diff --git a/s1/README.md b/s1/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s1/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s1/adapter_config.json b/s1/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..62e40fe2e14356ef40bafe4c78b2349cf487c2bd
--- /dev/null
+++ b/s1/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "k_proj",
+ "up_proj",
+ "down_proj",
+ "gate_proj",
+ "v_proj",
+ "q_proj",
+ "o_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s1/adapter_model.bin b/s1/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..30b4889b48b85545c02c6da7af3bb1640aa9b9e5
--- /dev/null
+++ b/s1/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d3febffb0fd3e2a025ce3cb029bebed8acaa25e1000903988aaeb13108c285e
+size 639786637
diff --git a/s1/config.json b/s1/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s1/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s1/global_step16000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s1/global_step16000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..48dae3a95d19ca5e7a9aa29272cb5c3304b4b7a7
--- /dev/null
+++ b/s1/global_step16000/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:d7f8eef59b5a3c0c4ec6e82319eeef44e4323b2c8ac07d75566dff8a351b157f
+size 4089599575
diff --git a/s1/global_step16000/mp_rank_00_model_states.pt b/s1/global_step16000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f3cb7436fbfb6d66ff3c92c757314f9eedfc34d7
--- /dev/null
+++ b/s1/global_step16000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d28eb31954dc91f37189a42f4995323cdb662187f21669e7b07705c7ccedd678
+size 28850200126
diff --git a/s1/latest b/s1/latest
new file mode 100644
index 0000000000000000000000000000000000000000..76800d601529826c2ec8d7453b614b895cf74f2a
--- /dev/null
+++ b/s1/latest
@@ -0,0 +1 @@
+global_step16000
\ No newline at end of file
diff --git a/s1/non_lora_trainables.bin b/s1/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..5d17682a3a9780abdaa6c825953ef87c1c9ff377
--- /dev/null
+++ b/s1/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4198193940865ffcf86b34f70405ce8adbc1c0071f1b49d6024360adce75538
+size 41961191
diff --git a/s1/rng_state.pth b/s1/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..4793aa7bb1f74fdc8a4fd1dd7b2f3f05452ff2c1
--- /dev/null
+++ b/s1/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f2b92e3dbbeb0695186b26a8876ed34fe0191fb75d370352321df9b44c28b69
+size 14575
diff --git a/s1/special_tokens_map.json b/s1/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s1/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s1/tokenizer.model b/s1/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s1/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s1/tokenizer_config.json b/s1/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s1/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s1/trainer_state.json b/s1/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8df87240ac0c543259492722357abf2f13b3c371
--- /dev/null
+++ b/s1/trainer_state.json
@@ -0,0 +1,9616 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 35.714285714285715,
+ "global_step": 16000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 7.3,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 7.4188,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.2781,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.4313,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 3.7781,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.1906,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 3.1437,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.8531,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.5203,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 2.3047,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 1.8922,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 1.6977,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.5375,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 1.3445,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.2848,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.1387,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.0102,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 0.8014,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 1.0227,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 0.9275,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 0.723,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 0.7512,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 0.7262,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 0.7047,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 0.5461,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 0.5694,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 0.5791,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 0.5969,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 0.8695,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 0.5192,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 0.5953,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 0.5244,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 0.4699,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 0.534,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 0.6418,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 0.5691,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 0.317,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 0.5244,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 0.4479,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 0.6146,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 0.4246,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 0.3758,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 0.4468,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 0.474,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 0.4896,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 0.4052,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 0.4098,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.479,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 0.5396,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 0.4161,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.4158,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.3825,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.3144,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.3358,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 0.3859,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.3961,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 0.2987,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.3139,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 0.3496,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.2793,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.441,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 0.3758,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.3647,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 0.4186,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.3077,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.4566,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 