
+      "epoch": 226.67,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 340
+    },
+    {
+      "epoch": 233.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 350
+    },
+    {
+      "epoch": 240.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 360
+    },
+    {
+      "epoch": 246.67,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 370
+    },
+    {
+      "epoch": 253.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 380
+    },
+    {
+      "epoch": 260.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 390
+    },
+    {
+      "epoch": 266.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 400
+    },
+    {
+      "epoch": 273.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 410
+    },
+    {
+      "epoch": 280.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 420
+    },
+    {
+      "epoch": 286.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 430
+    },
+    {
+      "epoch": 293.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 440
+    },
+    {
+      "epoch": 300.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 450
+    },
+    {
+      "epoch": 306.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 460
+    },
+    {
+      "epoch": 313.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 470
+    },
+    {
+      "epoch": 320.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 480
+    },
+    {
+      "epoch": 326.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 490
+    },
+    {
+      "epoch": 333.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 500
+    },
+    {
+      "epoch": 340.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 510
+    },
+    {
+      "epoch": 346.67,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 520
+    },
+    {
+      "epoch": 353.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 530
+    },
+    {
+      "epoch": 360.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 540
+    },
+    {
+      "epoch": 366.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0045,
+      "step": 550
+    },
+    {
+      "epoch": 373.33,
+      "learning_rate": 2e-05,
+      "loss": 0.005,
+      "step": 560
+    },
+    {
+      "epoch": 380.0,
+      "learning_rate": 2e-05,
+      "loss": 0.014,
+      "step": 570
+    },
+    {
+      "epoch": 386.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0149,
+      "step": 580
+    },
+    {
+      "epoch": 393.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0084,
+      "step": 590
+    },
+    {
+      "epoch": 400.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0072,
+      "step": 600
+    },
+    {
+      "epoch": 406.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0058,
+      "step": 610
+    },
+    {
+      "epoch": 413.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0053,
+      "step": 620
+    },
+    {
+      "epoch": 420.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0051,
+      "step": 630
+    },
+    {
+      "epoch": 426.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0047,
+      "step": 640
+    },
+    {
+      "epoch": 433.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0045,
+      "step": 650
+    },
+    {
+      "epoch": 440.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 660
+    },
+    {
+      "epoch": 446.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 670
+    },
+    {
+      "epoch": 453.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 680
+    },
+    {
+      "epoch": 460.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 690
+    },
+    {
+      "epoch": 466.67,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 700
+    },
+    {
+      "epoch": 473.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 710
+    },
+    {
+      "epoch": 480.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 720
+    },
+    {
+      "epoch": 486.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 730
+    },
+    {
+      "epoch": 493.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 740
+    },
+    {
+      "epoch": 500.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 750
+    },
+    {
+      "epoch": 506.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 760
+    },
+    {
+      "epoch": 513.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 770
+    },
+    {
+      "epoch": 520.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 780
+    },
+    {
+      "epoch": 526.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 790
+    },
+    {
+      "epoch": 533.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 800
+    },
+    {
+      "epoch": 540.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 810
+    },
+    {
+      "epoch": 546.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 820
+    },
+    {
+      "epoch": 553.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 830
+    },
+    {
+      "epoch": 560.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 840
+    },
+    {
+      "epoch": 566.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 850
+    },
+    {
+      "epoch": 573.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 860
+    },
+    {
+      "epoch": 580.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 870
+    },
+    {
