Upload folder using huggingface_hub

config.json

@@ -1,12 +1,12 @@
 {
-  "_name_or_path": "yentinglin/Llama-3-Taiwan-70B-Instruct-rc3",
+  "_name_or_path": "yentinglin/Llama-3-Taiwan-70B-Instruct",
   "architectures": [
     "LlamaForCausalLM"
   ],
   "attention_bias": false,
   "attention_dropout": 0.0,
   "bos_token_id": 128000,
-  "eos_token_id": 128009,
+  "eos_token_id": 128256,
   "hidden_act": "silu",
   "hidden_size": 8192,
   "initializer_range": 0.02,
@@ -24,5 +24,5 @@
   "torch_dtype": "bfloat16",
   "transformers_version": "4.40.2",
   "use_cache": false,
-  "vocab_size": 128256
+  "vocab_size": 128258
 }

latest

@@ -0,0 +1 @@
+global_step160

model-00001-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ceca59bf2672782b5d8dd32950298c93f8bbebff65e5114d59f184e8591dadc0
-size 4584408808
+oid sha256:3456c422f211ac69b2a0846a5bdf723b2d3776f16defc6a814ef9d2087685199
+size 4584441576

model-00002-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65311a1ae88a28f97756374bdfd3ac04819702e41b4e98a4ea3dc1e3f38da895
+oid sha256:c827cb32ca8351713bd296bf0ddbeb994df2d604fbf662391be1d6407d63cc5f
 size 4664167376

model-00003-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4fbbc92dd73a156be3e55bd4cad4c70c80f085f0a28119a356243ed99a6849a6
+oid sha256:1519c49a9afd51bc61f0f522986a9fb551a062097206b8b61147903e7c6c6496
 size 4999711704

model-00004-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21c12dbeba70e7520750904f4bdc8a5a5aa8d3ba711fdf05b40387c6a2112503
+oid sha256:97c0cdafd98d33a5308402961f181dee3fe7ed6204a2558440395b5a411d19bd
 size 4966157032

model-00005-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2049aa167f285c85dd9f57eb9dfabbd42939acf785572a549ee419935ed0c3f0
+oid sha256:3e879a6c47b37235fd975da6f9896976760d21ecf28ad70f771f9faf71713e3e
 size 4664134408

model-00006-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd2f73719934583118b06c53e8fe44b5697e9f38e4e07d57d5865a31d957ad3f
+oid sha256:4cc1b1e8b14a38c2f21085a0f55caf0f65d612227d6c169efa7735e9f81357f6
 size 4664167408

model-00007-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e626d86fb2893ac35ad28eb105601b1722b721f1c75db5f1516acd76f94d7c6b
+oid sha256:b0d47426adceabc7a4ebcfe5bc24b5830fc4af5c748d64e56f0cdd240ea6cbac
 size 4664167408

model-00008-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:827ddd50f669cab713cf36273822fab2ae87db5585c241a62ccdd880d16dc54f
+oid sha256:6ca3891ceaf01cd52f7d79ea8694378d5d0c53131cf0359e107ff0f03b45eeaa
 size 4999711728

model-00009-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:213b007783a739166d2be97c09573719695e98c70232a420a886965caad2d1cb
+oid sha256:c06eaa3666a65b4c64831ec4233270d56966e898f600cdba6cba07af0e8a559c
 size 4966157056

model-00010-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7f661f78dee33ad714f6bd6d19efdd243a49dad2f998def7c54feffde54e6e2
+oid sha256:c272793f8a90198705124b3bca6e36e605df5bd55c8fe6fa6ab21115d76e0ab1
 size 4664134408

model-00011-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72cb2d2ffea069ecdc2ea459edaf835b4a4b03cb6a7478e5dfa9b0a4fda8bbfd
+oid sha256:fde6dcdcc375f2f80e6edb57fc3c03493522b4d274209712ca2e1c1ff68eb868
 size 4664167408

model-00012-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ae342da72dabfd5dd05220feb33a03d8c307a492d025970e7ba7158ee20eda2
+oid sha256:7c31ffd3a2aa84fd522be76c3b71a756bcc05993d5336f8123cebe04248cd9b0
 size 4664167408

model-00013-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7abe91954470c523fca6880cf6344688f82f6be8456462aea18d14cca5d31d91
+oid sha256:b3c790db09692101d47da5261b8b9f363b5d94640a355a2e68ea6b8c1cc6ee80
 size 4999711728

model-00014-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb51329664948e7104e8061e3ae626c00bde9642d4639b4c1421d7690d06a1da
+oid sha256:7f62731dc23f1a8f8024aed62228ff5f2733b54e85584b500c459d4ed2b919fd
 size 4966157056

model-00015-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c66ad06b76130524832ece52e77d0b04d36689f4ad54ef3955bf29bddbd32264
+oid sha256:74b88be19757ab65ac11fe088852cf153d6299b1ee4ac8a04e8b94932b2ec6b8
 size 4664134408

model-00016-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:535a817159e772d8043eca0b5d4ac4249daf5e1ff263b00a35c5b0f5f80d1729
+oid sha256:4a39fa49652b72921656f86a2bbb3774d46d4e3769d46d102942cb47687d0a63
 size 4664167408

model-00017-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17f4484bce1cdb220889b175eeb3f5461318baffa1a9f611b175e741c08218af
+oid sha256:b53a2743088a1f168096db90806c903fa29bf0fb54bf348b6cd098c1842d0f0b
 size 4664167408

model-00018-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16497bd28ed92f99721a6bec89d1ce3048479b9f40744840da04480723b3e7f2
+oid sha256:70fef9f2a3914cdd3f685e0a6e04f83e9aeea249f65008d6750088c34ec25167
 size 4999711728