0.3208,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.3359,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.4198,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.3153,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.3498,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.4084,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 0.4025,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.3126,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.3003,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.2281,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.2701,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.4742,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.5,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.3385,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.375,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.4436,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 0.3574,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.2836,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.2996,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.3642,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.3561,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.3864,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.4053,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.2954,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.2247,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.2319,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.2724,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.2742,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.304,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.2624,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.1891,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.1701,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.2985,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.2728,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.2263,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.3595,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.2502,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.2602,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.2353,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.2864,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.2549,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.2569,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.346,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.2758,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.2818,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.2336,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.2459,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.2203,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.349,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.2357,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.2682,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.3163,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.2597,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.2586,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.2774,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.2304,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.2816,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.2818,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.2634,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.3213,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.2658,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.2895,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.2845,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.2872,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.1743,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.2832,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.1807,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.2162,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.2186,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.1903,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.1722,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.1993,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.2429,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.2275,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.1918,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.1879,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.1871,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.207,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.2684,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.1758,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.1617,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.1919,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.213,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.1901,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.1899,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.2066,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.208,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.1795,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.1484,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.2485,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.1768,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.2225,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.1879,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.1523,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.1863,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.1984,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.1988,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.2691,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.199,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.1507,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.2334,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.1837,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.2155,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.1941,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.2036,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.1872,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.1654,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.1707,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.2393,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.1844,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.1114,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.2215,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.2457,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.1481,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.184,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.1247,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.1488,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.1207,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.1539,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.1532,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.1507,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.1482,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.1539,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.1815,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.1599,