+      "epoch": 586.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 880
+    },
+    {
+      "epoch": 593.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 890
+    },
+    {
+      "epoch": 600.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 900
+    },
+    {
+      "epoch": 606.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 910
+    },
+    {
+      "epoch": 613.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 920
+    },
+    {
+      "epoch": 620.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 930
+    },
+    {
+      "epoch": 626.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 940
+    },
+    {
+      "epoch": 633.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 950
+    },
+    {
+      "epoch": 640.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 960
+    },
+    {
+      "epoch": 646.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 970
+    },
+    {
+      "epoch": 653.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 980
+    },
+    {
+      "epoch": 660.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 990
+    },
+    {
+      "epoch": 666.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1000
+    },
+    {
+      "epoch": 673.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1010
+    },
+    {
+      "epoch": 680.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1020
+    },
+    {
+      "epoch": 686.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1030
+    },
+    {
+      "epoch": 693.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1040
+    },
+    {
+      "epoch": 700.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1050
+    },
+    {
+      "epoch": 706.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1060
+    },
+    {
+      "epoch": 713.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1070
+    },
+    {
+      "epoch": 720.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1080
+    },
+    {
+      "epoch": 726.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1090
+    },
+    {
+      "epoch": 733.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1100
+    },
+    {
+      "epoch": 740.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1110
+    },
+    {
+      "epoch": 746.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1120
+    },
+    {
+      "epoch": 753.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1130
+    },
+    {
+      "epoch": 760.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1140
+    },
+    {
+      "epoch": 766.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 1150
+    },
+    {
+      "epoch": 773.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0049,
+      "step": 1160
+    },
+    {
+      "epoch": 780.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0049,
+      "step": 1170
+    },
+    {
+      "epoch": 786.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0048,
+      "step": 1180
+    },
+    {
+      "epoch": 793.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0048,
+      "step": 1190
+    },
+    {
+      "epoch": 800.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0046,
+      "step": 1200
+    },
+    {
+      "epoch": 806.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0041,
+      "step": 1210
+    },
+    {
+      "epoch": 813.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1220
+    },
+    {
+      "epoch": 820.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1230
+    },
+    {
+      "epoch": 826.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 1240
+    },
+    {
+      "epoch": 833.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1250
+    },
+    {
+      "epoch": 840.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1260
+    },
+    {
+      "epoch": 846.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1270
+    },
+    {
+      "epoch": 853.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1280
+    },
+    {
+      "epoch": 860.0,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1290
+    },
+    {
+      "epoch": 866.67,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1300
+    },
+    {
+      "epoch": 873.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0051,
+      "step": 1310
+    },
+    {
+      "epoch": 880.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0127,
+      "step": 1320
+    },
+    {
+      "epoch": 886.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0082,
+      "step": 1330
+    },
+    {
+      "epoch": 893.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0193,
+      "step": 1340
+    },
+    {
+      "epoch": 900.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0072,
+      "step": 1350
+    },
+    {
+      "epoch": 906.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0055,
+      "step": 1360
+    },
+    {
+      "epoch": 913.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0052,
+      "step": 1370
+    },
+    {
+      "epoch": 920.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0045,
+      "step": 1380
+    },
+    {
+      "epoch": 926.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 1390
+    },
+    {
+      "epoch": 933.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 1400
+    },
+    {
+      "epoch": 940.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1410
+    },
+    {