model-00019-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4649da440ca0a5ecf2db42bc13e5755e096df75b113bd6353a16885dc00419d4
+oid sha256:6f4610d20ada34f28002de4eaa8e3785b77d3df37e1cb612144e9bf9af8fe141
 size 4966157056

model-00020-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:593b4d531dd16d4e5d8c716d3b4ad7cbbd747163cca44e61e5ca995d5f9e93f9
+oid sha256:32b21b7f898f17d54d4796ab37c97bb09dd4bcea2d427ce6a3609fb70a331786
 size 4664134408

model-00021-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53c245da6fa1e9d64130ec768d57cc8ea82be2a0af2e244a7d8cef590df5a1d7
+oid sha256:b527974170cc2d8a21595dd33d3c38292eb261b6630283539aae5b62c6b2498a
 size 4664167408

model-00022-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c2b685bfae0ce5cecd5956011cc0266b47c559328e312a0b5a63b052786fc6a
+oid sha256:a59c96d221558078d5c4540e727e843500711df3e6ba33acb7d9067807907056
 size 4664167408

model-00023-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8c40cc207fb69f5b7e20ac3dcd62b582bbcbe4cb725d610092c9f1317c7a3de
+oid sha256:0f1e1ac772c05ccfad15bd3788ccb68cfbcc500026caed1593ca81ca5fcc12b5
 size 4999711728

model-00024-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ae3bf132dc1baa14d02efa0710c7abf4ce03ae27d18b4a305103608e7d2c795
+oid sha256:4dd2c2517501e87be7b5cc77f9e00a3e5bd2e5482a5f86890b97dbf8f7a58999
 size 4966157056

model-00025-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04226e40c4aadb6943a4b99263ef81b162c553023a272839afb394c561717bbf
+oid sha256:dc25480728d569497f3a8acaf6d9e1124fa215058a07d346a8794c74f3ec61ce
 size 4664134408

model-00026-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1ff1f40b6c163f26b8df9eafd78d5350d9c9c5cfcce7bb89f0ef367edb80daa
+oid sha256:4cf05dacab18c581ce789a0d55da092e3c655067adb330a1cb99699140d71aa1
 size 4664167408

model-00027-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8aff3ca5f4f9353d572d58ae20f9b60faa136ef9780d4195dca65597984905c8
+oid sha256:3f055d1be72dd2da7ac8714b05a2a071578f68a219c48e15fcfc499e5708a76c
 size 4664167408

model-00028-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3becbdc3e697907b0de01be570f7a9c8643d5dee888af6571413347f99db8a18
+oid sha256:1d4ddfa6156f632d291255b7f687cc9f82be9bd62f6216d9ca6b8d15e6da4d14
 size 4999711728

model-00029-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5a70e32356fb2d34e72dc2227bac53c6a7bb3f9c9fe914ef4749282f05752e7
+oid sha256:558dcd4052a8f80a09e6a201385783f73ffbb413b612c5f57177ee2489172abd
 size 4966173536

model-00030-of-00030.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19b1cef6d8979e225b57007f438d89807df98bec7a6623a384ae5ff0c5d05340
-size 2101346432
+oid sha256:fe325f3f5568f99a0bbf54bba1b1a7eada4262fa0ea2fe0b40f916bd4f5e24b1
+size 2101379200

model.safetensors.index.json

@@ -1,6 +1,6 @@
 {
   "metadata": {
-    "total_size": 141107412992
+    "total_size": 141107478528
   },
   "weight_map": {
     "lm_head.weight": "model-00030-of-00030.safetensors",

special_tokens_map.json

@@ -7,14 +7,14 @@
     "single_word": false
   },
   "eos_token": {
-    "content": "<|eot_id|>",
+    "content": "<|im_end|>",
     "lstrip": false,
     "normalized": false,
     "rstrip": false,
     "single_word": false
   },
   "pad_token": {
-    "content": "<|end_of_text|>",
+    "content": "<|endoftext|>",
     "lstrip": false,
     "normalized": false,
     "rstrip": false,

tokenizer.json

@@ -2306,6 +2306,24 @@
       "rstrip": false,
       "normalized": false,
       "special": true
+    },
+    {
+      "id": 128256,
+      "content": "<|im_end|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 128257,
+      "content": "<|endoftext|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
     }
   ],
   "normalizer": null,

tokenizer_config.json

@@ -2047,17 +2047,33 @@
       "rstrip": false,
       "single_word": false,
       "special": true
+    },
+    "128256": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "128257": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
     }
   },
   "bos_token": "<|begin_of_text|>",
   "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}{% else %}{{ eos_token }}{% endif %}",
   "clean_up_tokenization_spaces": true,
-  "eos_token": "<|eot_id|>",
+  "eos_token": "<|im_end|>",
   "model_input_names": [
     "input_ids",
     "attention_mask"
   ],
   "model_max_length": 1000000000000000019884624838656,
-  "pad_token": "<|end_of_text|>",
+  "pad_token": "<|endoftext|>",
   "tokenizer_class": "PreTrainedTokenizerFast"
 }