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.1358,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.2022,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.1565,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.1713,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.1467,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.1751,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.1524,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.1284,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.145,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.1589,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.1436,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.1622,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.1723,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.1154,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.1668,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.1912,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.1866,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.1316,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.1384,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.1384,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.1797,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.1349,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.1599,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.1425,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.1757,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.1569,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.188,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.1831,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.1615,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.143,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.1169,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.1214,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.1615,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.1083,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.0959,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.14,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.0995,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.1063,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.093,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.1012,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.0932,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.1145,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.1127,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.103,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.1051,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.1152,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.1153,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.1182,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.1193,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.1363,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.1243,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.0946,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.1122,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.1112,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.1459,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.1563,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.0993,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.1138,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.1105,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.1059,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.132,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.128,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.1497,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.1193,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.0972,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.1008,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.1331,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.1554,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.1272,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.1478,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.1736,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.1382,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.126,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.0965,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.1246,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.1268,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.1436,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.131,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.0908,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.1066,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.1081,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.0919,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.1009,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.1063,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.0909,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.0845,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.1061,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.1044,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.1013,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.1001,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.1005,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.0757,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.1091,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.1011,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.1287,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.1094,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.0935,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.116,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.1037,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.1147,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.1013,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.1175,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.0938,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.1181,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.0857,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.0974,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.0948,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.1108,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.1073,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.1024,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.1076,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.1438,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.1146,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.1113,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.0926,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.1224,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.0918,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.0947,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.1178,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.0936,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.092,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.109,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.1211,