Upload folder using huggingface_hub

README.md

@@ -0,0 +1,10 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+- PEFT 0.4.0
+
+- PEFT 0.4.0

adapter_config.json

@@ -0,0 +1,26 @@
+{
+  "auto_mapping": null,
+  "base_model_name_or_path": "/model_zoo/Vivid-7B-base",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "lora_alpha": 64,
+  "lora_dropout": 0.05,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 32,
+  "revision": null,
+  "target_modules": [
+    "o_proj",
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "down_proj",
+    "v_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM"
+}

adapter_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b85634ff64cb7542cf4817964b530b79171fbdd23435bc9352367fa68a0e42a
+size 167927754

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff14197a580bc8d117bf3c16e2eaab8aaded005ee90618bca3a0a563439295af
+size 167832688

config.json

@@ -0,0 +1,45 @@
+{
+  "_name_or_path": "/model_zoo/Vivid-7B-base",
+  "architectures": [
+    "LlavaMistralForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bert_type": "raw_bert_layer:12",
+  "bos_token_id": 1,
+  "compress_type": "mean",
+  "eos_token_id": 2,
+  "freeze_mm_mlp_adapter": false,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "image_aspect_ratio": "pad",
+  "image_grid_pinpoints": null,
+  "image_processor": "./llamavid/processor/clip-patch14-224",
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 32768,
+  "max_token": 4096,
+  "mm_hidden_size": 1024,
+  "mm_projector_type": "mlp2x_gelu",
+  "mm_use_im_patch_token": false,
+  "mm_use_im_start_end": false,
+  "mm_vision_select_feature": "patch",
+  "mm_vision_select_layer": -2,
+  "mm_vision_tower": "/model_zoo/openai-clip-vit-large-patch14",
+  "model_path": "/content/Vivid-7B-Instruct-Lora",
+  "model_type": "mistral",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "num_query": 32,
+  "pretrain_mm_mlp_adapter": "/content/Vivid-7B-PT/mm_projector.bin",
+  "rms_norm_eps": 1e-05,
+  "rope_theta": 10000.0,
+  "sliding_window": 4096,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.38.2",
+  "tune_mm_mlp_adapter": false,
+  "use_cache": false,
+  "use_mm_proj": true,
+  "vocab_size": 48384
+}

global_step150/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57262f60d6570023af26a520a89ac40f01d700d442a4f11a3d8fd45b0e13e8df
+size 6931872192

global_step150/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee8f5beb7c20bf85bed9e3f057f9524c5964e5c1773ffa4ff7387c20c1ca0875
+size 1155894712

latest

@@ -0,0 +1 @@
+global_step150

non_lora_trainables.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f70073b0da2fa52b0b7936fed4340dee3733ab45b897d7919c3be505bfefdd8a
+size 987789762

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f30c8c68b943b04231e947d488a0c4b71b56413e2f7f06b32c4e8377ce1df6a
+size 14244

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce76126167d620de50ef95ce12b983f6efa02e9a5112000d0901e807c78c2484
+size 1064

special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d88bdadaa2a065aa7c6e18a4b5999ce4c76cec14d9fea882102e7b4931d7ef0
+size 779539

tokenizer_config.json

@@ -0,0 +1,45 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '</s>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<unk>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

