Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+dataset-0/checkpoint-1064/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+dataset-0/checkpoint-300/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+dataset-0/checkpoint-600/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+dataset-0/checkpoint-900/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+dataset-0/last/tokenizer.json filter=lfs diff=lfs merge=lfs -text

dataset-0/all_results.json

@@ -0,0 +1,17 @@
+{
+    "dataset": 0,
+    "epoch": 4.0,
+    "eval_logits/chosen": -2.9694149494171143,
+    "eval_logits/rejected": -4.259973526000977,
+    "eval_logps/chosen": -733.276611328125,
+    "eval_logps/rejected": -979.574462890625,
+    "eval_loss": 1.8394471501537168e-09,
+    "eval_rewards/accuracies": 1.0,
+    "eval_rewards/chosen": 8.114361763000488,
+    "eval_rewards/margins": 60.553192138671875,
+    "eval_rewards/rejected": -52.42021179199219,
+    "eval_runtime": 8.4622,
+    "eval_samples_per_second": 177.258,
+    "eval_score": -0.7999978065490723,
+    "eval_steps_per_second": 5.554
+}

dataset-0/checkpoint-1064/added_tokens.json

@@ -0,0 +1,5 @@
+{
+  "<|endoftext|>": 151643,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644
+}

dataset-0/checkpoint-1064/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "Qwen/Qwen2-0.5B-Instruct",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.49.0",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

dataset-0/checkpoint-1064/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.1,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.49.0"
+}

dataset-0/checkpoint-1064/global_step1064/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb41c124e0049d045f80beb7f42c838172839ce64faa603b39c4a71ed3334741
+size 1482103280

dataset-0/checkpoint-1064/global_step1064/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9317be36b0188b7d0cabd74bc88a37eae8583f1eb9fca8452c732bfb863ac89e
+size 1482103280

dataset-0/checkpoint-1064/global_step1064/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c871fe241e3ee0c36dbf60f2197b7296f1fe8467072cd8bac69f469a67b8d12
+size 1482103280

dataset-0/checkpoint-1064/global_step1064/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd37c1ec03f596df7e6ab27c8571fbd9176aa9cb357983058b4f95732e5555c2
+size 1482103280

dataset-0/checkpoint-1064/global_step1064/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:273474a17de910f2ce838d710180ddd4cf5c5d0c045db20b089073b00b418a38
+size 142728

dataset-0/checkpoint-1064/global_step1064/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87df774a8a68d113a776d38e560d20d3a3c3eb0e451a21b14e8dfc9901c50b7e
+size 142664

dataset-0/checkpoint-1064/global_step1064/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f92a45d4d321ce6bbaa5e333c0f876250f8f234ecef3b53dff49cd3bd5602a81
+size 142664

dataset-0/checkpoint-1064/global_step1064/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:709f022824915a9e2a3db761537e1ba5da9b9ee138ea07043e5de559b14be90b
+size 142664

dataset-0/checkpoint-1064/latest

@@ -0,0 +1 @@
+global_step1064

dataset-0/checkpoint-1064/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f479a6e57886603c4731130ffc03926fa41d8302d97aeff35935bdc6de8a4b0
+size 988097824

dataset-0/checkpoint-1064/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb7ef461173a038e154f7168466a1a1cbf011caf013ac2c15a132c95d2c27fa3
+size 15024

dataset-0/checkpoint-1064/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5de94b1cfa98d2197934e9a8b4cf2db1290ca27f32e59d5f9055cd84d082a53
+size 15024

dataset-0/checkpoint-1064/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eeeae7b81b3ee26f32948132bc00a7bdb72d1903214f505d2fa94fda0c0753c
+size 15024

dataset-0/checkpoint-1064/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7da27fce9c3ca879a62610ae6a1a15d90818bbc1dd3f4a136dbb97608a01015
+size 15024

dataset-0/checkpoint-1064/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa7b3c71b5f5f0d87503ac4d850be0fea1a0582e2734991336a1973c6c64df67
+size 1064

dataset-0/checkpoint-1064/special_tokens_map.json

@@ -0,0 +1,20 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

dataset-0/checkpoint-1064/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcfe42da0a4497e8b2b172c1f9f4ec423a46dc12907f4349c55025f670422ba9
+size 11418266