+      "epoch": 946.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1420
+    },
+    {
+      "epoch": 953.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1430
+    },
+    {
+      "epoch": 960.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 1440
+    },
+    {
+      "epoch": 966.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1450
+    },
+    {
+      "epoch": 973.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1460
+    },
+    {
+      "epoch": 980.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0037,
+      "step": 1470
+    },
+    {
+      "epoch": 986.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1480
+    },
+    {
+      "epoch": 993.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1490
+    },
+    {
+      "epoch": 1000.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1500
+    },
+    {
+      "epoch": 1006.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1510
+    },
+    {
+      "epoch": 1013.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1520
+    },
+    {
+      "epoch": 1020.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1530
+    },
+    {
+      "epoch": 1026.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1540
+    },
+    {
+      "epoch": 1033.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1550
+    },
+    {
+      "epoch": 1040.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1560
+    },
+    {
+      "epoch": 1046.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1570
+    },
+    {
+      "epoch": 1053.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1580
+    },
+    {
+      "epoch": 1060.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1590
+    },
+    {
+      "epoch": 1066.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1600
+    },
+    {
+      "epoch": 1073.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1610
+    },
+    {
+      "epoch": 1080.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1620
+    },
+    {
+      "epoch": 1086.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1630
+    },
+    {
+      "epoch": 1093.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1640
+    },
+    {
+      "epoch": 1100.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1650
+    },
+    {
+      "epoch": 1106.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1660
+    },
+    {
+      "epoch": 1113.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1670
+    },
+    {
+      "epoch": 1120.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1680
+    },
+    {
+      "epoch": 1126.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1690
+    },
+    {
+      "epoch": 1133.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1700
+    },
+    {
+      "epoch": 1140.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1710
+    },
+    {
+      "epoch": 1146.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1720
+    },
+    {
+      "epoch": 1153.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1730
+    },
+    {
+      "epoch": 1160.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1740
+    },
+    {
+      "epoch": 1166.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1750
+    },
+    {
+      "epoch": 1173.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1760
+    },
+    {
+      "epoch": 1180.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1770
+    },
+    {
+      "epoch": 1186.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1780
+    },
+    {
+      "epoch": 1193.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1790
+    },
+    {
+      "epoch": 1200.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1800
+    },
+    {
+      "epoch": 1206.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1810
+    },
+    {
+      "epoch": 1213.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1820
+    },
+    {
+      "epoch": 1220.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1830
+    },
+    {
+      "epoch": 1226.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1840
+    },
+    {
+      "epoch": 1233.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1850
+    },
+    {
+      "epoch": 1240.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0063,
+      "step": 1860
+    },
+    {
+      "epoch": 1246.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1870
+    },
+    {
+      "epoch": 1253.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0061,
+      "step": 1880
+    },
+    {
+      "epoch": 1260.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0071,
+      "step": 1890
+    },
+    {
+      "epoch": 1266.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0055,
+      "step": 1900
+    },
+    {
+      "epoch": 1273.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0053,
+      "step": 1910
+    },
+    {
+      "epoch": 1280.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1920
+    },
+    {
+      "epoch": 1286.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 1930
+    },
+    {
+      "epoch": 1293.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 1940
+    },
+    {
+      "epoch": 1300.0,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1950
+    },
+    {
+      "epoch": 1306.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1960
+    },