trainer_state.json

@@ -0,0 +1,1141 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.001564945226917,
+  "eval_steps": 500,
+  "global_step": 160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.006259780907668232,
+      "grad_norm": 1.98288817639941,
+      "learning_rate": 5.000000000000001e-07,
+      "loss": 0.6599,
+      "step": 1
+    },
+    {
+      "epoch": 0.012519561815336464,
+      "grad_norm": 2.008513351833145,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 0.6744,
+      "step": 2
+    },
+    {
+      "epoch": 0.018779342723004695,
+      "grad_norm": 2.03144664277006,
+      "learning_rate": 1.5e-06,
+      "loss": 0.6721,
+      "step": 3
+    },
+    {
+      "epoch": 0.025039123630672927,
+      "grad_norm": 1.9480725202469245,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.6577,
+      "step": 4
+    },
+    {
+      "epoch": 0.03129890453834116,
+      "grad_norm": 1.8678118004054254,
+      "learning_rate": 2.5e-06,
+      "loss": 0.6484,
+      "step": 5
+    },
+    {
+      "epoch": 0.03755868544600939,
+      "grad_norm": 1.6583787538868422,
+      "learning_rate": 3e-06,
+      "loss": 0.6174,
+      "step": 6
+    },
+    {
+      "epoch": 0.04381846635367762,
+      "grad_norm": 1.5614405714896737,
+      "learning_rate": 3.5000000000000004e-06,
+      "loss": 0.5896,
+      "step": 7
+    },
+    {
+      "epoch": 0.050078247261345854,
+      "grad_norm": 0.5773143053283745,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.5557,
+      "step": 8
+    },
+    {
+      "epoch": 0.056338028169014086,
+      "grad_norm": 0.3043811484340276,
+      "learning_rate": 4.5e-06,
+      "loss": 0.541,
+      "step": 9
+    },
+    {
+      "epoch": 0.06259780907668232,
+      "grad_norm": 0.8131531353366078,
+      "learning_rate": 5e-06,
+      "loss": 0.5595,
+      "step": 10
+    },
+    {
+      "epoch": 0.06885758998435054,
+      "grad_norm": 0.424180567084822,
+      "learning_rate": 5.500000000000001e-06,
+      "loss": 0.5427,
+      "step": 11
+    },
+    {
+      "epoch": 0.07511737089201878,
+      "grad_norm": 0.2913041969769501,
+      "learning_rate": 6e-06,
+      "loss": 0.5274,
+      "step": 12
+    },
+    {
+      "epoch": 0.081377151799687,
+      "grad_norm": 0.34524917385772347,
+      "learning_rate": 6.5000000000000004e-06,
+      "loss": 0.5337,
+      "step": 13
+    },
+    {
+      "epoch": 0.08763693270735524,
+      "grad_norm": 0.36469195568794854,
+      "learning_rate": 7.000000000000001e-06,
+      "loss": 0.5279,
+      "step": 14
+    },
+    {
+      "epoch": 0.09389671361502347,
+      "grad_norm": 0.35209082489157323,
+      "learning_rate": 7.5e-06,
+      "loss": 0.5296,
+      "step": 15
+    },
+    {
+      "epoch": 0.10015649452269171,
+      "grad_norm": 0.28086156745404856,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.5319,
+      "step": 16
+    },
+    {
+      "epoch": 0.10641627543035993,
+      "grad_norm": 0.5457849868763605,
+      "learning_rate": 8.500000000000002e-06,
+      "loss": 0.5199,
+      "step": 17
+    },
+    {
+      "epoch": 0.11267605633802817,
+      "grad_norm": 0.264594169690208,
+      "learning_rate": 9e-06,
+      "loss": 0.5234,
+      "step": 18
+    },
+    {
+      "epoch": 0.1189358372456964,
+      "grad_norm": 0.2472097021778676,
+      "learning_rate": 9.5e-06,
+      "loss": 0.5248,
+      "step": 19
+    },
+    {
+      "epoch": 0.12519561815336464,
+      "grad_norm": 0.2560549908847749,
+      "learning_rate": 1e-05,
+      "loss": 0.5159,
+      "step": 20
+    },
+    {
+      "epoch": 0.13145539906103287,
+      "grad_norm": 0.4101523009554862,
+      "learning_rate": 1.05e-05,
+      "loss": 0.5058,
+      "step": 21
+    },
+    {
+      "epoch": 0.13771517996870108,
+      "grad_norm": 0.22290433425318873,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.5099,
+      "step": 22
+    },
+    {
+      "epoch": 0.14397496087636932,
+      "grad_norm": 0.2600145857043661,
+      "learning_rate": 1.1500000000000002e-05,
+      "loss": 0.5076,
+      "step": 23
+    },
+    {
+      "epoch": 0.15023474178403756,
+      "grad_norm": 1.1584269063197106,
+      "learning_rate": 1.2e-05,
+      "loss": 0.5133,
+      "step": 24
+    },
+    {
+      "epoch": 0.1564945226917058,
+      "grad_norm": 0.21303015786105067,
+      "learning_rate": 1.25e-05,
+      "loss": 0.5009,
+      "step": 25
+    },
+    {
+      "epoch": 0.162754303599374,
+      "grad_norm": 2.5709430754104345,
+      "learning_rate": 1.3000000000000001e-05,
+      "loss": 0.5067,
+      "step": 26
+    },
+    {
+      "epoch": 0.16901408450704225,
+      "grad_norm": 0.42260631876680255,
+      "learning_rate": 1.3500000000000001e-05,