trainer_state.json

@@ -0,0 +1,1071 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.33531260915123995,
+  "eval_steps": 500,
+  "global_step": 150,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.7522330163029802,
+      "learning_rate": 7.142857142857143e-06,
+      "loss": 1.3831,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.7473994784323043,
+      "learning_rate": 1.4285714285714285e-05,
+      "loss": 1.3916,
+      "step": 2
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.70544460641196,
+      "learning_rate": 2.1428571428571428e-05,
+      "loss": 1.4248,
+      "step": 3
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.5834904606723813,
+      "learning_rate": 2.857142857142857e-05,
+      "loss": 1.4045,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8295213706761098,
+      "learning_rate": 3.571428571428572e-05,
+      "loss": 1.3918,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.8133546220957368,
+      "learning_rate": 4.2857142857142856e-05,
+      "loss": 1.3865,
+      "step": 6
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.7903446421142379,
+      "learning_rate": 5e-05,
+      "loss": 1.373,
+      "step": 7
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6811759259942566,
+      "learning_rate": 5.714285714285714e-05,
+      "loss": 1.3682,
+      "step": 8
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6292547023357415,
+      "learning_rate": 6.428571428571429e-05,
+      "loss": 1.3916,
+      "step": 9
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.5931255904191401,
+      "learning_rate": 7.142857142857143e-05,
+      "loss": 1.3833,
+      "step": 10
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.6447974541785289,
+      "learning_rate": 7.857142857142858e-05,
+      "loss": 1.4023,
+      "step": 11
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.5941331466143945,
+      "learning_rate": 8.571428571428571e-05,
+      "loss": 1.3423,
+      "step": 12
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.6603903337024011,
+      "learning_rate": 9.285714285714286e-05,
+      "loss": 1.3762,
+      "step": 13
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.5633045289970319,
+      "learning_rate": 0.0001,
+      "loss": 1.3843,
+      "step": 14
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.5277710229216985,
+      "learning_rate": 9.999868398131282e-05,
+      "loss": 1.3872,
+      "step": 15
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5589957310018021,
+      "learning_rate": 9.999473599452745e-05,
+      "loss": 1.3452,
+      "step": 16
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5000831179822082,
+      "learning_rate": 9.99881562474689e-05,
+      "loss": 1.3813,
+      "step": 17
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.5290519156012868,
+      "learning_rate": 9.997894508649996e-05,
+      "loss": 1.3762,
+      "step": 18
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.519496248745277,
+      "learning_rate": 9.996710299650301e-05,
+      "loss": 1.3706,
+      "step": 19
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.47116309957086233,
+      "learning_rate": 9.995263060085454e-05,
+      "loss": 1.344,
+      "step": 20
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.4762611988791945,
+      "learning_rate": 9.993552866139229e-05,
+      "loss": 1.3513,
+      "step": 21
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.4669268035616973,
+      "learning_rate": 9.99157980783751e-05,
+      "loss": 1.3357,
+      "step": 22
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.46010212813136436,
+      "learning_rate": 9.989343989043563e-05,
+      "loss": 1.3391,
+      "step": 23
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.4608157209685965,
+      "learning_rate": 9.98684552745256e-05,
+      "loss": 1.3591,
+      "step": 24
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4483237749946448,
+      "learning_rate": 9.984084554585387e-05,
+      "loss": 1.3433,
+      "step": 25
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4491711395465447,
+      "learning_rate": 9.98106121578172e-05,
+      "loss": 1.3616,
+      "step": 26
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.42807282609895003,
+      "learning_rate": 9.977775670192371e-05,
+      "loss": 1.3342,
+      "step": 27
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4236536126545817,
+      "learning_rate": 9.97422809077092e-05,
+      "loss": 1.3572,
+      "step": 28
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.4237881873453837,
+      "learning_rate": 9.970418664264595e-05,
+      "loss": 1.3394,
+      "step": 29
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.4363640797964494,
+      "learning_rate": 9.96634759120446e-05,
+      "loss": 1.3611,
+      "step": 30
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.3953629543521621,