+    {
+      "epoch": 1313.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1970
+    },
+    {
+      "epoch": 1320.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1980
+    },
+    {
+      "epoch": 1326.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 1990
+    },
+    {
+      "epoch": 1333.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 2000
+    }
+  ],
+  "max_steps": 2001,
+  "num_train_epochs": 2001,
+  "total_flos": 841518856273920.0,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-2000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6f37d82e0ffda307f018bd024cb5411a96a38d758bc5f0c3bd5e051be3c5dc1
+size 5563

checkpoint-2000/zero_to_fp32.py

@@ -0,0 +1,483 @@
+#!/usr/bin/env python
+'''Copyright The Microsoft DeepSpeed Team'''
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# 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)
+
+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_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")),
+                         key=natural_keys)
+
+    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
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    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)

flyte_training_config.json

@@ -1 +1 @@
-{"base_model": "togethercomputer/RedPajama-INCITE-Base-3B-v1", "data_path": "wikipedia", "data_name": "20220301.simple", "num_epochs": 1, "max_steps": 1500, "learning_rate": 2e-05, "weight_decay": 0.02, "warmup_ratio": 0.03, "lr_scheduler_type": "cosine", "batch_size": 16, "micro_batch_size": 1, "val_set_size": 0, "group_by_length": false, "instruction_key": "instruction", "input_key": "input", "output_key": "output", "device_map": "auto", "cache_dir": null, "optim": "adamw_torch", "model_max_length": 512, "debug_mode": false, "debug_train_data_size": 1024, "wandb_project": ""}
+{"base_model": "togethercomputer/RedPajama-INCITE-Base-3B-v1", "data_path": "wikipedia", "data_name": "20220301.simple", "num_epochs": 1, "max_steps": 2001, "learning_rate": 2e-05, "weight_decay": 0.02, "warmup_ratio": 0.03, "lr_scheduler_type": "cosine", "batch_size": 16, "micro_batch_size": 1, "val_set_size": 0, "group_by_length": false, "instruction_key": "instruction", "input_key": "input", "output_key": "output", "device_map": "auto", "cache_dir": null, "optim": "adamw_torch", "model_max_length": 512, "debug_mode": false, "debug_train_data_size": 1024, "wandb_project": ""}

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9481ef4ff637f2e49e3f7aee08b56c685a6130dd015f5546e00a1bff515ff8b8
+oid sha256:6ffb1344823c7e872b10d296a16b98a5712cc86b52b776238f0b801f54d28833
 size 5686106713

trainer_state.json

@@ -1,88 +1,88 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 1000.0,
-  "global_step": 1500,
+  "epoch": 1334.0,
+  "global_step": 2001,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 6.67,
-      "learning_rate": 1.0223707272628328e-05,
-      "loss": 2.0543,
+      "learning_rate": 9.46713625058711e-06,
+      "loss": 2.0455,
       "step": 10
     },
     {
       "epoch": 13.33,
-      "learning_rate": 1.4885550540243259e-05,
-      "loss": 1.4127,
+      "learning_rate": 1.3783995508828243e-05,
+      "loss": 1.425,
       "step": 20
     },
     {
       "epoch": 20.0,
-      "learning_rate": 1.73161496447939e-05,
-      "loss": 0.5013,
+      "learning_rate": 1.603472631319529e-05,
+      "loss": 0.5237,
       "step": 30
     },
     {
       "epoch": 26.67,
-      "learning_rate": 1.897156851886858e-05,
-      "loss": 0.1121,
+      "learning_rate": 1.7567641489142956e-05,
+      "loss": 0.1184,
       "step": 40
     },
     {
       "epoch": 33.33,
-      "learning_rate": 2e-05,
-      "loss": 0.0576,
+      "learning_rate": 1.8731528764550483e-05,
+      "loss": 0.0585,
       "step": 50
     },
     {
       "epoch": 40.0,
-      "learning_rate": 2e-05,
-      "loss": 0.039,
+      "learning_rate": 1.9670033192067303e-05,
+      "loss": 0.0411,
       "step": 60
     },
     {
       "epoch": 46.67,
       "learning_rate": 2e-05,
-      "loss": 0.0296,
+      "loss": 0.0321,
       "step": 70
     },
     {
       "epoch": 53.33,
       "learning_rate": 2e-05,
-      "loss": 0.0221,
+      "loss": 0.0232,
       "step": 80
     },
     {
       "epoch": 60.0,
       "learning_rate": 2e-05,
-      "loss": 0.0173,
+      "loss": 0.0182,
       "step": 90
     },
     {
       "epoch": 66.67,
       "learning_rate": 2e-05,
-      "loss": 0.0134,
+      "loss": 0.0137,
       "step": 100
     },
     {
       "epoch": 73.33,
       "learning_rate": 2e-05,
-      "loss": 0.011,
+      "loss": 0.0111,
       "step": 110
     },
     {
       "epoch": 80.0,
       "learning_rate": 2e-05,
-      "loss": 0.0099,
+      "loss": 0.0096,
       "step": 120
     },
     {
       "epoch": 86.67,
       "learning_rate": 2e-05,
-      "loss": 0.0092,
+      "loss": 0.0085,
       "step": 130
     },
     {
@@ -94,73 +94,73 @@
     {
       "epoch": 100.0,
       "learning_rate": 2e-05,
-      "loss": 0.0072,
+      "loss": 0.007,
       "step": 150
     },
     {
       "epoch": 106.67,
       "learning_rate": 2e-05,
-      "loss": 0.01,
+      "loss": 0.0066,
       "step": 160
     },
     {
       "epoch": 113.33,
       "learning_rate": 2e-05,
-      "loss": 0.0169,