+      "loss": 0.4951,
+      "step": 27
+    },
+    {
+      "epoch": 0.1752738654147105,
+      "grad_norm": 0.2122989372030049,
+      "learning_rate": 1.4000000000000001e-05,
+      "loss": 0.4968,
+      "step": 28
+    },
+    {
+      "epoch": 0.18153364632237873,
+      "grad_norm": 0.36382001881720555,
+      "learning_rate": 1.45e-05,
+      "loss": 0.5035,
+      "step": 29
+    },
+    {
+      "epoch": 0.18779342723004694,
+      "grad_norm": 0.22094603076455596,
+      "learning_rate": 1.5e-05,
+      "loss": 0.5049,
+      "step": 30
+    },
+    {
+      "epoch": 0.19405320813771518,
+      "grad_norm": 0.17188920546056902,
+      "learning_rate": 1.55e-05,
+      "loss": 0.4979,
+      "step": 31
+    },
+    {
+      "epoch": 0.20031298904538342,
+      "grad_norm": 0.18515458685485783,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.4916,
+      "step": 32
+    },
+    {
+      "epoch": 0.20657276995305165,
+      "grad_norm": 0.783356101762532,
+      "learning_rate": 1.65e-05,
+      "loss": 0.4929,
+      "step": 33
+    },
+    {
+      "epoch": 0.21283255086071987,
+      "grad_norm": 0.19059224326067628,
+      "learning_rate": 1.7000000000000003e-05,
+      "loss": 0.4945,
+      "step": 34
+    },
+    {
+      "epoch": 0.2190923317683881,
+      "grad_norm": 0.2275442577977743,
+      "learning_rate": 1.75e-05,
+      "loss": 0.4936,
+      "step": 35
+    },
+    {
+      "epoch": 0.22535211267605634,
+      "grad_norm": 0.24798149507141237,
+      "learning_rate": 1.8e-05,
+      "loss": 0.4898,
+      "step": 36
+    },
+    {
+      "epoch": 0.23161189358372458,
+      "grad_norm": 0.20682357544778035,
+      "learning_rate": 1.85e-05,
+      "loss": 0.4888,
+      "step": 37
+    },
+    {
+      "epoch": 0.2378716744913928,
+      "grad_norm": 0.19518819682961547,
+      "learning_rate": 1.9e-05,
+      "loss": 0.4899,
+      "step": 38
+    },
+    {
+      "epoch": 0.24413145539906103,
+      "grad_norm": 0.18423871547579748,
+      "learning_rate": 1.9500000000000003e-05,
+      "loss": 0.4868,
+      "step": 39
+    },
+    {
+      "epoch": 0.25039123630672927,
+      "grad_norm": 0.1714820355275791,
+      "learning_rate": 2e-05,
+      "loss": 0.4795,
+      "step": 40
+    },
+    {
+      "epoch": 0.2566510172143975,
+      "grad_norm": 0.19187618384155788,
+      "learning_rate": 2.05e-05,
+      "loss": 0.4821,
+      "step": 41
+    },
+    {
+      "epoch": 0.26291079812206575,
+      "grad_norm": 0.1422378326228944,
+      "learning_rate": 2.1e-05,
+      "loss": 0.4829,
+      "step": 42
+    },
+    {
+      "epoch": 0.26917057902973396,
+      "grad_norm": 0.14724977757162294,
+      "learning_rate": 2.15e-05,
+      "loss": 0.4811,
+      "step": 43
+    },
+    {
+      "epoch": 0.27543035993740217,
+      "grad_norm": 0.16077227738580077,
+      "learning_rate": 2.2000000000000003e-05,
+      "loss": 0.477,
+      "step": 44
+    },
+    {
+      "epoch": 0.28169014084507044,
+      "grad_norm": 0.15993679259901028,
+      "learning_rate": 2.25e-05,
+      "loss": 0.4789,
+      "step": 45
+    },
+    {
+      "epoch": 0.28794992175273865,
+      "grad_norm": 0.14385134377084383,
+      "learning_rate": 2.3000000000000003e-05,
+      "loss": 0.4641,
+      "step": 46
+    },
+    {
+      "epoch": 0.2942097026604069,
+      "grad_norm": 0.14244559356804792,
+      "learning_rate": 2.35e-05,
+      "loss": 0.4767,
+      "step": 47
+    },
+    {
+      "epoch": 0.3004694835680751,
+      "grad_norm": 0.1481660114240819,
+      "learning_rate": 2.4e-05,
+      "loss": 0.4759,
+      "step": 48
+    },
+    {
+      "epoch": 0.30672926447574334,
+      "grad_norm": 0.14195363156015162,
+      "learning_rate": 2.45e-05,
+      "loss": 0.471,
+      "step": 49
+    },
+    {
+      "epoch": 0.3129890453834116,
+      "grad_norm": 0.15220552720898642,
+      "learning_rate": 2.5e-05,
+      "loss": 0.4715,
+      "step": 50
+    },
+    {
+      "epoch": 0.3192488262910798,
+      "grad_norm": 0.13409784658365015,
+      "learning_rate": 2.5500000000000003e-05,
+      "loss": 0.4692,
+      "step": 51
+    },
+    {
+      "epoch": 0.325508607198748,
+      "grad_norm": 0.13766694658848178,
+      "learning_rate": 2.6000000000000002e-05,
+      "loss": 0.47,
+      "step": 52
+    },
+    {
+      "epoch": 0.3317683881064163,
+      "grad_norm": 0.13097864679643595,
+      "learning_rate": 2.6500000000000004e-05,
+      "loss": 0.4651,
+      "step": 53
+    },
+    {
+      "epoch": 0.3380281690140845,
+      "grad_norm": 0.13207003285729219,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 0.4714,
+      "step": 54
+    },
+    {
+      "epoch": 0.3442879499217527,
