Upload with huggingface_hub

config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "EleutherAI/pythia-125m-deduped",
+  "architectures": [
+    "GPTNeoXForCausalLM"
+  ],
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "hidden_act": "gelu",
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 2048,
+  "model_type": "gpt_neox",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "rotary_emb_base": 10000,
+  "rotary_pct": 0.25,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.26.0.dev0",
+  "use_cache": true,
+  "use_parallel_residual": true,
+  "vocab_size": 50304
+}

global_step14299/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65f3e1388c1910f8300c6f1506f030b0990d48973d43a9fa46f64373d023d037
+size 375032614

global_step14299/zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0646ae5d67d19f2dc6f2aa1069bd3d67daed322175d8e73468e7ec16c30831d
+size 243487614

global_step14299/zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6cd397b71061ba43d3ebc80172c71db8ca3d42d8d3b3d30e63eed3fbe9dc2f4
+size 243488190

global_step14299/zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d1f6c0c029b5be6344a050913a0f1901dd77d4b9fa892674ffa665b1c2923f4
+size 243490174

global_step14299/zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f0ce3404a510dd659dc779aae11a57827723615aa791340ee55f6f8ca046bdc
+size 243490046

global_step14299/zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cac902664a41e18c2d1d9375b722fcc91f66606cd3dff694e992806abd0796eb
+size 243490110

global_step14299/zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd2258f43bd6efc51c74e3a25d2ce05a74bb6ff16b7f149c718c3d83cf67edcf
+size 243490366

global_step14299/zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52862fcaa4e3a8924e72ed3f7e558dbe32dc7dc1310bd36c6c0916e6fcc9deef
+size 243487934

global_step14299/zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3358f0c6d163e936078596147ae30e4d1c054b61a709222c10d1ed89664afa66
+size 243487614

latest

@@ -0,0 +1 @@
+global_step14299

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2338087afffad511f3542a2906678c21b1f6db0846eecd1247065b8fd9f57be4
+size 375009696

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a961c7aa8293531040f379f129f1930093670f11c91d3736bf6d8518af2b1507
+size 14583

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbc4475e8615a434844642d2b6db8c82a3eb4d1532c0358142650af35982f7d4
+size 14583

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:881eb4ea22a97cf8b6f92b147ff4c6e2ae57dcc7e36f52b570b5be3eeb7887b6
+size 14583

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b494daea51340989a686bc325649dfd2d62f65af56b0d25143e33790c8a957a
+size 14583

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a88b3a9b9d34f9c56a628c06043f3d84338d75c0f6730bf31d21d05f60a24e7
+size 14583

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1556e939ee7eca3367a22108d1bd8653197711d2ab29b9fc7633bec21b96125
+size 14583

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86e9ed27d35f586c7ec30600373ad7abf89086510211847110cb4a979365cfa0
+size 14583

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1828f5f1b78bdfd59c65a849162815831fbce941a063f6c4056e93648685a2ec
+size 14583