+      "learning_rate": 9.962015085894838e-05,
+      "loss": 1.3208,
+      "step": 31
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.4092487417981463,
+      "learning_rate": 9.957421376402051e-05,
+      "loss": 1.3457,
+      "step": 32
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.4026279934538214,
+      "learning_rate": 9.9525667045424e-05,
+      "loss": 1.344,
+      "step": 33
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.39575866758551026,
+      "learning_rate": 9.947451325869439e-05,
+      "loss": 1.3416,
+      "step": 34
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.40044778357348837,
+      "learning_rate": 9.942075509660527e-05,
+      "loss": 1.3181,
+      "step": 35
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.40607328352911176,
+      "learning_rate": 9.936439538902644e-05,
+      "loss": 1.321,
+      "step": 36
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.4080804869363711,
+      "learning_rate": 9.930543710277509e-05,
+      "loss": 1.3423,
+      "step": 37
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.39208662642207626,
+      "learning_rate": 9.924388334145943e-05,
+      "loss": 1.3333,
+      "step": 38
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.4056602363153877,
+      "learning_rate": 9.91797373453155e-05,
+      "loss": 1.3521,
+      "step": 39
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.40133586130843213,
+      "learning_rate": 9.911300249103645e-05,
+      "loss": 1.3284,
+      "step": 40
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.3839921181227527,
+      "learning_rate": 9.904368229159493e-05,
+      "loss": 1.334,
+      "step": 41
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.37819843842451634,
+      "learning_rate": 9.897178039605802e-05,
+      "loss": 1.3201,
+      "step": 42
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.38535218746253924,
+      "learning_rate": 9.889730058939528e-05,
+      "loss": 1.3418,
+      "step": 43
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.3816398827811328,
+      "learning_rate": 9.88202467922794e-05,
+      "loss": 1.292,
+      "step": 44
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.3872171757684603,
+      "learning_rate": 9.874062306087982e-05,
+      "loss": 1.3086,
+      "step": 45
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.40255942917819887,
+      "learning_rate": 9.865843358664932e-05,
+      "loss": 1.356,
+      "step": 46
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.38408063484536253,
+      "learning_rate": 9.857368269610324e-05,
+      "loss": 1.3123,
+      "step": 47
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.3858660663689368,
+      "learning_rate": 9.848637485059183e-05,
+      "loss": 1.3228,
+      "step": 48
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.3733195606949192,
+      "learning_rate": 9.839651464606529e-05,
+      "loss": 1.3191,
+      "step": 49
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.3843895669282248,
+      "learning_rate": 9.830410681283202e-05,
+      "loss": 1.3137,
+      "step": 50
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.3839890456700812,
+      "learning_rate": 9.82091562153094e-05,
+      "loss": 1.3037,
+      "step": 51
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3835789656095749,
+      "learning_rate": 9.811166785176784e-05,
+      "loss": 1.3281,
+      "step": 52
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3742526616293116,
+      "learning_rate": 9.801164685406771e-05,
+      "loss": 1.3083,
+      "step": 53
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.3896917243786738,
+      "learning_rate": 9.790909848738906e-05,
+      "loss": 1.3142,
+      "step": 54
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.37476225815080133,
+      "learning_rate": 9.780402814995459e-05,
+      "loss": 1.2991,
+      "step": 55
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.3715519529394886,
+      "learning_rate": 9.769644137274538e-05,
+      "loss": 1.2986,
+      "step": 56
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.3766009323599998,
+      "learning_rate": 9.758634381920981e-05,
+      "loss": 1.3054,
+      "step": 57
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.38112908922611866,
+      "learning_rate": 9.74737412849654e-05,
+      "loss": 1.2891,
+      "step": 58
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.37814911656520694,
+      "learning_rate": 9.735863969749372e-05,
+      "loss": 1.3025,
+      "step": 59
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.3825181116798945,
+      "learning_rate": 9.724104511582838e-05,
+      "loss": 1.3152,
+      "step": 60
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.37912087213705015,
+      "learning_rate": 9.712096373023604e-05,