+      "grad_norm": 0.14128427173382038,
+      "learning_rate": 2.7500000000000004e-05,
+      "loss": 0.4719,
+      "step": 55
+    },
+    {
+      "epoch": 0.350547730829421,
+      "grad_norm": 0.13599048333974484,
+      "learning_rate": 2.8000000000000003e-05,
+      "loss": 0.4657,
+      "step": 56
+    },
+    {
+      "epoch": 0.3568075117370892,
+      "grad_norm": 0.1547358977814178,
+      "learning_rate": 2.8499999999999998e-05,
+      "loss": 0.4599,
+      "step": 57
+    },
+    {
+      "epoch": 0.36306729264475746,
+      "grad_norm": 0.1357320992255676,
+      "learning_rate": 2.9e-05,
+      "loss": 0.4615,
+      "step": 58
+    },
+    {
+      "epoch": 0.36932707355242567,
+      "grad_norm": 0.14465717873045295,
+      "learning_rate": 2.95e-05,
+      "loss": 0.4738,
+      "step": 59
+    },
+    {
+      "epoch": 0.3755868544600939,
+      "grad_norm": 0.5900603203611421,
+      "learning_rate": 3e-05,
+      "loss": 0.4702,
+      "step": 60
+    },
+    {
+      "epoch": 0.38184663536776214,
+      "grad_norm": 0.17729474902277623,
+      "learning_rate": 3.05e-05,
+      "loss": 0.4592,
+      "step": 61
+    },
+    {
+      "epoch": 0.38810641627543035,
+      "grad_norm": 0.22055664690525556,
+      "learning_rate": 3.1e-05,
+      "loss": 0.47,
+      "step": 62
+    },
+    {
+      "epoch": 0.39436619718309857,
+      "grad_norm": 0.22917133262033845,
+      "learning_rate": 3.15e-05,
+      "loss": 0.4668,
+      "step": 63
+    },
+    {
+      "epoch": 0.40062597809076683,
+      "grad_norm": 0.23278911760289017,
+      "learning_rate": 3.2000000000000005e-05,
+      "loss": 0.4691,
+      "step": 64
+    },
+    {
+      "epoch": 0.40688575899843504,
+      "grad_norm": 0.23911939507472177,
+      "learning_rate": 3.2500000000000004e-05,
+      "loss": 0.4662,
+      "step": 65
+    },
+    {
+      "epoch": 0.4131455399061033,
+      "grad_norm": 0.19447041878105836,
+      "learning_rate": 3.3e-05,
+      "loss": 0.4633,
+      "step": 66
+    },
+    {
+      "epoch": 0.4194053208137715,
+      "grad_norm": 0.17498726962496755,
+      "learning_rate": 3.35e-05,
+      "loss": 0.4654,
+      "step": 67
+    },
+    {
+      "epoch": 0.42566510172143973,
+      "grad_norm": 0.24918375228266929,
+      "learning_rate": 3.4000000000000007e-05,
+      "loss": 0.477,
+      "step": 68
+    },
+    {
+      "epoch": 0.431924882629108,
+      "grad_norm": 0.2850664865678729,
+      "learning_rate": 3.45e-05,
+      "loss": 0.4648,
+      "step": 69
+    },
+    {
+      "epoch": 0.4381846635367762,
+      "grad_norm": 0.27562629972396513,
+      "learning_rate": 3.5e-05,
+      "loss": 0.4667,
+      "step": 70
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 0.22637202856522412,
+      "learning_rate": 3.55e-05,
+      "loss": 0.4653,
+      "step": 71
+    },
+    {
+      "epoch": 0.4507042253521127,
+      "grad_norm": 0.2295442026728235,
+      "learning_rate": 3.6e-05,
+      "loss": 0.4622,
+      "step": 72
+    },
+    {
+      "epoch": 0.4569640062597809,
+      "grad_norm": 0.26572612655057165,
+      "learning_rate": 3.65e-05,
+      "loss": 0.4673,
+      "step": 73
+    },
+    {
+      "epoch": 0.46322378716744916,
+      "grad_norm": 0.2496817546620412,
+      "learning_rate": 3.7e-05,
+      "loss": 0.4611,
+      "step": 74
+    },
+    {
+      "epoch": 0.4694835680751174,
+      "grad_norm": 0.21430723659191686,
+      "learning_rate": 3.7500000000000003e-05,
+      "loss": 0.4637,
+      "step": 75
+    },
+    {
+      "epoch": 0.4757433489827856,
+      "grad_norm": 0.1799606207168491,
+      "learning_rate": 3.8e-05,
+      "loss": 0.4612,
+      "step": 76
+    },
+    {
+      "epoch": 0.48200312989045385,
+      "grad_norm": 0.2329269891744439,
+      "learning_rate": 3.85e-05,
+      "loss": 0.4569,
+      "step": 77
+    },
+    {
+      "epoch": 0.48826291079812206,
+      "grad_norm": 0.2859704851548014,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.4677,
+      "step": 78
+    },
+    {
+      "epoch": 0.4945226917057903,
+      "grad_norm": 0.3153100598444141,
+      "learning_rate": 3.9500000000000005e-05,
+      "loss": 0.465,
+      "step": 79
+    },
+    {
+      "epoch": 0.5007824726134585,
+      "grad_norm": 0.3165950932566608,
+      "learning_rate": 4e-05,
+      "loss": 0.4755,
+      "step": 80
+    },
+    {
+      "epoch": 0.5070422535211268,
+      "grad_norm": 0.3018577292754275,
+      "learning_rate": 4.05e-05,
+      "loss": 0.464,
+      "step": 81
+    },
+    {
+      "epoch": 0.513302034428795,
+      "grad_norm": 0.39363558044861696,
+      "learning_rate": 4.1e-05,
+      "loss": 0.4701,
+      "step": 82
+    },
+    {