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.0925,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.0824,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.0845,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.0835,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.0783,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.0878,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.0801,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.0836,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.0643,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.0643,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.0996,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.0918,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.0928,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.0855,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.0825,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.0839,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.0694,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.0892,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.0861,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.0997,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.0998,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.0812,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.0952,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.0956,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.0839,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.0962,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.1016,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.1068,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.1113,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.1018,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.1005,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.0917,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.0819,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.095,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.0829,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.102,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.0858,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.1062,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.0797,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.1038,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.0892,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.1083,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.1154,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.0999,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.0845,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.0834,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.0728,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.0952,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.0746,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.082,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.0686,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.0743,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.0874,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.0841,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.069,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.0934,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.0861,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.0859,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.0878,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.0893,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.0787,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.0717,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.0753,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.0805,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.0775,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.0731,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.0833,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.0736,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.07,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.0734,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.1058,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.0829,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.0993,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.0743,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.0878,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.0746,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.0912,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.078,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.0827,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.0905,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.0847,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.087,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.0956,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.0832,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.0844,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.1053,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.0903,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.0692,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.0813,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.0725,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.0726,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.0786,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.0751,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.0569,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.0659,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.112,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.0757,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.0827,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.0715,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.0709,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.0658,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.09,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.0654,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.0894,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.0701,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.075,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.073,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.0802,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.0721,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.071,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.0996,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.0915,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.0887,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.0699,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.0902,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.0858,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.0821,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.078,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.0736,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.0705,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.0804,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.0862,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.0801,