+      "loss": 0.0061,
       "step": 170
     },
     {
       "epoch": 120.0,
       "learning_rate": 2e-05,
-      "loss": 0.0064,
+      "loss": 0.0057,
       "step": 180
     },
     {
       "epoch": 126.67,
       "learning_rate": 2e-05,
-      "loss": 0.0058,
+      "loss": 0.0054,
       "step": 190
     },
     {
       "epoch": 133.33,
       "learning_rate": 2e-05,
-      "loss": 0.0154,
+      "loss": 0.0052,
       "step": 200
     },
     {
       "epoch": 140.0,
       "learning_rate": 2e-05,
-      "loss": 0.0126,
+      "loss": 0.0049,
       "step": 210
     },
     {
       "epoch": 146.67,
       "learning_rate": 2e-05,
-      "loss": 0.0052,
+      "loss": 0.0048,
       "step": 220
     },
     {
       "epoch": 153.33,
       "learning_rate": 2e-05,
-      "loss": 0.0076,
+      "loss": 0.0048,
       "step": 230
     },
     {
       "epoch": 160.0,
       "learning_rate": 2e-05,
-      "loss": 0.006,
+      "loss": 0.0045,
       "step": 240
     },
     {
       "epoch": 166.67,
       "learning_rate": 2e-05,
-      "loss": 0.005,
+      "loss": 0.0046,
       "step": 250
     },
     {
       "epoch": 173.33,
       "learning_rate": 2e-05,
-      "loss": 0.0047,
+      "loss": 0.0045,
       "step": 260
     },
     {
@@ -184,19 +184,19 @@
     {
       "epoch": 200.0,
       "learning_rate": 2e-05,
-      "loss": 0.0038,
+      "loss": 0.0042,
       "step": 300
     },
     {
       "epoch": 206.67,
       "learning_rate": 2e-05,
-      "loss": 0.004,
+      "loss": 0.0041,
       "step": 310
     },
     {
       "epoch": 213.33,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.0041,
       "step": 320
     },
     {
@@ -208,43 +208,43 @@
     {
       "epoch": 226.67,
       "learning_rate": 2e-05,
-      "loss": 0.0042,
+      "loss": 0.004,
       "step": 340
     },
     {
       "epoch": 233.33,
       "learning_rate": 2e-05,
-      "loss": 0.0041,
+      "loss": 0.004,
       "step": 350
     },
     {
       "epoch": 240.0,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.0038,
       "step": 360
     },
     {
       "epoch": 246.67,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.004,
       "step": 370
     },
     {
       "epoch": 253.33,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.004,
       "step": 380
     },
     {
       "epoch": 260.0,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.0041,
       "step": 390
     },
     {
       "epoch": 266.67,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0039,
       "step": 400
     },
     {
@@ -262,25 +262,25 @@
     {
       "epoch": 286.67,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.0038,
       "step": 430
     },
     {
       "epoch": 293.33,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.004,
       "step": 440
     },
     {
       "epoch": 300.0,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0039,
       "step": 450
     },
     {
       "epoch": 306.67,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0038,
       "step": 460
     },
     {
@@ -292,19 +292,19 @@
     {
       "epoch": 320.0,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0038,
       "step": 480
     },
     {
       "epoch": 326.67,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0038,
       "step": 490
     },
     {
       "epoch": 333.33,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.0037,
       "step": 500
     },
     {
@@ -316,205 +316,205 @@
     {
       "epoch": 346.67,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.004,
       "step": 520
     },
     {
       "epoch": 353.33,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0037,
       "step": 530
     },
     {
       "epoch": 360.0,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0039,
       "step": 540
     },
     {
       "epoch": 366.67,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.0045,
       "step": 550
     },
     {
       "epoch": 373.33,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.005,
       "step": 560
     },
     {
       "epoch": 380.0,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.014,
       "step": 570
     },
     {
       "epoch": 386.67,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.0149,
       "step": 580
     },
     {
       "epoch": 393.33,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0084,
       "step": 590
     },
     {
       "epoch": 400.0,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0072,
       "step": 600
     },
     {
       "epoch": 406.67,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.0058,
       "step": 610
     },
     {
       "epoch": 413.33,
       "learning_rate": 2e-05,
-      "loss": 0.0038,
+      "loss": 0.0053,
       "step": 620
     },
     {
       "epoch": 420.0,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0051,
       "step": 630
     },
     {
       "epoch": 426.67,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0047,
       "step": 640
     },
     {
       "epoch": 433.33,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0045,
       "step": 650
     },
     {
       "epoch": 440.0,
       "learning_rate": 2e-05,
-      "loss": 0.0054,
+      "loss": 0.0043,
       "step": 660
     },
     {
       "epoch": 446.67,
       "learning_rate": 2e-05,