+      "epoch": 0.5195618153364632,
+      "grad_norm": 0.44171413078007776,
+      "learning_rate": 4.15e-05,
+      "loss": 0.4697,
+      "step": 83
+    },
+    {
+      "epoch": 0.5258215962441315,
+      "grad_norm": 0.4086449510625894,
+      "learning_rate": 4.2e-05,
+      "loss": 0.4611,
+      "step": 84
+    },
+    {
+      "epoch": 0.5320813771517997,
+      "grad_norm": 0.3156689305434587,
+      "learning_rate": 4.25e-05,
+      "loss": 0.4633,
+      "step": 85
+    },
+    {
+      "epoch": 0.5383411580594679,
+      "grad_norm": 0.37582415992669976,
+      "learning_rate": 4.3e-05,
+      "loss": 0.4689,
+      "step": 86
+    },
+    {
+      "epoch": 0.5446009389671361,
+      "grad_norm": 0.3751728997948819,
+      "learning_rate": 4.35e-05,
+      "loss": 0.4658,
+      "step": 87
+    },
+    {
+      "epoch": 0.5508607198748043,
+      "grad_norm": 0.2622604607003995,
+      "learning_rate": 4.4000000000000006e-05,
+      "loss": 0.4641,
+      "step": 88
+    },
+    {
+      "epoch": 0.5571205007824727,
+      "grad_norm": 0.27806769516567914,
+      "learning_rate": 4.4500000000000004e-05,
+      "loss": 0.4689,
+      "step": 89
+    },
+    {
+      "epoch": 0.5633802816901409,
+      "grad_norm": 0.37193892514568727,
+      "learning_rate": 4.5e-05,
+      "loss": 0.4645,
+      "step": 90
+    },
+    {
+      "epoch": 0.5696400625978091,
+      "grad_norm": 0.319234610988282,
+      "learning_rate": 4.55e-05,
+      "loss": 0.4697,
+      "step": 91
+    },
+    {
+      "epoch": 0.5758998435054773,
+      "grad_norm": 0.24391835650924631,
+      "learning_rate": 4.600000000000001e-05,
+      "loss": 0.4605,
+      "step": 92
+    },
+    {
+      "epoch": 0.5821596244131455,
+      "grad_norm": 0.3860119064167233,
+      "learning_rate": 4.6500000000000005e-05,
+      "loss": 0.4721,
+      "step": 93
+    },
+    {
+      "epoch": 0.5884194053208138,
+      "grad_norm": 0.43978262147491526,
+      "learning_rate": 4.7e-05,
+      "loss": 0.4692,
+      "step": 94
+    },
+    {
+      "epoch": 0.594679186228482,
+      "grad_norm": 0.2869109051387356,
+      "learning_rate": 4.75e-05,
+      "loss": 0.4644,
+      "step": 95
+    },
+    {
+      "epoch": 0.6009389671361502,
+      "grad_norm": 0.33046074741721215,
+      "learning_rate": 4.8e-05,
+      "loss": 0.4711,
+      "step": 96
+    },
+    {
+      "epoch": 0.6071987480438185,
+      "grad_norm": 0.3874189152162858,
+      "learning_rate": 4.85e-05,
+      "loss": 0.4694,
+      "step": 97
+    },
+    {
+      "epoch": 0.6134585289514867,
+      "grad_norm": 0.46318630797414556,
+      "learning_rate": 4.9e-05,
+      "loss": 0.4741,
+      "step": 98
+    },
+    {
+      "epoch": 0.6197183098591549,
+      "grad_norm": 0.6037444606802089,
+      "learning_rate": 4.9500000000000004e-05,
+      "loss": 0.4754,
+      "step": 99
+    },
+    {
+      "epoch": 0.6259780907668232,
+      "grad_norm": 0.5037059436389102,
+      "learning_rate": 5e-05,
+      "loss": 0.4739,
+      "step": 100
+    },
+    {
+      "epoch": 0.6322378716744914,
+      "grad_norm": 0.5631190436137139,
+      "learning_rate": 4.9997404092249336e-05,
+      "loss": 0.4699,
+      "step": 101
+    },
+    {
+      "epoch": 0.6384976525821596,
+      "grad_norm": 0.39119483297638863,
+      "learning_rate": 4.998961690809628e-05,
+      "loss": 0.4703,
+      "step": 102
+    },
+    {
+      "epoch": 0.6447574334898278,
+      "grad_norm": 0.40196303529424704,
+      "learning_rate": 4.997664006472579e-05,
+      "loss": 0.4749,
+      "step": 103
+    },
+    {
+      "epoch": 0.651017214397496,
+      "grad_norm": 0.3397733110278162,
+      "learning_rate": 4.9958476257072914e-05,
+      "loss": 0.4654,
+      "step": 104
+    },
+    {
+      "epoch": 0.6572769953051644,
+      "grad_norm": 0.2670846226151608,
+      "learning_rate": 4.993512925726319e-05,
+      "loss": 0.4716,
+      "step": 105
+    },
+    {
+      "epoch": 0.6635367762128326,
+      "grad_norm": 0.36681659702689784,
+      "learning_rate": 4.990660391382923e-05,
+      "loss": 0.4704,
+      "step": 106
+    },
+    {
+      "epoch": 0.6697965571205008,
+      "grad_norm": 0.26058292855009557,
+      "learning_rate": 4.987290615070385e-05,
+      "loss": 0.4647,
+      "step": 107
+    },
+    {
+      "epoch": 0.676056338028169,
+      "grad_norm": 0.25205128219384887,
+      "learning_rate": 4.983404296598979e-05,
+      "loss": 0.4725,
+      "step": 108
+    },
+    {
+      "epoch": 0.6823161189358372,
+      "grad_norm": 0.3208687051782515,
+      "learning_rate": 4.9790022430506463e-05,
+      "loss": 0.471,
+      "step": 109
+    },
+    {
+      "epoch": 0.6885758998435054,
+      "grad_norm": 0.2306209439140453,