dataset-0/checkpoint-1064/tokenizer_config.json

@@ -0,0 +1,44 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

dataset-0/checkpoint-1064/trainer_state.json

@@ -0,0 +1,913 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 200,
+  "global_step": 1064,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.07518796992481203,
+      "grad_norm": 0.00031717625653774825,
+      "learning_rate": 4.906015037593986e-06,
+      "logits/chosen": -3.08984375,
+      "logits/rejected": -3.359375,
+      "logps/chosen": -759.5999755859375,
+      "logps/rejected": -601.7999877929688,
+      "loss": 0.0505,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": 3.9212889671325684,
+      "rewards/margins": 19.23046875,
+      "rewards/rejected": -15.315332412719727,
+      "step": 20
+    },
+    {
+      "epoch": 0.15037593984962405,
+      "grad_norm": 2.5674383538843824e-13,
+      "learning_rate": 4.81203007518797e-06,
+      "logits/chosen": -3.37890625,
+      "logits/rejected": -3.901562452316284,
+      "logps/chosen": -797.4000244140625,
+      "logps/rejected": -834.2000122070312,
+      "loss": 0.0018,
+      "rewards/accuracies": 0.9984375238418579,
+      "rewards/chosen": 0.5521484613418579,
+      "rewards/margins": 38.98125076293945,
+      "rewards/rejected": -38.412498474121094,
+      "step": 40
+    },
+    {
+      "epoch": 0.22556390977443608,
+      "grad_norm": 4.184555188916154e-12,
+      "learning_rate": 4.718045112781955e-06,
+      "logits/chosen": -3.237499952316284,
+      "logits/rejected": -4.165625095367432,
+      "logps/chosen": -774.5999755859375,
+      "logps/rejected": -944.7999877929688,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 3.255664110183716,
+      "rewards/margins": 53.625,
+      "rewards/rejected": -50.36249923706055,
+      "step": 60
+    },
+    {
+      "epoch": 0.3007518796992481,
+      "grad_norm": 3.674299650107461e-07,
+      "learning_rate": 4.62406015037594e-06,
+      "logits/chosen": -3.0601563453674316,
+      "logits/rejected": -4.301562309265137,
+      "logps/chosen": -744.2000122070312,
+      "logps/rejected": -968.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 6.635937690734863,
+      "rewards/margins": 58.375,
+      "rewards/rejected": -51.724998474121094,
+      "step": 80
+    },
+    {
+      "epoch": 0.37593984962406013,
+      "grad_norm": 2.9469468290533765e-13,
+      "learning_rate": 4.530075187969925e-06,
+      "logits/chosen": -3.035937547683716,
+      "logits/rejected": -4.3359375,
+      "logps/chosen": -754.0,
+      "logps/rejected": -981.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 6.212500095367432,
+      "rewards/margins": 58.912498474121094,
+      "rewards/rejected": -52.724998474121094,
+      "step": 100
+    },
+    {
+      "epoch": 0.45112781954887216,
+      "grad_norm": 2.450421719466682e-10,
+      "learning_rate": 4.43609022556391e-06,
+      "logits/chosen": -3.0367188453674316,
+      "logits/rejected": -4.3203125,
+      "logps/chosen": -740.0,
+      "logps/rejected": -961.7999877929688,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 6.285937309265137,
+      "rewards/margins": 57.474998474121094,
+      "rewards/rejected": -51.1875,
+      "step": 120
+    },
+    {
+      "epoch": 0.5263157894736842,
+      "grad_norm": 1.6351303999698253e-10,
+      "learning_rate": 4.342105263157895e-06,
+      "logits/chosen": -3.0234375,
+      "logits/rejected": -4.3046875,
+      "logps/chosen": -754.0,
+      "logps/rejected": -969.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 6.207812309265137,
+      "rewards/margins": 57.662498474121094,
+      "rewards/rejected": -51.412498474121094,
+      "step": 140
+    },
+    {
+      "epoch": 0.6015037593984962,
+      "grad_norm": 1.5915083665281199e-09,
+      "learning_rate": 4.24812030075188e-06,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -4.318749904632568,
+      "logps/chosen": -748.2000122070312,
+      "logps/rejected": -958.7999877929688,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 6.248437404632568,
+      "rewards/margins": 57.525001525878906,
+      "rewards/rejected": -51.25,
+      "step": 160
+    },
+    {
+      "epoch": 0.6766917293233082,
+      "grad_norm": 4.297014302694241e-12,
+      "learning_rate": 4.1541353383458646e-06,
+      "logits/chosen": -3.0406250953674316,
+      "logits/rejected": -4.279687404632568,
+      "logps/chosen": -741.4000244140625,
+      "logps/rejected": -961.4000244140625,
+      "loss": 0.0132,
+      "rewards/accuracies": 0.9906250238418579,
+      "rewards/chosen": 6.785937309265137,
+      "rewards/margins": 57.4375,
+      "rewards/rejected": -50.625,
+      "step": 180
+    },
+    {
+      "epoch": 0.7518796992481203,
+      "grad_norm": 9.095973454791659e-12,
+      "learning_rate": 4.06015037593985e-06,
+      "logits/chosen": -3.055468797683716,