+      "loss": 1.2844,
+      "step": 61
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3776600267484215,
+      "learning_rate": 9.69984018618906e-05,
+      "loss": 1.3042,
+      "step": 62
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3928544712245226,
+      "learning_rate": 9.687336596254044e-05,
+      "loss": 1.3259,
+      "step": 63
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.3652188942778873,
+      "learning_rate": 9.674586261416873e-05,
+      "loss": 1.2854,
+      "step": 64
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.3846903345562418,
+      "learning_rate": 9.66158985286471e-05,
+      "loss": 1.3091,
+      "step": 65
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.3872756941099026,
+      "learning_rate": 9.648348054738208e-05,
+      "loss": 1.2981,
+      "step": 66
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.38743186857828676,
+      "learning_rate": 9.634861564095524e-05,
+      "loss": 1.303,
+      "step": 67
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.39711005882387096,
+      "learning_rate": 9.621131090875603e-05,
+      "loss": 1.313,
+      "step": 68
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.3775435690594975,
+      "learning_rate": 9.607157357860823e-05,
+      "loss": 1.2766,
+      "step": 69
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.3875463212418074,
+      "learning_rate": 9.59294110063893e-05,
+      "loss": 1.3157,
+      "step": 70
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.3858636973285557,
+      "learning_rate": 9.578483067564335e-05,
+      "loss": 1.2834,
+      "step": 71
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.37541703932578996,
+      "learning_rate": 9.563784019718703e-05,
+      "loss": 1.2957,
+      "step": 72
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.38113820507589535,
+      "learning_rate": 9.548844730870901e-05,
+      "loss": 1.2932,
+      "step": 73
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.38509417483302427,
+      "learning_rate": 9.533665987436263e-05,
+      "loss": 1.311,
+      "step": 74
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.36678005983919304,
+      "learning_rate": 9.518248588435185e-05,
+      "loss": 1.2852,
+      "step": 75
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.3875959550686585,
+      "learning_rate": 9.502593345451078e-05,
+      "loss": 1.2961,
+      "step": 76
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.3690516151813626,
+      "learning_rate": 9.486701082587634e-05,
+      "loss": 1.2837,
+      "step": 77
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.37636909155427106,
+      "learning_rate": 9.470572636425449e-05,
+      "loss": 1.3242,
+      "step": 78
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3723573615312267,
+      "learning_rate": 9.454208855977986e-05,
+      "loss": 1.2837,
+      "step": 79
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3849981556406106,
+      "learning_rate": 9.437610602646878e-05,
+      "loss": 1.2944,
+      "step": 80
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.3778470741048999,
+      "learning_rate": 9.420778750176588e-05,
+      "loss": 1.2871,
+      "step": 81
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.37645381040425563,
+      "learning_rate": 9.403714184608411e-05,
+      "loss": 1.3088,
+      "step": 82
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.397847096161663,
+      "learning_rate": 9.386417804233835e-05,
+      "loss": 1.3196,
+      "step": 83
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.39589602529348583,
+      "learning_rate": 9.368890519547251e-05,
+      "loss": 1.2812,
+      "step": 84
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.3755562583506324,
+      "learning_rate": 9.351133253198028e-05,
+      "loss": 1.2759,
+      "step": 85
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.3757465002070439,
+      "learning_rate": 9.333146939941938e-05,
+      "loss": 1.2949,
+      "step": 86
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.38318152261822297,
+      "learning_rate": 9.314932526591956e-05,
+      "loss": 1.2795,
+      "step": 87
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.3756550651404947,
+      "learning_rate": 9.296490971968415e-05,
+      "loss": 1.2651,
+      "step": 88
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.38942044400837683,
+      "learning_rate": 9.277823246848536e-05,
+      "loss": 1.2903,
+      "step": 89
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.37348810668822124,
+      "learning_rate": 9.258930333915324e-05,
+      "loss": 1.2612,
+      "step": 90
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.3832983019734946,
+      "learning_rate": 9.239813227705838e-05,
+      "loss": 1.2805,
+      "step": 91
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.37762671923250346,