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.0794,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.0726,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.0806,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.0738,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.0756,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.0844,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.0892,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.0844,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.0683,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.0776,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.0686,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.0724,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.0603,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.0674,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.0748,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.065,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.066,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.0586,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.0672,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.0595,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.0797,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.053,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.0658,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.0763,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.0655,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.0699,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.0848,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.0705,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.0687,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.0635,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.0712,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.0647,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.0784,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.0726,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.078,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.0822,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.09,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.089,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.0726,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.0814,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.0645,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.0692,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.0703,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.0803,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.0782,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.0807,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.0814,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.0716,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.0668,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.0736,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.0672,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.0713,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.074,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.0845,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.076,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.0778,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.0679,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.0628,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.067,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.0652,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.0649,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.0616,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.0677,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.0691,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.0674,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.074,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.0656,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.0621,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.0506,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.0647,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.0679,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.0696,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.0724,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.073,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.0599,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.083,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.0745,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.067,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.0705,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.0739,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.0716,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.071,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.0675,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.0712,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.0737,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.0666,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.0625,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.0786,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.0667,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.0687,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.0738,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.0686,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.0709,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.0571,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.0787,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.0724,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.0927,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.0923,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.0708,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.071,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.0684,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.0652,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.0686,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.059,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.0643,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.0611,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.0559,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.0766,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.0704,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.0643,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.0585,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.0691,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.0669,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.0665,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.0629,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.0782,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.0715,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.0629,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0552,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.0649,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0619,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.0785,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.0717,