+      "logits/rejected": -4.318749904632568,
+      "logps/chosen": -733.0,
+      "logps/rejected": -968.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.356249809265137,
+      "rewards/margins": 59.5625,
+      "rewards/rejected": -52.224998474121094,
+      "step": 200
+    },
+    {
+      "epoch": 0.7518796992481203,
+      "eval_logits/chosen": -3.058178186416626,
+      "eval_logits/rejected": -4.299867153167725,
+      "eval_logps/chosen": -741.7021484375,
+      "eval_logps/rejected": -972.0850830078125,
+      "eval_loss": 2.180002622864663e-09,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/chosen": 7.262632846832275,
+      "eval_rewards/margins": 58.87765884399414,
+      "eval_rewards/rejected": -51.62765884399414,
+      "eval_runtime": 8.4937,
+      "eval_samples_per_second": 176.601,
+      "eval_score": -0.6606304049491882,
+      "eval_steps_per_second": 5.533,
+      "step": 200
+    },
+    {
+      "epoch": 0.8270676691729323,
+      "grad_norm": 2.3694597465927517e-14,
+      "learning_rate": 3.966165413533835e-06,
+      "logits/chosen": -3.057812452316284,
+      "logits/rejected": -4.2890625,
+      "logps/chosen": -739.0,
+      "logps/rejected": -973.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.339062690734863,
+      "rewards/margins": 58.67499923706055,
+      "rewards/rejected": -51.32500076293945,
+      "step": 220
+    },
+    {
+      "epoch": 0.9022556390977443,
+      "grad_norm": 4.7239515256796625e-09,
+      "learning_rate": 3.87218045112782e-06,
+      "logits/chosen": -3.0648436546325684,
+      "logits/rejected": -4.301562309265137,
+      "logps/chosen": -739.4000244140625,
+      "logps/rejected": -980.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.296875,
+      "rewards/margins": 59.57500076293945,
+      "rewards/rejected": -52.25,
+      "step": 240
+    },
+    {
+      "epoch": 0.9774436090225563,
+      "grad_norm": 1.1010824468107971e-09,
+      "learning_rate": 3.778195488721805e-06,
+      "logits/chosen": -3.059375047683716,
+      "logits/rejected": -4.317187309265137,
+      "logps/chosen": -730.0,
+      "logps/rejected": -975.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.485937595367432,
+      "rewards/margins": 60.162498474121094,
+      "rewards/rejected": -52.67499923706055,
+      "step": 260
+    },
+    {
+      "epoch": 1.0526315789473684,
+      "grad_norm": 5.510844973332184e-09,
+      "learning_rate": 3.6842105263157896e-06,
+      "logits/chosen": -3.06640625,
+      "logits/rejected": -4.3046875,
+      "logps/chosen": -736.4000244140625,
+      "logps/rejected": -961.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.443749904632568,
+      "rewards/margins": 58.42499923706055,
+      "rewards/rejected": -51.025001525878906,
+      "step": 280
+    },
+    {
+      "epoch": 1.1278195488721805,
+      "grad_norm": 1.0805968862929627e-08,
+      "learning_rate": 3.590225563909775e-06,
+      "logits/chosen": -3.067187547683716,
+      "logits/rejected": -4.329687595367432,
+      "logps/chosen": -737.0,
+      "logps/rejected": -978.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.373437404632568,
+      "rewards/margins": 59.849998474121094,
+      "rewards/rejected": -52.474998474121094,
+      "step": 300
+    },
+    {
+      "epoch": 1.2030075187969924,
+      "grad_norm": 1.8280615029499892e-13,
+      "learning_rate": 3.4962406015037596e-06,
+      "logits/chosen": -3.063281297683716,
+      "logits/rejected": -4.317187309265137,
+      "logps/chosen": -739.2000122070312,
+      "logps/rejected": -977.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.349999904632568,
+      "rewards/margins": 59.67499923706055,
+      "rewards/rejected": -52.337501525878906,
+      "step": 320
+    },
+    {
+      "epoch": 1.2781954887218046,
+      "grad_norm": 8.685091807936255e-12,
+      "learning_rate": 3.4022556390977448e-06,
+      "logits/chosen": -3.057812452316284,
+      "logits/rejected": -4.293749809265137,
+      "logps/chosen": -740.4000244140625,
+      "logps/rejected": -961.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.3203125,
+      "rewards/margins": 58.587501525878906,
+      "rewards/rejected": -51.25,
+      "step": 340
+    },
+    {
+      "epoch": 1.3533834586466165,
+      "grad_norm": 1.6244164681709312e-07,
+      "learning_rate": 3.3082706766917295e-06,
+      "logits/chosen": -3.059375047683716,
+      "logits/rejected": -4.295312404632568,
+      "logps/chosen": -738.2000122070312,
+      "logps/rejected": -958.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.278124809265137,
+      "rewards/margins": 58.337501525878906,
+      "rewards/rejected": -51.04999923706055,
+      "step": 360
+    },
+    {
+      "epoch": 1.4285714285714286,
+      "grad_norm": 3.8272192733826815e-13,
+      "learning_rate": 3.2142857142857147e-06,