trainer_state.json

@@ -0,0 +1,876 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "global_step": 14299,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.488135456452206e-05,
+      "loss": 3.4925,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0001,
+      "loss": 1.9277,
+      "step": 200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0001,
+      "loss": 1.9662,
+      "step": 300
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001,
+      "loss": 1.9422,
+      "step": 400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001,
+      "loss": 1.9676,
+      "step": 500
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0001,
+      "loss": 1.9137,
+      "step": 600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0001,
+      "loss": 1.9491,
+      "step": 700
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0001,
+      "loss": 1.9413,
+      "step": 800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0001,
+      "loss": 1.9424,
+      "step": 900
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0001,
+      "loss": 1.916,
+      "step": 1000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0001,
+      "loss": 1.8919,
+      "step": 1100
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0001,
+      "loss": 1.9228,
+      "step": 1200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0001,
+      "loss": 1.9008,
+      "step": 1300
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0001,
+      "loss": 1.9189,
+      "step": 1400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0001,
+      "loss": 1.8898,
+      "step": 1500
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0001,
+      "loss": 1.8724,
+      "step": 1600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0001,
+      "loss": 1.9123,
+      "step": 1700
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0001,
+      "loss": 1.934,
+      "step": 1800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0001,
+      "loss": 1.904,
+      "step": 1900
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0001,
+      "loss": 1.9295,
+      "step": 2000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0001,
+      "loss": 1.926,
+      "step": 2100
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0001,
+      "loss": 1.8874,
+      "step": 2200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0001,
+      "loss": 1.9097,
+      "step": 2300
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0001,
+      "loss": 1.9119,
+      "step": 2400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0001,
+      "loss": 1.8833,
+      "step": 2500
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0001,
+      "loss": 1.894,
+      "step": 2600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0001,
+      "loss": 1.9345,
+      "step": 2700
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0001,
+      "loss": 1.8902,
+      "step": 2800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0001,
+      "loss": 1.8798,
+      "step": 2900
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0001,
+      "loss": 1.8891,
+      "step": 3000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0001,
+      "loss": 1.8853,
+      "step": 3100
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0001,
+      "loss": 1.865,
+      "step": 3200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0001,
+      "loss": 1.9037,
+      "step": 3300
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0001,
+      "loss": 1.8672,
+      "step": 3400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0001,
+      "loss": 1.868,
+      "step": 3500
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0001,
+      "loss": 1.8933,
+      "step": 3600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0001,
+      "loss": 1.8709,
+      "step": 3700
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0001,
+      "loss": 1.8901,
+      "step": 3800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0001,
+      "loss": 1.87,
+      "step": 3900
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0001,
+      "loss": 1.9241,
+      "step": 4000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0001,
+      "loss": 1.8981,
+      "step": 4100
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0001,
+      "loss": 1.8564,
+      "step": 4200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0001,
+      "loss": 1.8906,
+      "step": 4300
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0001,
+      "loss": 1.8751,
+      "step": 4400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0001,
+      "loss": 1.8649,
+      "step": 4500
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0001,
+      "loss": 1.8901,
+      "step": 4600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0001,
+      "loss": 1.8843,
+      "step": 4700
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0001,
+      "loss": 1.876,
+      "step": 4800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0001,
+      "loss": 1.8486,
+      "step": 4900
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0001,
+      "loss": 1.8679,
+      "step": 5000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0001,
+      "loss": 1.8295,
+      "step": 5100
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0001,
+      "loss": 1.8675,
+      "step": 5200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0001,
+      "loss": 1.8501,
+      "step": 5300
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0001,
+      "loss": 1.861,
+      "step": 5400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0001,
+      "loss": 1.8825,
+      "step": 5500
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0001,
+      "loss": 1.8431,
+      "step": 5600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0001,
+      "loss": 1.8278,
+      "step": 5700
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0001,
+      "loss": 1.8671,
+      "step": 5800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0001,
+      "loss": 1.8763,
+      "step": 5900
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0001,
+      "loss": 1.8645,
+      "step": 6000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0001,
+      "loss": 1.8367,
+      "step": 6100
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0001,
+      "loss": 1.8838,
+      "step": 6200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0001,
+      "loss": 1.8699,
+      "step": 6300
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0001,
+      "loss": 1.87,
+      "step": 6400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0001,
+      "loss": 1.843,
+      "step": 6500
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0001,
+      "loss": 1.8441,
+      "step": 6600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0001,
+      "loss": 1.8656,
+      "step": 6700
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0001,
+      "loss": 1.8527,
+      "step": 6800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0001,
+      "loss": 1.8294,
+      "step": 6900
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0001,
+      "loss": 1.871,
+      "step": 7000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0001,
+      "loss": 1.8634,
+      "step": 7100
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0001,
+      "loss": 1.8614,
+      "step": 7200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0001,
+      "loss": 1.8229,
+      "step": 7300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0001,
+      "loss": 1.8955,