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.0623,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.0711,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.0623,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.0686,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.0586,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.0569,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.0613,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.061,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.0579,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.0658,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.0633,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.0597,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.0583,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.0621,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.0644,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.0758,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.062,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.0758,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.0809,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.0745,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.0659,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.0792,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.0553,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.0641,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.0648,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.0508,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.0543,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.059,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.0695,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.0657,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.0698,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.0662,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.0664,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.0633,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.059,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.0727,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.0596,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.0794,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0643,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.0663,
+ "step": 6000
+ },
+ {
+ "epoch": 13.42,
+ "learning_rate": 1.5495257562515308e-05,
+ "loss": 0.0799,
+ "step": 6010
+ },
+ {
+ "epoch": 13.44,
+ "learning_rate": 1.5480148332320562e-05,
+ "loss": 0.0676,
+ "step": 6020
+ },
+ {
+ "epoch": 13.46,
+ "learning_rate": 1.5465021200501046e-05,
+ "loss": 0.0634,
+ "step": 6030
+ },
+ {
+ "epoch": 13.48,
+ "learning_rate": 1.5449876216471525e-05,
+ "loss": 0.0636,
+ "step": 6040
+ },
+ {
+ "epoch": 13.5,
+ "learning_rate": 1.5434713429705078e-05,
+ "loss": 0.062,
+ "step": 6050
+ },
+ {
+ "epoch": 13.53,
+ "learning_rate": 1.5419532889732943e-05,
+ "loss": 0.0562,
+ "step": 6060
+ },
+ {
+ "epoch": 13.55,
+ "learning_rate": 1.540433464614435e-05,
+ "loss": 0.064,
+ "step": 6070
+ },
+ {
+ "epoch": 13.57,
+ "learning_rate": 1.5389118748586357e-05,
+ "loss": 0.0563,
+ "step": 6080
+ },
+ {
+ "epoch": 13.59,
+ "learning_rate": 1.537388524676369e-05,
+ "loss": 0.0739,
+ "step": 6090
+ },
+ {
+ "epoch": 13.62,
+ "learning_rate": 1.5358634190438592e-05,
+ "loss": 0.055,
+ "step": 6100
+ },
+ {
+ "epoch": 13.64,
+ "learning_rate": 1.5343365629430638e-05,
+ "loss": 0.056,
+ "step": 6110
+ },
+ {
+ "epoch": 13.66,
+ "learning_rate": 1.5328079613616592e-05,
+ "loss": 0.0601,
+ "step": 6120
+ },
+ {
+ "epoch": 13.68,
+ "learning_rate": 1.531277619293023e-05,
+ "loss": 0.0845,
+ "step": 6130
+ },
+ {
+ "epoch": 13.71,
+ "learning_rate": 1.5297455417362194e-05,
+ "loss": 0.0808,
+ "step": 6140
+ },
+ {
+ "epoch": 13.73,
+ "learning_rate": 1.52821173369598e-05,
+ "loss": 0.0654,
+ "step": 6150
+ },
+ {
+ "epoch": 13.75,
+ "learning_rate": 1.526676200182691e-05,
+ "loss": 0.0638,
+ "step": 6160
+ },
+ {
+ "epoch": 13.77,
+ "learning_rate": 1.5251389462123748e-05,
+ "loss": 0.0692,
+ "step": 6170
+ },
+ {
+ "epoch": 13.79,
+ "learning_rate": 1.5235999768066729e-05,
+ "loss": 0.0699,
+ "step": 6180
+ },
+ {
+ "epoch": 13.82,
+ "learning_rate": 1.5220592969928313e-05,
+ "loss": 0.0719,
+ "step": 6190
+ },
+ {
+ "epoch": 13.84,
+ "learning_rate": 1.5205169118036831e-05,
+ "loss": 0.0645,
+ "step": 6200
+ },
+ {
+ "epoch": 13.86,
+ "learning_rate": 1.5189728262776325e-05,
+ "loss": 0.0621,
+ "step": 6210
+ },
+ {
+ "epoch": 13.88,
+ "learning_rate": 1.5174270454586375e-05,
+ "loss": 0.0638,
+ "step": 6220
+ },
+ {
+ "epoch": 13.91,
+ "learning_rate": 1.5158795743961942e-05,
+ "loss": 0.066,
+ "step": 6230
+ },
+ {
+ "epoch": 13.93,
+ "learning_rate": 1.5143304181453204e-05,
+ "loss": 0.0657,
+ "step": 6240
+ },
+ {
+ "epoch": 13.95,
+ "learning_rate": 1.5127795817665389e-05,
+ "loss": 0.0544,
+ "step": 6250
+ },
+ {
+ "epoch": 13.97,
+ "learning_rate": 1.5112270703258602e-05,
+ "loss": 0.0689,
+ "step": 6260
+ },
+ {
+ "epoch": 14.0,
+ "learning_rate": 1.5096728888947669e-05,
+ "loss": 0.0641,
+ "step": 6270
+ },
+ {
+ "epoch": 14.02,
+ "learning_rate": 1.508117042550197e-05,
+ "loss": 0.0563,
+ "step": 6280
+ },
+ {
+ "epoch": 14.04,
+ "learning_rate": 1.5065595363745272e-05,
+ "loss": 0.0564,
+ "step": 6290
+ },
+ {
+ "epoch": 14.06,
+ "learning_rate": 1.505000375455556e-05,
+ "loss": 0.0503,
+ "step": 6300
+ },
+ {
+ "epoch": 14.08,
+ "learning_rate": 1.503439564886487e-05,
+ "loss": 0.0578,
+ "step": 6310
+ },
+ {
+ "epoch": 14.11,
+ "learning_rate": 1.501877109765914e-05,
+ "loss": 0.0562,
+ "step": 6320
+ },
+ {
+ "epoch": 14.13,
+ "learning_rate": 1.5003130151978012e-05,
+ "loss": 0.0533,
+ "step": 6330
+ },
+ {
+ "epoch": 14.15,
+ "learning_rate": 1.4987472862914697e-05,
+ "loss": 0.0536,
+ "step": 6340
+ },
+ {
+ "epoch": 14.17,
+ "learning_rate": 1.4971799281615782e-05,
+ "loss": 0.0577,
+ "step": 6350
+ },
+ {
+ "epoch": 14.2,
+ "learning_rate": 1.4956109459281083e-05,
+ "loss": 0.0517,
+ "step": 6360
+ },
+ {
+ "epoch": 14.22,
+ "learning_rate": 1.4940403447163467e-05,
+ "loss": 0.0559,
+ "step": 6370
+ },
+ {
+ "epoch": 14.24,
+ "learning_rate": 1.4924681296568689e-05,
+ "loss": 0.0509,
+ "step": 6380
+ },
+ {
+ "epoch": 14.26,
+ "learning_rate": 1.4908943058855213e-05,
+ "loss": 0.0703,
+ "step": 6390
+ },
+ {
+ "epoch": 14.29,
+ "learning_rate": 1.4893188785434067e-05,
+ "loss": 0.0534,
+ "step": 6400
+ },
+ {
+ "epoch": 14.31,
+ "learning_rate": 1.4877418527768654e-05,
+ "loss": 0.0651,
+ "step": 6410
+ },
+ {
+ "epoch": 14.33,
+ "learning_rate": 1.4861632337374596e-05,
+ "loss": 0.0488,
+ "step": 6420
+ },
+ {
+ "epoch": 14.35,
+ "learning_rate": 1.4845830265819552e-05,
+ "loss": 0.0588,
+ "step": 6430
+ },
+ {
+ "epoch": 14.38,
+ "learning_rate": 1.483001236472307e-05,
+ "loss": 0.0554,
+ "step": 6440
+ },
+ {
+ "epoch": 14.4,
+ "learning_rate": 1.4814178685756405e-05,
+ "loss": 0.0624,
+ "step": 6450
+ },
+ {
+ "epoch": 14.42,
+ "learning_rate": 1.4798329280642345e-05,
+ "loss": 0.0527,
+ "step": 6460
+ },
+ {
+ "epoch": 14.44,
+ "learning_rate": 1.4782464201155057e-05,
+ "loss": 0.0509,
+ "step": 6470
+ },
+ {
+ "epoch": 14.46,
+ "learning_rate": 1.476658349911991e-05,