+      "logits/chosen": -3.0648436546325684,
+      "logits/rejected": -4.318749904632568,
+      "logps/chosen": -731.0,
+      "logps/rejected": -984.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.357812404632568,
+      "rewards/margins": 60.38750076293945,
+      "rewards/rejected": -53.04999923706055,
+      "step": 380
+    },
+    {
+      "epoch": 1.5037593984962405,
+      "grad_norm": 6.450130800887369e-09,
+      "learning_rate": 3.1203007518796995e-06,
+      "logits/chosen": -3.0570311546325684,
+      "logits/rejected": -4.317187309265137,
+      "logps/chosen": -735.5999755859375,
+      "logps/rejected": -965.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.471875190734863,
+      "rewards/margins": 59.13750076293945,
+      "rewards/rejected": -51.650001525878906,
+      "step": 400
+    },
+    {
+      "epoch": 1.5037593984962405,
+      "eval_logits/chosen": -3.0611701011657715,
+      "eval_logits/rejected": -4.303191661834717,
+      "eval_logps/chosen": -741.872314453125,
+      "eval_logps/rejected": -972.85107421875,
+      "eval_loss": 1.5668466524232372e-09,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/chosen": 7.266622543334961,
+      "eval_rewards/margins": 58.98404312133789,
+      "eval_rewards/rejected": -51.70744705200195,
+      "eval_runtime": 8.4851,
+      "eval_samples_per_second": 176.78,
+      "eval_score": -0.6956531405448914,
+      "eval_steps_per_second": 5.539,
+      "step": 400
+    },
+    {
+      "epoch": 1.5789473684210527,
+      "grad_norm": 1.5739247094468196e-10,
+      "learning_rate": 3.0263157894736843e-06,
+      "logits/chosen": -3.063281297683716,
+      "logits/rejected": -4.301562309265137,
+      "logps/chosen": -732.5999755859375,
+      "logps/rejected": -961.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.067187309265137,
+      "rewards/margins": 58.025001525878906,
+      "rewards/rejected": -50.95000076293945,
+      "step": 420
+    },
+    {
+      "epoch": 1.6541353383458648,
+      "grad_norm": 5.037244548596992e-09,
+      "learning_rate": 2.9323308270676694e-06,
+      "logits/chosen": -3.0679688453674316,
+      "logits/rejected": -4.328125,
+      "logps/chosen": -735.5999755859375,
+      "logps/rejected": -983.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.542187690734863,
+      "rewards/margins": 60.11249923706055,
+      "rewards/rejected": -52.54999923706055,
+      "step": 440
+    },
+    {
+      "epoch": 1.7293233082706767,
+      "grad_norm": 1.907719763871417e-13,
+      "learning_rate": 2.8383458646616546e-06,
+      "logits/chosen": -3.065624952316284,
+      "logits/rejected": -4.318749904632568,
+      "logps/chosen": -732.0,
+      "logps/rejected": -971.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.348437309265137,
+      "rewards/margins": 59.599998474121094,
+      "rewards/rejected": -52.275001525878906,
+      "step": 460
+    },
+    {
+      "epoch": 1.8045112781954886,
+      "grad_norm": 2.413237608760496e-12,
+      "learning_rate": 2.7443609022556394e-06,
+      "logits/chosen": -3.06640625,
+      "logits/rejected": -4.303124904632568,
+      "logps/chosen": -734.0,
+      "logps/rejected": -962.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.337500095367432,
+      "rewards/margins": 58.900001525878906,
+      "rewards/rejected": -51.54999923706055,
+      "step": 480
+    },
+    {
+      "epoch": 1.8796992481203008,
+      "grad_norm": 3.8499016698807266e-05,
+      "learning_rate": 2.650375939849624e-06,
+      "logits/chosen": -3.063281297683716,
+      "logits/rejected": -4.34375,
+      "logps/chosen": -741.0,
+      "logps/rejected": -975.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.368750095367432,
+      "rewards/margins": 60.42499923706055,
+      "rewards/rejected": -53.0625,
+      "step": 500
+    },
+    {
+      "epoch": 1.954887218045113,
+      "grad_norm": 1.6477178899318827e-11,
+      "learning_rate": 2.556390977443609e-06,
+      "logits/chosen": -3.051562547683716,
+      "logits/rejected": -4.314062595367432,
+      "logps/chosen": -734.2000122070312,
+      "logps/rejected": -970.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.464062690734863,
+      "rewards/margins": 59.70000076293945,
+      "rewards/rejected": -52.224998474121094,
+      "step": 520
+    },
+    {
+      "epoch": 2.030075187969925,
+      "grad_norm": 2.0829162250893226e-08,
+      "learning_rate": 2.462406015037594e-06,
+      "logits/chosen": -3.022656202316284,
+      "logits/rejected": -4.27734375,
+      "logps/chosen": -713.4000244140625,
+      "logps/rejected": -979.4000244140625,
+      "loss": 0.0045,
+      "rewards/accuracies": 0.995312511920929,
+      "rewards/chosen": 8.104687690734863,
+      "rewards/margins": 60.17499923706055,
+      "rewards/rejected": -52.0625,
+      "step": 540
+    },
+    {
+      "epoch": 2.1052631578947367,