+      "learning_rate": 4.974085368611381e-05,
+      "loss": 0.473,
+      "step": 110
+    },
+    {
+      "epoch": 0.6948356807511737,
+      "grad_norm": 0.21458192536569118,
+      "learning_rate": 4.968654694381379e-05,
+      "loss": 0.4692,
+      "step": 111
+    },
+    {
+      "epoch": 0.701095461658842,
+      "grad_norm": 0.24400329234341836,
+      "learning_rate": 4.962711348162987e-05,
+      "loss": 0.4742,
+      "step": 112
+    },
+    {
+      "epoch": 0.7073552425665102,
+      "grad_norm": 0.5445701250609367,
+      "learning_rate": 4.956256564226487e-05,
+      "loss": 0.4677,
+      "step": 113
+    },
+    {
+      "epoch": 0.7136150234741784,
+      "grad_norm": 0.2485591152431222,
+      "learning_rate": 4.949291683053769e-05,
+      "loss": 0.478,
+      "step": 114
+    },
+    {
+      "epoch": 0.7198748043818466,
+      "grad_norm": 0.2683190648451619,
+      "learning_rate": 4.941818151059956e-05,
+      "loss": 0.468,
+      "step": 115
+    },
+    {
+      "epoch": 0.7261345852895149,
+      "grad_norm": 0.17377296116604452,
+      "learning_rate": 4.933837520293017e-05,
+      "loss": 0.4682,
+      "step": 116
+    },
+    {
+      "epoch": 0.7323943661971831,
+      "grad_norm": 0.19892874090328266,
+      "learning_rate": 4.9253514481114535e-05,
+      "loss": 0.4716,
+      "step": 117
+    },
+    {
+      "epoch": 0.7386541471048513,
+      "grad_norm": 0.22470516800088272,
+      "learning_rate": 4.91636169684011e-05,
+      "loss": 0.4807,
+      "step": 118
+    },
+    {
+      "epoch": 0.7449139280125195,
+      "grad_norm": 0.23033947133081567,
+      "learning_rate": 4.906870133404187e-05,
+      "loss": 0.4721,
+      "step": 119
+    },
+    {
+      "epoch": 0.7511737089201878,
+      "grad_norm": 0.2764527709442302,
+      "learning_rate": 4.896878728941531e-05,
+      "loss": 0.4693,
+      "step": 120
+    },
+    {
+      "epoch": 0.7574334898278561,
+      "grad_norm": 0.28746556965081915,
+      "learning_rate": 4.8863895583932836e-05,
+      "loss": 0.4767,
+      "step": 121
+    },
+    {
+      "epoch": 0.7636932707355243,
+      "grad_norm": 0.32061574884194566,
+      "learning_rate": 4.875404800072977e-05,
+      "loss": 0.4643,
+      "step": 122
+    },
+    {
+      "epoch": 0.7699530516431925,
+      "grad_norm": 0.34181281337669966,
+      "learning_rate": 4.86392673521415e-05,
+      "loss": 0.4602,
+      "step": 123
+    },
+    {
+      "epoch": 0.7762128325508607,
+      "grad_norm": 0.30941984507586506,
+      "learning_rate": 4.8519577474966074e-05,
+      "loss": 0.4711,
+      "step": 124
+    },
+    {
+      "epoch": 0.7824726134585289,
+      "grad_norm": 0.23600978038755785,
+      "learning_rate": 4.839500322551386e-05,
+      "loss": 0.4696,
+      "step": 125
+    },
+    {
+      "epoch": 0.7887323943661971,
+      "grad_norm": 0.2577164285099203,
+      "learning_rate": 4.8265570474445636e-05,
+      "loss": 0.4644,
+      "step": 126
+    },
+    {
+      "epoch": 0.7949921752738655,
+      "grad_norm": 0.27823451721774306,
+      "learning_rate": 4.813130610139994e-05,
+      "loss": 0.479,
+      "step": 127
+    },
+    {
+      "epoch": 0.8012519561815337,
+      "grad_norm": 0.22061524932206344,
+      "learning_rate": 4.7992237989410904e-05,
+      "loss": 0.4711,
+      "step": 128
+    },
+    {
+      "epoch": 0.8075117370892019,
+      "grad_norm": 0.20216340578684158,
+      "learning_rate": 4.784839501911771e-05,
+      "loss": 0.468,
+      "step": 129
+    },
+    {
+      "epoch": 0.8137715179968701,
+      "grad_norm": 0.27542745611047786,
+      "learning_rate": 4.7699807062766876e-05,
+      "loss": 0.4754,
+      "step": 130
+    },
+    {
+      "epoch": 0.8200312989045383,
+      "grad_norm": 0.21954180738847087,
+      "learning_rate": 4.75465049780086e-05,
+      "loss": 0.4595,
+      "step": 131
+    },
+    {
+      "epoch": 0.8262910798122066,
+      "grad_norm": 0.19430624161738624,
+      "learning_rate": 4.738852060148849e-05,
+      "loss": 0.4747,
+      "step": 132
+    },
+    {
+      "epoch": 0.8325508607198748,
+      "grad_norm": 0.1884671644058954,
+      "learning_rate": 4.722588674223594e-05,
+      "loss": 0.4748,
+      "step": 133
+    },
+    {
+      "epoch": 0.838810641627543,
+      "grad_norm": 0.20913369047927102,
+      "learning_rate": 4.7058637174850604e-05,
+      "loss": 0.4653,
+      "step": 134
+    },
+    {
+      "epoch": 0.8450704225352113,
+      "grad_norm": 0.19564021089464265,
+      "learning_rate": 4.688680663248837e-05,
+      "loss": 0.4644,
+      "step": 135
+    },
+    {
+      "epoch": 0.8513302034428795,
+      "grad_norm": 0.17437877798570775,