+ "loss": 0.0486,
+ "step": 6480
+ },
+ {
+ "epoch": 14.49,
+ "learning_rate": 1.4750687226413305e-05,
+ "loss": 0.0623,
+ "step": 6490
+ },
+ {
+ "epoch": 14.51,
+ "learning_rate": 1.4734775434962504e-05,
+ "loss": 0.0464,
+ "step": 6500
+ },
+ {
+ "epoch": 14.53,
+ "learning_rate": 1.471884817674546e-05,
+ "loss": 0.0514,
+ "step": 6510
+ },
+ {
+ "epoch": 14.55,
+ "learning_rate": 1.4702905503790668e-05,
+ "loss": 0.0558,
+ "step": 6520
+ },
+ {
+ "epoch": 14.58,
+ "learning_rate": 1.4686947468176955e-05,
+ "loss": 0.06,
+ "step": 6530
+ },
+ {
+ "epoch": 14.6,
+ "learning_rate": 1.467097412203334e-05,
+ "loss": 0.0583,
+ "step": 6540
+ },
+ {
+ "epoch": 14.62,
+ "learning_rate": 1.4654985517538864e-05,
+ "loss": 0.0605,
+ "step": 6550
+ },
+ {
+ "epoch": 14.64,
+ "learning_rate": 1.4638981706922401e-05,
+ "loss": 0.0641,
+ "step": 6560
+ },
+ {
+ "epoch": 14.67,
+ "learning_rate": 1.4622962742462503e-05,
+ "loss": 0.0669,
+ "step": 6570
+ },
+ {
+ "epoch": 14.69,
+ "learning_rate": 1.4606928676487223e-05,
+ "loss": 0.0691,
+ "step": 6580
+ },
+ {
+ "epoch": 14.71,
+ "learning_rate": 1.459087956137394e-05,
+ "loss": 0.0713,
+ "step": 6590
+ },
+ {
+ "epoch": 14.73,
+ "learning_rate": 1.4574815449549209e-05,
+ "loss": 0.0704,
+ "step": 6600
+ },
+ {
+ "epoch": 14.75,
+ "learning_rate": 1.4558736393488553e-05,
+ "loss": 0.0554,
+ "step": 6610
+ },
+ {
+ "epoch": 14.78,
+ "learning_rate": 1.4542642445716326e-05,
+ "loss": 0.0519,
+ "step": 6620
+ },
+ {
+ "epoch": 14.8,
+ "learning_rate": 1.4526533658805517e-05,
+ "loss": 0.0614,
+ "step": 6630
+ },
+ {
+ "epoch": 14.82,
+ "learning_rate": 1.4510410085377606e-05,
+ "loss": 0.0683,
+ "step": 6640
+ },
+ {
+ "epoch": 14.84,
+ "learning_rate": 1.4494271778102358e-05,
+ "loss": 0.0525,
+ "step": 6650
+ },
+ {
+ "epoch": 14.87,
+ "learning_rate": 1.4478118789697675e-05,
+ "loss": 0.0677,
+ "step": 6660
+ },
+ {
+ "epoch": 14.89,
+ "learning_rate": 1.4461951172929419e-05,
+ "loss": 0.0625,
+ "step": 6670
+ },
+ {
+ "epoch": 14.91,
+ "learning_rate": 1.4445768980611233e-05,
+ "loss": 0.0592,
+ "step": 6680
+ },
+ {
+ "epoch": 14.93,
+ "learning_rate": 1.4429572265604375e-05,
+ "loss": 0.0616,
+ "step": 6690
+ },
+ {
+ "epoch": 14.96,
+ "learning_rate": 1.4413361080817545e-05,
+ "loss": 0.0844,
+ "step": 6700
+ },
+ {
+ "epoch": 14.98,
+ "learning_rate": 1.4397135479206705e-05,
+ "loss": 0.0702,
+ "step": 6710
+ },
+ {
+ "epoch": 15.0,
+ "learning_rate": 1.4380895513774922e-05,
+ "loss": 0.067,
+ "step": 6720
+ },
+ {
+ "epoch": 15.02,
+ "learning_rate": 1.436464123757217e-05,
+ "loss": 0.0541,
+ "step": 6730
+ },
+ {
+ "epoch": 15.04,
+ "learning_rate": 1.4348372703695184e-05,
+ "loss": 0.0521,
+ "step": 6740
+ },
+ {
+ "epoch": 15.07,
+ "learning_rate": 1.4332089965287266e-05,
+ "loss": 0.0576,
+ "step": 6750
+ },
+ {
+ "epoch": 15.09,
+ "learning_rate": 1.431579307553812e-05,
+ "loss": 0.0666,
+ "step": 6760
+ },
+ {
+ "epoch": 15.11,
+ "learning_rate": 1.429948208768368e-05,
+ "loss": 0.0724,
+ "step": 6770
+ },
+ {
+ "epoch": 15.13,
+ "learning_rate": 1.4283157055005928e-05,
+ "loss": 0.1002,
+ "step": 6780
+ },
+ {
+ "epoch": 15.16,
+ "learning_rate": 1.4266818030832732e-05,
+ "loss": 0.0808,
+ "step": 6790
+ },
+ {
+ "epoch": 15.18,
+ "learning_rate": 1.4250465068537664e-05,
+ "loss": 0.0491,
+ "step": 6800
+ },
+ {
+ "epoch": 15.2,
+ "learning_rate": 1.4234098221539818e-05,
+ "loss": 0.0626,
+ "step": 6810
+ },
+ {
+ "epoch": 15.22,
+ "learning_rate": 1.4217717543303657e-05,
+ "loss": 0.0664,
+ "step": 6820
+ },
+ {
+ "epoch": 15.25,
+ "learning_rate": 1.4201323087338816e-05,
+ "loss": 0.071,
+ "step": 6830
+ },
+ {
+ "epoch": 15.27,
+ "learning_rate": 1.4184914907199942e-05,
+ "loss": 0.0525,
+ "step": 6840
+ },
+ {
+ "epoch": 15.29,
+ "learning_rate": 1.4168493056486512e-05,
+ "loss": 0.067,
+ "step": 6850
+ },
+ {
+ "epoch": 15.31,
+ "learning_rate": 1.4152057588842657e-05,
+ "loss": 0.0594,
+ "step": 6860
+ },
+ {
+ "epoch": 15.33,
+ "learning_rate": 1.4135608557956992e-05,
+ "loss": 0.0624,
+ "step": 6870
+ },
+ {
+ "epoch": 15.36,
+ "learning_rate": 1.4119146017562441e-05,
+ "loss": 0.0559,
+ "step": 6880
+ },
+ {
+ "epoch": 15.38,
+ "learning_rate": 1.4102670021436059e-05,
+ "loss": 0.0622,
+ "step": 6890
+ },
+ {
+ "epoch": 15.4,
+ "learning_rate": 1.4086180623398842e-05,
+ "loss": 0.0514,
+ "step": 6900
+ },
+ {
+ "epoch": 15.42,
+ "learning_rate": 1.4069677877315587e-05,
+ "loss": 0.0725,
+ "step": 6910
+ },
+ {
+ "epoch": 15.45,
+ "learning_rate": 1.4053161837094675e-05,
+ "loss": 0.0686,
+ "step": 6920
+ },
+ {
+ "epoch": 15.47,
+ "learning_rate": 1.4036632556687927e-05,
+ "loss": 0.0502,
+ "step": 6930
+ },
+ {
+ "epoch": 15.49,
+ "learning_rate": 1.4020090090090408e-05,
+ "loss": 0.0618,
+ "step": 6940
+ },
+ {
+ "epoch": 15.51,
+ "learning_rate": 1.4003534491340259e-05,
+ "loss": 0.0582,
+ "step": 6950
+ },
+ {
+ "epoch": 15.54,
+ "learning_rate": 1.3986965814518521e-05,
+ "loss": 0.0679,
+ "step": 6960
+ },
+ {
+ "epoch": 15.56,
+ "learning_rate": 1.3970384113748951e-05,
+ "loss": 0.0717,
+ "step": 6970
+ },
+ {
+ "epoch": 15.58,
+ "learning_rate": 1.3953789443197857e-05,
+ "loss": 0.0585,
+ "step": 6980
+ },
+ {
+ "epoch": 15.6,
+ "learning_rate": 1.3937181857073912e-05,
+ "loss": 0.065,
+ "step": 6990
+ },
+ {
+ "epoch": 15.62,
+ "learning_rate": 1.3920561409627974e-05,
+ "loss": 0.0636,
+ "step": 7000
+ },
+ {
+ "epoch": 15.65,
+ "learning_rate": 1.3903928155152926e-05,
+ "loss": 0.0653,
+ "step": 7010
+ },
+ {
+ "epoch": 15.67,
+ "learning_rate": 1.3887282147983472e-05,
+ "loss": 0.0607,
+ "step": 7020
+ },
+ {
+ "epoch": 15.69,
+ "learning_rate": 1.3870623442495987e-05,
+ "loss": 0.0604,
+ "step": 7030
+ },
+ {
+ "epoch": 15.71,
+ "learning_rate": 1.3853952093108323e-05,
+ "loss": 0.0619,
+ "step": 7040
+ },
+ {
+ "epoch": 15.74,
+ "learning_rate": 1.3837268154279628e-05,
+ "loss": 0.0623,
+ "step": 7050
+ },
+ {
+ "epoch": 15.76,
+ "learning_rate": 1.3820571680510187e-05,
+ "loss": 0.0515,
+ "step": 7060
+ },
+ {
+ "epoch": 15.78,
+ "learning_rate": 1.3803862726341224e-05,
+ "loss": 0.0511,
+ "step": 7070
+ },
+ {
+ "epoch": 15.8,
+ "learning_rate": 1.3787141346354733e-05,
+ "loss": 0.0516,
+ "step": 7080
+ },
+ {
+ "epoch": 15.83,
+ "learning_rate": 1.3770407595173301e-05,
+ "loss": 0.0529,
+ "step": 7090
+ },
+ {
+ "epoch": 15.85,
+ "learning_rate": 1.375366152745992e-05,
+ "loss": 0.0549,
+ "step": 7100
+ },
+ {
+ "epoch": 15.87,
+ "learning_rate": 1.373690319791783e-05,
+ "loss": 0.0506,
+ "step": 7110
+ },
+ {
+ "epoch": 15.89,
+ "learning_rate": 1.3720132661290311e-05,
+ "loss": 0.0682,
+ "step": 7120
+ },
+ {
+ "epoch": 15.92,
+ "learning_rate": 1.3703349972360527e-05,
+ "loss": 0.0675,
+ "step": 7130
+ },
+ {
+ "epoch": 15.94,
+ "learning_rate": 1.3686555185951334e-05,
+ "loss": 0.068,
+ "step": 7140
+ },
+ {
+ "epoch": 15.96,
+ "learning_rate": 1.3669748356925112e-05,
+ "loss": 0.0608,
+ "step": 7150
+ },
+ {
+ "epoch": 15.98,
+ "learning_rate": 1.3652929540183578e-05,
+ "loss": 0.0833,
+ "step": 7160
+ },
+ {
+ "epoch": 16.0,
+ "learning_rate": 1.3636098790667605e-05,
+ "loss": 0.0644,
+ "step": 7170
+ },
+ {
+ "epoch": 16.03,
+ "learning_rate": 1.3619256163357046e-05,
+ "loss": 0.049,
+ "step": 7180
+ },
+ {