+      "grad_norm": 1.0373247696775237e-09,
+      "learning_rate": 2.368421052631579e-06,
+      "logits/chosen": -2.9625000953674316,
+      "logits/rejected": -4.248437404632568,
+      "logps/chosen": -729.2000122070312,
+      "logps/rejected": -967.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.295312881469727,
+      "rewards/margins": 59.724998474121094,
+      "rewards/rejected": -51.42499923706055,
+      "step": 560
+    },
+    {
+      "epoch": 2.180451127819549,
+      "grad_norm": 1.0771024319146517e-10,
+      "learning_rate": 2.274436090225564e-06,
+      "logits/chosen": -2.964062452316284,
+      "logits/rejected": -4.268750190734863,
+      "logps/chosen": -727.2000122070312,
+      "logps/rejected": -964.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.871874809265137,
+      "rewards/margins": 59.3125,
+      "rewards/rejected": -51.42499923706055,
+      "step": 580
+    },
+    {
+      "epoch": 2.255639097744361,
+      "grad_norm": 3.252682975051824e-07,
+      "learning_rate": 2.180451127819549e-06,
+      "logits/chosen": -2.9429688453674316,
+      "logits/rejected": -4.2734375,
+      "logps/chosen": -725.0,
+      "logps/rejected": -963.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.995312690734863,
+      "rewards/margins": 59.537498474121094,
+      "rewards/rejected": -51.537498474121094,
+      "step": 600
+    },
+    {
+      "epoch": 2.255639097744361,
+      "eval_logits/chosen": -2.953125,
+      "eval_logits/rejected": -4.254654407501221,
+      "eval_logps/chosen": -734.9786987304688,
+      "eval_logps/rejected": -968.7659301757812,
+      "eval_loss": 3.946915239083637e-09,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/chosen": 7.918882846832275,
+      "eval_rewards/margins": 59.25,
+      "eval_rewards/rejected": -51.32978820800781,
+      "eval_runtime": 8.4849,
+      "eval_samples_per_second": 176.784,
+      "eval_score": -0.8367462158203125,
+      "eval_steps_per_second": 5.539,
+      "step": 600
+    },
+    {
+      "epoch": 2.3308270676691727,
+      "grad_norm": 1.0373911797620403e-10,
+      "learning_rate": 2.086466165413534e-06,
+      "logits/chosen": -2.9546875953674316,
+      "logits/rejected": -4.296875,
+      "logps/chosen": -730.0,
+      "logps/rejected": -982.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 7.956250190734863,
+      "rewards/margins": 60.900001525878906,
+      "rewards/rejected": -52.92499923706055,
+      "step": 620
+    },
+    {
+      "epoch": 2.406015037593985,
+      "grad_norm": 8.524041237745346e-09,
+      "learning_rate": 1.9924812030075188e-06,
+      "logits/chosen": -2.9507813453674316,
+      "logits/rejected": -4.275000095367432,
+      "logps/chosen": -725.5999755859375,
+      "logps/rejected": -961.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.017187118530273,
+      "rewards/margins": 59.662498474121094,
+      "rewards/rejected": -51.63750076293945,
+      "step": 640
+    },
+    {
+      "epoch": 2.481203007518797,
+      "grad_norm": 1.7859258054396646e-12,
+      "learning_rate": 1.898496240601504e-06,
+      "logits/chosen": -2.953906297683716,
+      "logits/rejected": -4.225781440734863,
+      "logps/chosen": -725.0,
+      "logps/rejected": -964.7999877929688,
+      "loss": 0.0037,
+      "rewards/accuracies": 0.996874988079071,
+      "rewards/chosen": 7.964062690734863,
+      "rewards/margins": 58.587501525878906,
+      "rewards/rejected": -50.599998474121094,
+      "step": 660
+    },
+    {
+      "epoch": 2.556390977443609,
+      "grad_norm": 1.0660900262877694e-07,
+      "learning_rate": 1.8045112781954887e-06,
+      "logits/chosen": -2.977343797683716,
+      "logits/rejected": -4.276562690734863,
+      "logps/chosen": -729.4000244140625,
+      "logps/rejected": -979.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.3203125,
+      "rewards/margins": 60.9375,
+      "rewards/rejected": -52.61249923706055,
+      "step": 680
+    },
+    {
+      "epoch": 2.6315789473684212,
+      "grad_norm": 1.4321429190376608e-11,
+      "learning_rate": 1.710526315789474e-06,
+      "logits/chosen": -2.97265625,
+      "logits/rejected": -4.278124809265137,
+      "logps/chosen": -723.4000244140625,
+      "logps/rejected": -970.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.307812690734863,
+      "rewards/margins": 60.48749923706055,
+      "rewards/rejected": -52.162498474121094,
+      "step": 700
+    },
+    {
+      "epoch": 2.706766917293233,
+      "grad_norm": 1.3597127845174747e-07,
+      "learning_rate": 1.6165413533834587e-06,
+      "logits/chosen": -2.9710936546325684,
+      "logits/rejected": -4.296875,
+      "logps/chosen": -730.7999877929688,
+      "logps/rejected": -986.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.112500190734863,
+      "rewards/margins": 61.724998474121094,