-      "loss": 0.0048,
+      "loss": 0.0041,
       "step": 670
     },
     {
       "epoch": 453.33,
       "learning_rate": 2e-05,
-      "loss": 0.0041,
+      "loss": 0.0039,
       "step": 680
     },
     {
       "epoch": 460.0,
       "learning_rate": 2e-05,
-      "loss": 0.004,
+      "loss": 0.0038,
       "step": 690
     },
     {
       "epoch": 466.67,
       "learning_rate": 2e-05,
-      "loss": 0.005,
+      "loss": 0.004,
       "step": 700
     },
     {
       "epoch": 473.33,
       "learning_rate": 2e-05,
-      "loss": 0.0052,
+      "loss": 0.004,
       "step": 710
     },
     {
       "epoch": 480.0,
       "learning_rate": 2e-05,
-      "loss": 0.0049,
+      "loss": 0.0036,
       "step": 720
     },
     {
       "epoch": 486.67,
       "learning_rate": 2e-05,
-      "loss": 0.0053,
+      "loss": 0.0036,
       "step": 730
     },
     {
       "epoch": 493.33,
       "learning_rate": 2e-05,
-      "loss": 0.0046,
+      "loss": 0.0037,
       "step": 740
     },
     {
       "epoch": 500.0,
       "learning_rate": 2e-05,
-      "loss": 0.0049,
+      "loss": 0.0036,
       "step": 750
     },
     {
       "epoch": 506.67,
       "learning_rate": 2e-05,
-      "loss": 0.0046,
+      "loss": 0.0034,
       "step": 760
     },
     {
       "epoch": 513.33,
       "learning_rate": 2e-05,
-      "loss": 0.0043,
+      "loss": 0.0035,
       "step": 770
     },
     {
       "epoch": 520.0,
       "learning_rate": 2e-05,
-      "loss": 0.0045,
+      "loss": 0.0035,
       "step": 780
     },
     {
       "epoch": 526.67,
       "learning_rate": 2e-05,
-      "loss": 0.0044,
+      "loss": 0.0035,
       "step": 790
     },
     {
       "epoch": 533.33,
       "learning_rate": 2e-05,
-      "loss": 0.0044,
+      "loss": 0.0034,
       "step": 800
     },
     {
       "epoch": 540.0,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.0035,
       "step": 810
     },
     {
       "epoch": 546.67,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.0034,
       "step": 820
     },
     {
       "epoch": 553.33,
       "learning_rate": 2e-05,
-      "loss": 0.0039,
+      "loss": 0.0037,
       "step": 830
     },
     {
       "epoch": 560.0,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0034,
       "step": 840
     },
     {
       "epoch": 566.67,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0034,
       "step": 850
     },
     {
@@ -532,91 +532,91 @@
     {
       "epoch": 586.67,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0035,
       "step": 880
     },
     {
       "epoch": 593.33,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0033,
       "step": 890
     },
     {
       "epoch": 600.0,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.0035,
       "step": 900
     },
     {
       "epoch": 606.67,
       "learning_rate": 2e-05,
-      "loss": 0.0047,
+      "loss": 0.0035,
       "step": 910
     },
     {
       "epoch": 613.33,
       "learning_rate": 2e-05,
-      "loss": 0.004,
+      "loss": 0.0034,
       "step": 920
     },
     {
       "epoch": 620.0,
       "learning_rate": 2e-05,
-      "loss": 0.004,
+      "loss": 0.0033,
       "step": 930
     },
     {
       "epoch": 626.67,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0034,
       "step": 940
     },
     {
       "epoch": 633.33,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.0035,
       "step": 950
     },
     {
       "epoch": 640.0,
       "learning_rate": 2e-05,
-      "loss": 0.004,
+      "loss": 0.0034,
       "step": 960
     },
     {
       "epoch": 646.67,
       "learning_rate": 2e-05,
-      "loss": 0.0042,
+      "loss": 0.0035,
       "step": 970
     },
     {
       "epoch": 653.33,
       "learning_rate": 2e-05,
-      "loss": 0.0037,
+      "loss": 0.0034,
       "step": 980
     },
     {
       "epoch": 660.0,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0035,
       "step": 990
     },
     {
       "epoch": 666.67,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0034,
       "step": 1000
     },
     {
       "epoch": 673.33,
       "learning_rate": 2e-05,
-      "loss": 0.0036,
+      "loss": 0.0035,
       "step": 1010
     },
     {
       "epoch": 680.0,
       "learning_rate": 2e-05,
-      "loss": 0.0035,
+      "loss": 0.0034,
       "step": 1020
     },
     {
@@ -628,7 +628,7 @@
     {
       "epoch": 693.33,
       "learning_rate": 2e-05,
-      "loss": 0.0034,
+      "loss": 0.0033,
       "step": 1040
     },
     {
@@ -646,280 +646,580 @@
     {
       "epoch": 713.33,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0035,
       "step": 1070
     },
     {
       "epoch": 720.0,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0035,
       "step": 1080
     },
     {
       "epoch": 726.67,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0035,
       "step": 1090
     },
     {
       "epoch": 733.33,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0034,
       "step": 1100
     },
     {
       "epoch": 740.0,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0034,
       "step": 1110
     },
     {
       "epoch": 746.67,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0034,
       "step": 1120
     },
     {