+      "learning_rate": 9.220472934558837e-05,
+      "loss": 1.3,
+      "step": 92
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3791093875126196,
+      "learning_rate": 9.20091047256181e-05,
+      "loss": 1.2925,
+      "step": 93
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.383463697649199,
+      "learning_rate": 9.181126871497378e-05,
+      "loss": 1.2815,
+      "step": 94
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.379130797624887,
+      "learning_rate": 9.16112317278909e-05,
+      "loss": 1.2805,
+      "step": 95
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.3816039801864187,
+      "learning_rate": 9.140900429446602e-05,
+      "loss": 1.2825,
+      "step": 96
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.37629435581707915,
+      "learning_rate": 9.120459706010233e-05,
+      "loss": 1.2896,
+      "step": 97
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.38890563473703343,
+      "learning_rate": 9.099802078494948e-05,
+      "loss": 1.2734,
+      "step": 98
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.3782158636576143,
+      "learning_rate": 9.078928634333698e-05,
+      "loss": 1.2803,
+      "step": 99
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.38350433679936774,
+      "learning_rate": 9.057840472320192e-05,
+      "loss": 1.2639,
+      "step": 100
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.38940074349705917,
+      "learning_rate": 9.036538702551036e-05,
+      "loss": 1.2732,
+      "step": 101
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.39603306187884135,
+      "learning_rate": 9.015024446367315e-05,
+      "loss": 1.2976,
+      "step": 102
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.39112427722222776,
+      "learning_rate": 8.993298836295556e-05,
+      "loss": 1.2588,
+      "step": 103
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3962255296124105,
+      "learning_rate": 8.971363015988114e-05,
+      "loss": 1.2915,
+      "step": 104
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.3858473392543198,
+      "learning_rate": 8.949218140162966e-05,
+      "loss": 1.2695,
+      "step": 105
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3820634901255279,
+      "learning_rate": 8.926865374542928e-05,
+      "loss": 1.2876,
+      "step": 106
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.40421307595226363,
+      "learning_rate": 8.904305895794292e-05,
+      "loss": 1.2817,
+      "step": 107
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3716456005592202,
+      "learning_rate": 8.881540891464879e-05,
+      "loss": 1.2761,
+      "step": 108
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.3815567099868862,
+      "learning_rate": 8.858571559921538e-05,
+      "loss": 1.2644,
+      "step": 109
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.3848695348441116,
+      "learning_rate": 8.835399110287045e-05,
+      "loss": 1.2622,
+      "step": 110
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.38017284820864067,
+      "learning_rate": 8.812024762376477e-05,
+      "loss": 1.2642,
+      "step": 111
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.3779202181266027,
+      "learning_rate": 8.788449746632977e-05,
+      "loss": 1.2766,
+      "step": 112
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.39015638614586473,
+      "learning_rate": 8.764675304062993e-05,
+      "loss": 1.2881,
+      "step": 113
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.39447312823941305,
+      "learning_rate": 8.740702686170955e-05,
+      "loss": 1.2776,
+      "step": 114
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.38606430769475053,
+      "learning_rate": 8.716533154893389e-05,
+      "loss": 1.283,
+      "step": 115
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.37710007527264683,
+      "learning_rate": 8.692167982532486e-05,
+      "loss": 1.2754,
+      "step": 116
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.3953761864482322,
+      "learning_rate": 8.667608451689134e-05,
+      "loss": 1.2891,
+      "step": 117
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.38041030780699725,
+      "learning_rate": 8.642855855195394e-05,
+      "loss": 1.2717,
+      "step": 118
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.37534026330530307,
+      "learning_rate": 8.617911496046446e-05,
+      "loss": 1.2861,
+      "step": 119
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.3642851567224028,
+      "learning_rate": 8.592776687332002e-05,
+      "loss": 1.2546,
+      "step": 120
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.37390620756705745,
+      "learning_rate": 8.567452752167182e-05,
+      "loss": 1.2583,
+      "step": 121
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.37970332189538375,
+      "learning_rate": 8.54194102362286e-05,