+      "rewards/rejected": -53.599998474121094,
+      "step": 720
+    },
+    {
+      "epoch": 2.781954887218045,
+      "grad_norm": 2.181569008354852e-13,
+      "learning_rate": 1.5225563909774439e-06,
+      "logits/chosen": -2.9749999046325684,
+      "logits/rejected": -4.271874904632568,
+      "logps/chosen": -721.0,
+      "logps/rejected": -971.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.370312690734863,
+      "rewards/margins": 60.67499923706055,
+      "rewards/rejected": -52.32500076293945,
+      "step": 740
+    },
+    {
+      "epoch": 2.857142857142857,
+      "grad_norm": 1.4223787675268966e-11,
+      "learning_rate": 1.4285714285714286e-06,
+      "logits/chosen": -2.979687452316284,
+      "logits/rejected": -4.292187690734863,
+      "logps/chosen": -735.5999755859375,
+      "logps/rejected": -979.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.193750381469727,
+      "rewards/margins": 61.3125,
+      "rewards/rejected": -53.125,
+      "step": 760
+    },
+    {
+      "epoch": 2.932330827067669,
+      "grad_norm": 6.992803813674245e-08,
+      "learning_rate": 1.3345864661654136e-06,
+      "logits/chosen": -2.973437547683716,
+      "logits/rejected": -4.25,
+      "logps/chosen": -724.0,
+      "logps/rejected": -967.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.217187881469727,
+      "rewards/margins": 59.900001525878906,
+      "rewards/rejected": -51.6875,
+      "step": 780
+    },
+    {
+      "epoch": 3.007518796992481,
+      "grad_norm": 1.660423123855119e-08,
+      "learning_rate": 1.2406015037593986e-06,
+      "logits/chosen": -2.975781202316284,
+      "logits/rejected": -4.293749809265137,
+      "logps/chosen": -731.7999877929688,
+      "logps/rejected": -990.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.190625190734863,
+      "rewards/margins": 61.974998474121094,
+      "rewards/rejected": -53.76250076293945,
+      "step": 800
+    },
+    {
+      "epoch": 3.007518796992481,
+      "eval_logits/chosen": -2.9724068641662598,
+      "eval_logits/rejected": -4.260638236999512,
+      "eval_logps/chosen": -733.4468383789062,
+      "eval_logps/rejected": -979.574462890625,
+      "eval_loss": 1.755893541677267e-09,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/chosen": 8.110372543334961,
+      "eval_rewards/margins": 60.537235260009766,
+      "eval_rewards/rejected": -52.441490173339844,
+      "eval_runtime": 8.4913,
+      "eval_samples_per_second": 176.652,
+      "eval_score": -0.7882758378982544,
+      "eval_steps_per_second": 5.535,
+      "step": 800
+    },
+    {
+      "epoch": 3.082706766917293,
+      "grad_norm": 8.005659023834536e-12,
+      "learning_rate": 1.1466165413533836e-06,
+      "logits/chosen": -2.9710936546325684,
+      "logits/rejected": -4.2890625,
+      "logps/chosen": -731.2000122070312,
+      "logps/rejected": -993.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.2578125,
+      "rewards/margins": 62.13750076293945,
+      "rewards/rejected": -53.900001525878906,
+      "step": 820
+    },
+    {
+      "epoch": 3.1578947368421053,
+      "grad_norm": 5.843963868032614e-09,
+      "learning_rate": 1.0526315789473685e-06,
+      "logits/chosen": -2.975781202316284,
+      "logits/rejected": -4.279687404632568,
+      "logps/chosen": -723.7999877929688,
+      "logps/rejected": -973.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.0703125,
+      "rewards/margins": 60.662498474121094,
+      "rewards/rejected": -52.57500076293945,
+      "step": 840
+    },
+    {
+      "epoch": 3.2330827067669174,
+      "grad_norm": 1.4081920596327731e-08,
+      "learning_rate": 9.586466165413535e-07,
+      "logits/chosen": -2.969531297683716,
+      "logits/rejected": -4.287499904632568,
+      "logps/chosen": -732.4000244140625,
+      "logps/rejected": -984.7999877929688,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.153124809265137,
+      "rewards/margins": 61.599998474121094,
+      "rewards/rejected": -53.45000076293945,
+      "step": 860
+    },
+    {
+      "epoch": 3.308270676691729,
+      "grad_norm": 6.352507576934744e-09,
+      "learning_rate": 8.646616541353384e-07,
+      "logits/chosen": -2.9742188453674316,
+      "logits/rejected": -4.279687404632568,
+      "logps/chosen": -727.7999877929688,
+      "logps/rejected": -974.5999755859375,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.331250190734863,
+      "rewards/margins": 60.92499923706055,
+      "rewards/rejected": -52.650001525878906,
+      "step": 880
+    },
+    {
+      "epoch": 3.3834586466165413,
+      "grad_norm": 1.3822216777037244e-07,
+      "learning_rate": 7.706766917293233e-07,
+      "logits/chosen": -2.9828124046325684,
+      "logits/rejected": -4.2734375,
+      "logps/chosen": -725.4000244140625,
+      "logps/rejected": -971.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.231249809265137,