+      "step": 7400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0001,
+      "loss": 1.8379,
+      "step": 7500
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0001,
+      "loss": 1.8533,
+      "step": 7600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0001,
+      "loss": 1.8454,
+      "step": 7700
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0001,
+      "loss": 1.8168,
+      "step": 7800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0001,
+      "loss": 1.8283,
+      "step": 7900
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0001,
+      "loss": 1.8545,
+      "step": 8000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0001,
+      "loss": 1.8133,
+      "step": 8100
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0001,
+      "loss": 1.8389,
+      "step": 8200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0001,
+      "loss": 1.8246,
+      "step": 8300
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0001,
+      "loss": 1.8351,
+      "step": 8400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0001,
+      "loss": 1.8244,
+      "step": 8500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0001,
+      "loss": 1.8137,
+      "step": 8600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0001,
+      "loss": 1.8016,
+      "step": 8700
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001,
+      "loss": 1.8237,
+      "step": 8800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001,
+      "loss": 1.812,
+      "step": 8900
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001,
+      "loss": 1.8345,
+      "step": 9000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001,
+      "loss": 1.8411,
+      "step": 9100
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0001,
+      "loss": 1.8077,
+      "step": 9200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001,
+      "loss": 1.8581,
+      "step": 9300
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001,
+      "loss": 1.8242,
+      "step": 9400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001,
+      "loss": 1.8166,
+      "step": 9500
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0001,
+      "loss": 1.8227,
+      "step": 9600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001,
+      "loss": 1.8011,
+      "step": 9700
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001,
+      "loss": 1.8253,
+      "step": 9800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001,
+      "loss": 1.7924,
+      "step": 9900
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0001,
+      "loss": 1.8091,
+      "step": 10000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001,
+      "loss": 1.8178,
+      "step": 10100
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001,
+      "loss": 1.811,
+      "step": 10200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001,
+      "loss": 1.8134,
+      "step": 10300
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001,
+      "loss": 1.8311,
+      "step": 10400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001,
+      "loss": 1.8269,
+      "step": 10500
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001,
+      "loss": 1.8286,
+      "step": 10600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0001,
+      "loss": 1.805,
+      "step": 10700
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001,
+      "loss": 1.7735,
+      "step": 10800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001,
+      "loss": 1.761,
+      "step": 10900
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001,
+      "loss": 1.7869,
+      "step": 11000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001,
+      "loss": 1.8054,
+      "step": 11100
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0001,
+      "loss": 1.7958,
+      "step": 11200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001,
+      "loss": 1.8049,
+      "step": 11300
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001,
+      "loss": 1.7943,
+      "step": 11400
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001,
+      "loss": 1.7771,
+      "step": 11500
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001,
+      "loss": 1.7981,
+      "step": 11600
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0001,
+      "loss": 1.8178,
+      "step": 11700
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0001,
+      "loss": 1.7732,
+      "step": 11800
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0001,
+      "loss": 1.7991,
+      "step": 11900
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001,
+      "loss": 1.7949,
+      "step": 12000
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001,
+      "loss": 1.7829,
+      "step": 12100
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001,
+      "loss": 1.7816,
+      "step": 12200
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0001,
+      "loss": 1.7896,
+      "step": 12300
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001,
+      "loss": 1.7797,
+      "step": 12400
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001,
+      "loss": 1.7965,
+      "step": 12500
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001,
+      "loss": 1.7855,
+      "step": 12600
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0001,
+      "loss": 1.7883,
+      "step": 12700
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001,
+      "loss": 1.7606,
+      "step": 12800
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001,
+      "loss": 1.7792,
+      "step": 12900
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001,
+      "loss": 1.7773,
+      "step": 13000
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001,
+      "loss": 1.7914,
+      "step": 13100
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001,
+      "loss": 1.787,
+      "step": 13200
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001,
+      "loss": 1.7892,
+      "step": 13300
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001,
+      "loss": 1.7676,
+      "step": 13400
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001,
+      "loss": 1.7775,
+      "step": 13500
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0001,
+      "loss": 1.7874,
+      "step": 13600
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001,
+      "loss": 1.7517,
+      "step": 13700
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0001,
+      "loss": 1.7669,
+      "step": 13800
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0001,
+      "loss": 1.7647,
+      "step": 13900
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001,
+      "loss": 1.7818,
+      "step": 14000
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001,
+      "loss": 1.7826,
+      "step": 14100
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001,
+      "loss": 1.7929,
+      "step": 14200
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.7685546875,
+      "eval_runtime": 3.0679,
+      "eval_samples_per_second": 761.101,
+      "eval_steps_per_second": 95.178,
+      "step": 14299
+    }
+  ],
+  "max_steps": 14299,
+  "num_train_epochs": 1,
+  "total_flos": 4.847476972453888e+16,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8322db256e84b969591d3b4f264ac692df836629441824bc6ceee156e244871
+size 4347

zero_to_fp32.py

@@ -0,0 +1,482 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
+                                                "*_optim_states.pt")),
+                         key=natural_keys)
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)