       "epoch": 753.33,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0035,
       "step": 1130
     },
     {
       "epoch": 760.0,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0035,
       "step": 1140
     },
     {
       "epoch": 766.67,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0039,
       "step": 1150
     },
     {
       "epoch": 773.33,
       "learning_rate": 2e-05,
-      "loss": 0.003,
+      "loss": 0.0049,
       "step": 1160
     },
     {
       "epoch": 780.0,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0049,
       "step": 1170
     },
     {
       "epoch": 786.67,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0048,
       "step": 1180
     },
     {
       "epoch": 793.33,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0048,
       "step": 1190
     },
     {
       "epoch": 800.0,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0046,
       "step": 1200
     },
     {
       "epoch": 806.67,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0041,
       "step": 1210
     },
     {
       "epoch": 813.33,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0038,
       "step": 1220
     },
     {
       "epoch": 820.0,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0043,
       "step": 1230
     },
     {
       "epoch": 826.67,
       "learning_rate": 2e-05,
-      "loss": 0.0029,
+      "loss": 0.0042,
       "step": 1240
     },
     {
       "epoch": 833.33,
       "learning_rate": 2e-05,
-      "loss": 0.003,
+      "loss": 0.004,
       "step": 1250
     },
     {
       "epoch": 840.0,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0037,
       "step": 1260
     },
     {
       "epoch": 846.67,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0043,
       "step": 1270
     },
     {
       "epoch": 853.33,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0037,
       "step": 1280
     },
     {
       "epoch": 860.0,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.004,
       "step": 1290
     },
     {
       "epoch": 866.67,
       "learning_rate": 2e-05,
-      "loss": 0.0029,
+      "loss": 0.004,
       "step": 1300
     },
     {
       "epoch": 873.33,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0051,
       "step": 1310
     },
     {
       "epoch": 880.0,
       "learning_rate": 2e-05,
-      "loss": 0.003,
+      "loss": 0.0127,
       "step": 1320
     },
     {
       "epoch": 886.67,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0082,
       "step": 1330
     },
     {
       "epoch": 893.33,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0193,
       "step": 1340
     },
     {
       "epoch": 900.0,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0072,
       "step": 1350
     },
     {
       "epoch": 906.67,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0055,
       "step": 1360
     },
     {
       "epoch": 913.33,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0052,
       "step": 1370
     },
     {
       "epoch": 920.0,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0045,
       "step": 1380
     },
     {
       "epoch": 926.67,
       "learning_rate": 2e-05,
-      "loss": 0.0031,
+      "loss": 0.0042,
       "step": 1390
     },
     {
       "epoch": 933.33,
       "learning_rate": 2e-05,
-      "loss": 0.0033,
+      "loss": 0.0042,
       "step": 1400
     },
     {
       "epoch": 940.0,
       "learning_rate": 2e-05,
-      "loss": 0.0032,
+      "loss": 0.0037,
       "step": 1410
     },
     {
       "epoch": 946.67,
       "learning_rate": 2e-05,
-      "loss": 0.0038,
+      "loss": 0.0037,
       "step": 1420
     },
     {
       "epoch": 953.33,
       "learning_rate": 2e-05,
-      "loss": 0.0044,
+      "loss": 0.0038,
       "step": 1430
     },
     {
       "epoch": 960.0,
       "learning_rate": 2e-05,
-      "loss": 0.0045,
+      "loss": 0.0036,
       "step": 1440
     },
     {
       "epoch": 966.67,
       "learning_rate": 2e-05,
-      "loss": 0.0049,
+      "loss": 0.0037,
       "step": 1450
     },
     {
       "epoch": 973.33,
       "learning_rate": 2e-05,
-      "loss": 0.005,
+      "loss": 0.0035,
       "step": 1460
     },
     {
       "epoch": 980.0,
       "learning_rate": 2e-05,
-      "loss": 0.0048,
+      "loss": 0.0037,
       "step": 1470
     },
     {
       "epoch": 986.67,
       "learning_rate": 2e-05,
-      "loss": 0.0051,
+      "loss": 0.0035,
       "step": 1480
     },
     {
       "epoch": 993.33,
       "learning_rate": 2e-05,
-      "loss": 0.0049,
+      "loss": 0.0034,
       "step": 1490
     },
     {
       "epoch": 1000.0,
       "learning_rate": 2e-05,
-      "loss": 0.005,
+      "loss": 0.0035,
       "step": 1500
     },
     {
-      "epoch": 1000.0,
-      "step": 1500,
-      "total_flos": 631132567633920.0,
-      "train_loss": 0.032416117899119856,
-      "train_runtime": 64000.2264,
+      "epoch": 1006.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1510
+    },
+    {
+      "epoch": 1013.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1520
+    },
+    {
+      "epoch": 1020.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1530
+    },
+    {
+      "epoch": 1026.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0035,
+      "step": 1540
+    },
+    {