+      "rewards/margins": 60.712501525878906,
+      "rewards/rejected": -52.5,
+      "step": 900
+    },
+    {
+      "epoch": 3.4586466165413534,
+      "grad_norm": 9.571201627686208e-09,
+      "learning_rate": 6.766917293233083e-07,
+      "logits/chosen": -2.9742188453674316,
+      "logits/rejected": -4.264062404632568,
+      "logps/chosen": -731.2000122070312,
+      "logps/rejected": -973.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.154687881469727,
+      "rewards/margins": 60.537498474121094,
+      "rewards/rejected": -52.38750076293945,
+      "step": 920
+    },
+    {
+      "epoch": 3.5338345864661656,
+      "grad_norm": 1.749306374149707e-12,
+      "learning_rate": 5.827067669172933e-07,
+      "logits/chosen": -2.975781202316284,
+      "logits/rejected": -4.282812595367432,
+      "logps/chosen": -724.4000244140625,
+      "logps/rejected": -971.7999877929688,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.28125,
+      "rewards/margins": 60.92499923706055,
+      "rewards/rejected": -52.63750076293945,
+      "step": 940
+    },
+    {
+      "epoch": 3.6090225563909772,
+      "grad_norm": 1.2949258395470178e-10,
+      "learning_rate": 4.887218045112782e-07,
+      "logits/chosen": -2.973437547683716,
+      "logits/rejected": -4.268750190734863,
+      "logps/chosen": -720.7999877929688,
+      "logps/rejected": -976.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.112500190734863,
+      "rewards/margins": 60.787498474121094,
+      "rewards/rejected": -52.67499923706055,
+      "step": 960
+    },
+    {
+      "epoch": 3.6842105263157894,
+      "grad_norm": 7.884964535382202e-09,
+      "learning_rate": 3.9473684210526315e-07,
+      "logits/chosen": -2.9671874046325684,
+      "logits/rejected": -4.268750190734863,
+      "logps/chosen": -730.4000244140625,
+      "logps/rejected": -975.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.253125190734863,
+      "rewards/margins": 60.787498474121094,
+      "rewards/rejected": -52.54999923706055,
+      "step": 980
+    },
+    {
+      "epoch": 3.7593984962406015,
+      "grad_norm": 1.6358064493089766e-10,
+      "learning_rate": 3.007518796992481e-07,
+      "logits/chosen": -2.973437547683716,
+      "logits/rejected": -4.2578125,
+      "logps/chosen": -730.0,
+      "logps/rejected": -983.2000122070312,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.1875,
+      "rewards/margins": 60.962501525878906,
+      "rewards/rejected": -52.787498474121094,
+      "step": 1000
+    },
+    {
+      "epoch": 3.7593984962406015,
+      "eval_logits/chosen": -2.9727394580841064,
+      "eval_logits/rejected": -4.261303424835205,
+      "eval_logps/chosen": -733.3616943359375,
+      "eval_logps/rejected": -980.0,
+      "eval_loss": 1.7744486990878272e-09,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/chosen": 8.113032341003418,
+      "eval_rewards/margins": 60.5638313293457,
+      "eval_rewards/rejected": -52.462764739990234,
+      "eval_runtime": 8.5637,
+      "eval_samples_per_second": 175.158,
+      "eval_score": -0.7953461408615112,
+      "eval_steps_per_second": 5.488,
+      "step": 1000
+    },
+    {
+      "epoch": 3.8345864661654137,
+      "grad_norm": 7.702424009927947e-10,
+      "learning_rate": 2.067669172932331e-07,
+      "logits/chosen": -2.971874952316284,
+      "logits/rejected": -4.2578125,
+      "logps/chosen": -729.5999755859375,
+      "logps/rejected": -980.0,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.28125,
+      "rewards/margins": 60.537498474121094,
+      "rewards/rejected": -52.20000076293945,
+      "step": 1020
+    },
+    {
+      "epoch": 3.909774436090226,
+      "grad_norm": 5.643256636003421e-08,
+      "learning_rate": 1.1278195488721805e-07,
+      "logits/chosen": -2.9820313453674316,
+      "logits/rejected": -4.292187690734863,
+      "logps/chosen": -729.7999877929688,
+      "logps/rejected": -988.4000244140625,
+      "loss": 0.0,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 8.293749809265137,
+      "rewards/margins": 61.79999923706055,
+      "rewards/rejected": -53.525001525878906,
+      "step": 1040
+    },
+    {
+      "epoch": 3.9849624060150375,
+      "grad_norm": 6.672698844431039e-09,
+      "learning_rate": 1.8796992481203008e-08,
+      "logits/chosen": -2.9671874046325684,
+      "logits/rejected": -4.241406440734863,
+      "logps/chosen": -710.4000244140625,
+      "logps/rejected": -972.4000244140625,
+      "loss": 0.0049,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": 8.0546875,
+      "rewards/margins": 59.900001525878906,
+      "rewards/rejected": -51.837501525878906,
+      "step": 1060
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 1064,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 300,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