+      "learning_rate": 4.671043079964815e-05,
+      "loss": 0.4666,
+      "step": 136
+    },
+    {
+      "epoch": 0.8575899843505478,
+      "grad_norm": 0.18658537333186465,
+      "learning_rate": 4.652954630476127e-05,
+      "loss": 0.463,
+      "step": 137
+    },
+    {
+      "epoch": 0.863849765258216,
+      "grad_norm": 0.1916983418252378,
+      "learning_rate": 4.634419071258472e-05,
+      "loss": 0.4801,
+      "step": 138
+    },
+    {
+      "epoch": 0.8701095461658842,
+      "grad_norm": 0.18269150591223743,
+      "learning_rate": 4.615440251639995e-05,
+      "loss": 0.465,
+      "step": 139
+    },
+    {
+      "epoch": 0.8763693270735524,
+      "grad_norm": 0.19124021712384207,
+      "learning_rate": 4.5960221130018946e-05,
+      "loss": 0.4624,
+      "step": 140
+    },
+    {
+      "epoch": 0.8826291079812206,
+      "grad_norm": 0.17751289300487907,
+      "learning_rate": 4.576168687959895e-05,
+      "loss": 0.4667,
+      "step": 141
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 0.16256598664527863,
+      "learning_rate": 4.555884099526794e-05,
+      "loss": 0.4724,
+      "step": 142
+    },
+    {
+      "epoch": 0.8951486697965572,
+      "grad_norm": 0.17306660659668968,
+      "learning_rate": 4.535172560256218e-05,
+      "loss": 0.4764,
+      "step": 143
+    },
+    {
+      "epoch": 0.9014084507042254,
+      "grad_norm": 0.15311694878287935,
+      "learning_rate": 4.5140383713677916e-05,
+      "loss": 0.4633,
+      "step": 144
+    },
+    {
+      "epoch": 0.9076682316118936,
+      "grad_norm": 0.16327033693685952,
+      "learning_rate": 4.492485921853894e-05,
+      "loss": 0.4626,
+      "step": 145
+    },
+    {
+      "epoch": 0.9139280125195618,
+      "grad_norm": 0.1577015015575217,
+      "learning_rate": 4.4705196875681854e-05,
+      "loss": 0.465,
+      "step": 146
+    },
+    {
+      "epoch": 0.92018779342723,
+      "grad_norm": 0.14976303947345634,
+      "learning_rate": 4.448144230296093e-05,
+      "loss": 0.4732,
+      "step": 147
+    },
+    {
+      "epoch": 0.9264475743348983,
+      "grad_norm": 0.1799041852337434,
+      "learning_rate": 4.425364196807451e-05,
+      "loss": 0.4638,
+      "step": 148
+    },
+    {
+      "epoch": 0.9327073552425665,
+      "grad_norm": 0.25582934784311545,
+      "learning_rate": 4.402184317891501e-05,
+      "loss": 0.4687,
+      "step": 149
+    },
+    {
+      "epoch": 0.9389671361502347,
+      "grad_norm": 0.14767269207211267,
+      "learning_rate": 4.37860940737443e-05,
+      "loss": 0.4622,
+      "step": 150
+    },
+    {
+      "epoch": 0.945226917057903,
+      "grad_norm": 0.18510146862998086,
+      "learning_rate": 4.354644361119672e-05,
+      "loss": 0.4714,
+      "step": 151
+    },
+    {
+      "epoch": 0.9514866979655712,
+      "grad_norm": 0.1834113544053396,
+      "learning_rate": 4.330294156011172e-05,
+      "loss": 0.4665,
+      "step": 152
+    },
+    {
+      "epoch": 0.9577464788732394,
+      "grad_norm": 0.16106024098596552,
+      "learning_rate": 4.305563848919824e-05,
+      "loss": 0.4612,
+      "step": 153
+    },
+    {
+      "epoch": 0.9640062597809077,
+      "grad_norm": 0.1582714001537092,
+      "learning_rate": 4.2804585756532965e-05,
+      "loss": 0.4656,
+      "step": 154
+    },
+    {
+      "epoch": 0.9702660406885759,
+      "grad_norm": 0.1838011411088347,
+      "learning_rate": 4.254983549889467e-05,
+      "loss": 0.4585,
+      "step": 155
+    },
+    {
+      "epoch": 0.9765258215962441,
+      "grad_norm": 0.22256207898681857,
+      "learning_rate": 4.2291440620936796e-05,
+      "loss": 0.4712,
+      "step": 156
+    },
+    {
+      "epoch": 0.9827856025039123,
+      "grad_norm": 0.16842112143070276,
+      "learning_rate": 4.2029454784200676e-05,
+      "loss": 0.4691,
+      "step": 157
+    },
+    {
+      "epoch": 0.9890453834115805,
+      "grad_norm": 0.15122887832488566,
+      "learning_rate": 4.176393239597144e-05,
+      "loss": 0.4778,
+      "step": 158
+    },
+    {
+      "epoch": 0.9953051643192489,
+      "grad_norm": 0.1902639072378955,
+      "learning_rate": 4.149492859797912e-05,
+      "loss": 0.4688,
+      "step": 159
+    },
+    {
+      "epoch": 1.001564945226917,
+      "grad_norm": 0.17200971150006397,
+      "learning_rate": 4.122249925494726e-05,
+      "loss": 0.464,
+      "step": 160
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 318,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 16,
+  "total_flos": 1.0356139229184e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6838be8a1bc22ac23837c7b4fd5b41797dd1b59a6f54d379a9faeeae90388906
+size 7800

zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)