+      "epoch": 1033.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1550
+    },
+    {
+      "epoch": 1040.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1560
+    },
+    {
+      "epoch": 1046.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1570
+    },
+    {
+      "epoch": 1053.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1580
+    },
+    {
+      "epoch": 1060.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1590
+    },
+    {
+      "epoch": 1066.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1600
+    },
+    {
+      "epoch": 1073.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1610
+    },
+    {
+      "epoch": 1080.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1620
+    },
+    {
+      "epoch": 1086.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1630
+    },
+    {
+      "epoch": 1093.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1640
+    },
+    {
+      "epoch": 1100.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1650
+    },
+    {
+      "epoch": 1106.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1660
+    },
+    {
+      "epoch": 1113.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1670
+    },
+    {
+      "epoch": 1120.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1680
+    },
+    {
+      "epoch": 1126.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1690
+    },
+    {
+      "epoch": 1133.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1700
+    },
+    {
+      "epoch": 1140.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1710
+    },
+    {
+      "epoch": 1146.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1720
+    },
+    {
+      "epoch": 1153.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1730
+    },
+    {
+      "epoch": 1160.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1740
+    },
+    {
+      "epoch": 1166.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1750
+    },
+    {
+      "epoch": 1173.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0033,
+      "step": 1760
+    },
+    {
+      "epoch": 1180.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1770
+    },
+    {
+      "epoch": 1186.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1780
+    },
+    {
+      "epoch": 1193.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1790
+    },
+    {
+      "epoch": 1200.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1800
+    },
+    {
+      "epoch": 1206.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1810
+    },
+    {
+      "epoch": 1213.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1820
+    },
+    {
+      "epoch": 1220.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1830
+    },
+    {
+      "epoch": 1226.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1840
+    },
+    {
+      "epoch": 1233.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0034,
+      "step": 1850
+    },
+    {
+      "epoch": 1240.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0063,
+      "step": 1860
+    },
+    {
+      "epoch": 1246.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0032,
+      "step": 1870
+    },
+    {
+      "epoch": 1253.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0061,
+      "step": 1880
+    },
+    {
+      "epoch": 1260.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0071,
+      "step": 1890
+    },
+    {
+      "epoch": 1266.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0055,
+      "step": 1900
+    },
+    {
+      "epoch": 1273.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0053,
+      "step": 1910
+    },
+    {
+      "epoch": 1280.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0043,
+      "step": 1920
+    },
+    {
+      "epoch": 1286.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0042,
+      "step": 1930
+    },
+    {
+      "epoch": 1293.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0039,
+      "step": 1940
+    },
+    {
+      "epoch": 1300.0,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1950
+    },
+    {
+      "epoch": 1306.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1960
+    },
+    {
+      "epoch": 1313.33,
+      "learning_rate": 2e-05,
+      "loss": 0.004,
+      "step": 1970
+    },
+    {
+      "epoch": 1320.0,
+      "learning_rate": 2e-05,
+      "loss": 0.0038,
+      "step": 1980
+    },
+    {
+      "epoch": 1326.67,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 1990
+    },
+    {
+      "epoch": 1333.33,
+      "learning_rate": 2e-05,
+      "loss": 0.0036,
+      "step": 2000
+    },
+    {
+      "epoch": 1334.0,
+      "step": 2001,
+      "total_flos": 841939628851200.0,
+      "train_loss": 0.02562973479824564,
+      "train_runtime": 85372.5158,
       "train_samples_per_second": 3.0,
       "train_steps_per_second": 0.023
     }
   ],
-  "max_steps": 1500,
-  "num_train_epochs": 1500,
-  "total_flos": 631132567633920.0,
+  "max_steps": 2001,
+  "num_train_epochs": 2001,
+  "total_flos": 841939628851200.0,
   "trial_name": null,
   "trial_params": null
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c4e2371bf98be78aa001aad3916f770787fe84501537792f54f2133de363db2
+oid sha256:a6f37d82e0ffda307f018bd024cb5411a96a38d758bc5f0c3bd5e051be3c5dc1
 size 5563