dataset-0/checkpoint-1064/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53083fe0b244af4e0f853144e008d73e36663ab1424b0e34148d1834df62622b
+size 7992

dataset-0/checkpoint-1064/zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+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, weights_only=False)
+
+        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 tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # 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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+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 = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        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}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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 to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

dataset-0/checkpoint-300/added_tokens.json

@@ -0,0 +1,5 @@
+{
+  "<|endoftext|>": 151643,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644
+}

dataset-0/checkpoint-300/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "Qwen/Qwen2-0.5B-Instruct",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.49.0",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

dataset-0/checkpoint-300/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.1,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.49.0"
+}

dataset-0/checkpoint-300/global_step300/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:215f5aac3c4e731f92aa12b04bff8afafed83f8ab6c33772f26f719cd5ddf46e
+size 1482103280

dataset-0/checkpoint-300/global_step300/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdfd3549f8e5284cd53888cf8277ce0168895e97add700aa89e9be8c696252bf
+size 1482103280

dataset-0/checkpoint-300/global_step300/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c8cd97699549e19af2a3caee10ceb7185de3aca2b0e28110ee3df2bfbbd2260
+size 1482103280

dataset-0/checkpoint-300/global_step300/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70b55a1986d6646a9b6893fa56167c9d916e6366ab601d3574b638cb8d4f6ba6
+size 1482103280

dataset-0/checkpoint-300/global_step300/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2905f3d8ef914e8f3e13498720c98815c28cc1405201df8666e3553413880e13
+size 142728

dataset-0/checkpoint-300/global_step300/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5c06da8d026e399810f9277cfcefcac4c4683e884491ed22f1abf2a813a587a
+size 142664

dataset-0/checkpoint-300/global_step300/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac32b8a1f16fe6c858519ab8dad2047cf25a1c7f7856186607712140845a5082
+size 142664

dataset-0/checkpoint-300/global_step300/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5f8fb4df2e19b18ec38d918ad176a802b4d6ffa28b4dc07fd12f7ac442fc6ec
+size 142664

dataset-0/checkpoint-300/latest

@@ -0,0 +1 @@
+global_step300

dataset-0/checkpoint-300/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:795daca6b0e7f3796545eb712b58861323b4b8a0096962fc1682bd6d4d616899
+size 988097824

dataset-0/checkpoint-300/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e60baca04b746326a6b7ae6b3aa189980b7cc51bdd14b3ed53746093b56189df
+size 15024

dataset-0/checkpoint-300/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e234cb41cc8c7b367d4683147cea9d4111b7c0fe205f6a4a39a690414894dec3
+size 15024

dataset-0/checkpoint-300/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91b117f6cd91485bd56ca6b914c1e5a29b7b4cb2e7f29545be13eb60b37277c1
+size 15024

dataset-0/checkpoint-300/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3990446e832529a8aa9f5a964d189b3a2eea28ebfcfefcd038028ca040f4236
+size 15024

dataset-0/checkpoint-300/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:148067466c5e2463648afe08dbbbfd0cb93920366b97202b7c6c137055747dbb
+size 1064

dataset-0/checkpoint-300/special_tokens_map.json

@@ -0,0 +1,20 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

dataset-0/checkpoint-300/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcfe42da0a4497e8b2b172c1f9f4ec423a46dc12907f4349c55025f670422ba9
+size 11418266

dataset-0/checkpoint-300/tokenizer_config.json

@@ -0,0 +1,44 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}