+      "loss": 1.2827,
+      "step": 122
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.3873078508435452,
+      "learning_rate": 8.516242844655499e-05,
+      "loss": 1.2708,
+      "step": 123
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.38706228040025265,
+      "learning_rate": 8.490359568036446e-05,
+      "loss": 1.269,
+      "step": 124
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3745301218341843,
+      "learning_rate": 8.464292556280733e-05,
+      "loss": 1.2434,
+      "step": 125
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.38170858987166395,
+      "learning_rate": 8.43804318157534e-05,
+      "loss": 1.2456,
+      "step": 126
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.3735646632404992,
+      "learning_rate": 8.411612825706977e-05,
+      "loss": 1.2573,
+      "step": 127
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.4099110289018266,
+      "learning_rate": 8.385002879989328e-05,
+      "loss": 1.2915,
+      "step": 128
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3877538310898807,
+      "learning_rate": 8.35821474518983e-05,
+      "loss": 1.2793,
+      "step": 129
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.3902454161481885,
+      "learning_rate": 8.33124983145592e-05,
+      "loss": 1.2683,
+      "step": 130
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.38878925776724954,
+      "learning_rate": 8.304109558240815e-05,
+      "loss": 1.2417,
+      "step": 131
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.39723877066251073,
+      "learning_rate": 8.276795354228786e-05,
+      "loss": 1.2751,
+      "step": 132
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.38395165663929565,
+      "learning_rate": 8.249308657259943e-05,
+      "loss": 1.2576,
+      "step": 133
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.39643029338267877,
+      "learning_rate": 8.221650914254566e-05,
+      "loss": 1.2742,
+      "step": 134
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.39444322211869104,
+      "learning_rate": 8.193823581136918e-05,
+      "loss": 1.2671,
+      "step": 135
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.3941951730886792,
+      "learning_rate": 8.165828122758615e-05,
+      "loss": 1.2612,
+      "step": 136
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.3877727475185371,
+      "learning_rate": 8.137666012821514e-05,
+      "loss": 1.2441,
+      "step": 137
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.40860172723134963,
+      "learning_rate": 8.109338733800132e-05,
+      "loss": 1.2839,
+      "step": 138
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.39465054288851986,
+      "learning_rate": 8.08084777686361e-05,
+      "loss": 1.2595,
+      "step": 139
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.38699274061736677,
+      "learning_rate": 8.052194641797217e-05,
+      "loss": 1.2539,
+      "step": 140
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3901027157370492,
+      "learning_rate": 8.023380836923405e-05,
+      "loss": 1.269,
+      "step": 141
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3818216948779056,
+      "learning_rate": 7.994407879022396e-05,
+      "loss": 1.2632,
+      "step": 142
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.364402831939757,
+      "learning_rate": 7.965277293252353e-05,
+      "loss": 1.2329,
+      "step": 143
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.3815713950573408,
+      "learning_rate": 7.935990613069086e-05,
+      "loss": 1.2734,
+      "step": 144
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.370245164697694,
+      "learning_rate": 7.906549380145329e-05,
+      "loss": 1.2412,
+      "step": 145
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.37768543350286454,
+      "learning_rate": 7.876955144289592e-05,
+      "loss": 1.2681,
+      "step": 146
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.37390287320878,
+      "learning_rate": 7.847209463364573e-05,
+      "loss": 1.2668,
+      "step": 147
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.37508076646147226,
+      "learning_rate": 7.817313903205148e-05,
+      "loss": 1.2566,
+      "step": 148
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.3775081062527898,
+      "learning_rate": 7.78727003753595e-05,
+      "loss": 1.2434,
+      "step": 149
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.3901559194494511,
+      "learning_rate": 7.757079447888527e-05,
+      "loss": 1.2351,
+      "step": 150
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 447,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 25,
+  "total_flos": 2.222372777610969e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68c32ad566d970e6f873dc4e4345382ecb94322012eeed8b360cdbecf9bc1dba
+size 6584

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)