+ "epoch": 16.05,
+ "learning_rate": 1.3602401713270566e-05,
+ "loss": 0.0577,
+ "step": 7190
+ },
+ {
+ "epoch": 16.07,
+ "learning_rate": 1.3585535495465432e-05,
+ "loss": 0.0615,
+ "step": 7200
+ },
+ {
+ "epoch": 16.09,
+ "learning_rate": 1.3568657565037365e-05,
+ "loss": 0.0522,
+ "step": 7210
+ },
+ {
+ "epoch": 16.12,
+ "learning_rate": 1.3551767977120341e-05,
+ "loss": 0.0461,
+ "step": 7220
+ },
+ {
+ "epoch": 16.14,
+ "learning_rate": 1.353486678688642e-05,
+ "loss": 0.0538,
+ "step": 7230
+ },
+ {
+ "epoch": 16.16,
+ "learning_rate": 1.351795404954556e-05,
+ "loss": 0.0513,
+ "step": 7240
+ },
+ {
+ "epoch": 16.18,
+ "learning_rate": 1.3501029820345446e-05,
+ "loss": 0.0535,
+ "step": 7250
+ },
+ {
+ "epoch": 16.21,
+ "learning_rate": 1.3484094154571286e-05,
+ "loss": 0.048,
+ "step": 7260
+ },
+ {
+ "epoch": 16.23,
+ "learning_rate": 1.3467147107545668e-05,
+ "loss": 0.0614,
+ "step": 7270
+ },
+ {
+ "epoch": 16.25,
+ "learning_rate": 1.3450188734628344e-05,
+ "loss": 0.0501,
+ "step": 7280
+ },
+ {
+ "epoch": 16.27,
+ "learning_rate": 1.3433219091216069e-05,
+ "loss": 0.0547,
+ "step": 7290
+ },
+ {
+ "epoch": 16.29,
+ "learning_rate": 1.3416238232742414e-05,
+ "loss": 0.0592,
+ "step": 7300
+ },
+ {
+ "epoch": 16.32,
+ "learning_rate": 1.3399246214677583e-05,
+ "loss": 0.0519,
+ "step": 7310
+ },
+ {
+ "epoch": 16.34,
+ "learning_rate": 1.338224309252824e-05,
+ "loss": 0.0604,
+ "step": 7320
+ },
+ {
+ "epoch": 16.36,
+ "learning_rate": 1.3365228921837314e-05,
+ "loss": 0.0633,
+ "step": 7330
+ },
+ {
+ "epoch": 16.38,
+ "learning_rate": 1.3348203758183831e-05,
+ "loss": 0.0501,
+ "step": 7340
+ },
+ {
+ "epoch": 16.41,
+ "learning_rate": 1.3331167657182726e-05,
+ "loss": 0.0611,
+ "step": 7350
+ },
+ {
+ "epoch": 16.43,
+ "learning_rate": 1.3314120674484663e-05,
+ "loss": 0.0504,
+ "step": 7360
+ },
+ {
+ "epoch": 16.45,
+ "learning_rate": 1.3297062865775851e-05,
+ "loss": 0.0458,
+ "step": 7370
+ },
+ {
+ "epoch": 16.47,
+ "learning_rate": 1.327999428677786e-05,
+ "loss": 0.0605,
+ "step": 7380
+ },
+ {
+ "epoch": 16.5,
+ "learning_rate": 1.3262914993247454e-05,
+ "loss": 0.0466,
+ "step": 7390
+ },
+ {
+ "epoch": 16.52,
+ "learning_rate": 1.324582504097638e-05,
+ "loss": 0.0542,
+ "step": 7400
+ },
+ {
+ "epoch": 16.54,
+ "learning_rate": 1.3228724485791225e-05,
+ "loss": 0.0618,
+ "step": 7410
+ },
+ {
+ "epoch": 16.56,
+ "learning_rate": 1.321161338355319e-05,
+ "loss": 0.0499,
+ "step": 7420
+ },
+ {
+ "epoch": 16.58,
+ "learning_rate": 1.3194491790157947e-05,
+ "loss": 0.0446,
+ "step": 7430
+ },
+ {
+ "epoch": 16.61,
+ "learning_rate": 1.3177359761535427e-05,
+ "loss": 0.0519,
+ "step": 7440
+ },
+ {
+ "epoch": 16.63,
+ "learning_rate": 1.3160217353649652e-05,
+ "loss": 0.0693,
+ "step": 7450
+ },
+ {
+ "epoch": 16.65,
+ "learning_rate": 1.3143064622498551e-05,
+ "loss": 0.0491,
+ "step": 7460
+ },
+ {
+ "epoch": 16.67,
+ "learning_rate": 1.312590162411378e-05,
+ "loss": 0.0634,
+ "step": 7470
+ },
+ {
+ "epoch": 16.7,
+ "learning_rate": 1.310872841456052e-05,
+ "loss": 0.0636,
+ "step": 7480
+ },
+ {
+ "epoch": 16.72,
+ "learning_rate": 1.3091545049937322e-05,
+ "loss": 0.0569,
+ "step": 7490
+ },
+ {
+ "epoch": 16.74,
+ "learning_rate": 1.3074351586375906e-05,
+ "loss": 0.0464,
+ "step": 7500
+ },
+ {
+ "epoch": 16.76,
+ "learning_rate": 1.305714808004098e-05,
+ "loss": 0.057,
+ "step": 7510
+ },
+ {
+ "epoch": 16.79,
+ "learning_rate": 1.3039934587130056e-05,
+ "loss": 0.0512,
+ "step": 7520
+ },
+ {
+ "epoch": 16.81,
+ "learning_rate": 1.3022711163873272e-05,
+ "loss": 0.0521,
+ "step": 7530
+ },
+ {
+ "epoch": 16.83,
+ "learning_rate": 1.3005477866533202e-05,
+ "loss": 0.06,
+ "step": 7540
+ },
+ {
+ "epoch": 16.85,
+ "learning_rate": 1.2988234751404683e-05,
+ "loss": 0.0666,
+ "step": 7550
+ },
+ {
+ "epoch": 16.88,
+ "learning_rate": 1.2970981874814613e-05,
+ "loss": 0.0698,
+ "step": 7560
+ },
+ {
+ "epoch": 16.9,
+ "learning_rate": 1.2953719293121775e-05,
+ "loss": 0.0471,
+ "step": 7570
+ },
+ {
+ "epoch": 16.92,
+ "learning_rate": 1.2936447062716668e-05,
+ "loss": 0.047,
+ "step": 7580
+ },
+ {
+ "epoch": 16.94,
+ "learning_rate": 1.2919165240021303e-05,
+ "loss": 0.0578,
+ "step": 7590
+ },
+ {
+ "epoch": 16.96,
+ "learning_rate": 1.2901873881489021e-05,
+ "loss": 0.0688,
+ "step": 7600
+ },
+ {
+ "epoch": 16.99,
+ "learning_rate": 1.288457304360432e-05,
+ "loss": 0.0538,
+ "step": 7610
+ },
+ {
+ "epoch": 17.01,
+ "learning_rate": 1.2867262782882662e-05,
+ "loss": 0.0549,
+ "step": 7620
+ },
+ {
+ "epoch": 17.03,
+ "learning_rate": 1.2849943155870284e-05,
+ "loss": 0.0577,
+ "step": 7630
+ },
+ {
+ "epoch": 17.05,
+ "learning_rate": 1.2832614219144027e-05,
+ "loss": 0.0405,
+ "step": 7640
+ },
+ {
+ "epoch": 17.08,
+ "learning_rate": 1.2815276029311138e-05,
+ "loss": 0.0487,
+ "step": 7650
+ },
+ {
+ "epoch": 17.1,
+ "learning_rate": 1.2797928643009097e-05,
+ "loss": 0.0517,
+ "step": 7660
+ },
+ {
+ "epoch": 17.12,
+ "learning_rate": 1.2780572116905418e-05,
+ "loss": 0.048,
+ "step": 7670
+ },
+ {
+ "epoch": 17.14,
+ "learning_rate": 1.276320650769748e-05,
+ "loss": 0.0614,
+ "step": 7680
+ },
+ {
+ "epoch": 17.17,
+ "learning_rate": 1.2745831872112318e-05,
+ "loss": 0.0568,
+ "step": 7690
+ },
+ {
+ "epoch": 17.19,
+ "learning_rate": 1.2728448266906468e-05,
+ "loss": 0.0484,
+ "step": 7700
+ },
+ {
+ "epoch": 17.21,
+ "learning_rate": 1.2711055748865765e-05,
+ "loss": 0.0491,
+ "step": 7710
+ },
+ {
+ "epoch": 17.23,
+ "learning_rate": 1.2693654374805148e-05,
+ "loss": 0.0507,
+ "step": 7720
+ },
+ {
+ "epoch": 17.25,
+ "learning_rate": 1.2676244201568498e-05,
+ "loss": 0.0489,
+ "step": 7730
+ },
+ {
+ "epoch": 17.28,
+ "learning_rate": 1.2658825286028428e-05,
+ "loss": 0.05,
+ "step": 7740
+ },
+ {
+ "epoch": 17.3,
+ "learning_rate": 1.2641397685086124e-05,
+ "loss": 0.0606,
+ "step": 7750
+ },
+ {
+ "epoch": 17.32,
+ "learning_rate": 1.2623961455671125e-05,
+ "loss": 0.0548,
+ "step": 7760
+ },
+ {
+ "epoch": 17.34,
+ "learning_rate": 1.2606516654741172e-05,
+ "loss": 0.0446,
+ "step": 7770
+ },
+ {
+ "epoch": 17.37,
+ "learning_rate": 1.2589063339281995e-05,
+ "loss": 0.0643,
+ "step": 7780
+ },
+ {
+ "epoch": 17.39,
+ "learning_rate": 1.257160156630715e-05,
+ "loss": 0.0438,
+ "step": 7790
+ },
+ {
+ "epoch": 17.41,
+ "learning_rate": 1.2554131392857812e-05,
+ "loss": 0.0501,
+ "step": 7800
+ },
+ {
+ "epoch": 17.43,
+ "learning_rate": 1.253665287600259e-05,
+ "loss": 0.0748,
+ "step": 7810
+ },
+ {
+ "epoch": 17.46,
+ "learning_rate": 1.2519166072837368e-05,
+ "loss": 0.0482,
+ "step": 7820
+ },
+ {
+ "epoch": 17.48,
+ "learning_rate": 1.250167104048508e-05,
+ "loss": 0.0579,
+ "step": 7830
+ },
+ {
+ "epoch": 17.5,
+ "learning_rate": 1.248416783609555e-05,
+ "loss": 0.055,
+ "step": 7840
+ },
+ {
+ "epoch": 17.52,
+ "learning_rate": 1.2466656516845293e-05,
+ "loss": 0.0455,
+ "step": 7850
+ },
+ {
+ "epoch": 17.54,
+ "learning_rate": 1.244913713993734e-05,
+ "loss": 0.0495,
+ "step": 7860
+ },
+ {
+ "epoch": 17.57,
+ "learning_rate": 1.2431609762601036e-05,
+ "loss": 0.0436,
+ "step": 7870
+ },
+ {
+ "epoch": 17.59,
+ "learning_rate": 1.241407444209186e-05,
+ "loss": 0.0562,
+ "step": 7880
+ },
+ {
+ "epoch": 17.61,
+ "learning_rate": 1.2396531235691245e-05,
+ "loss": 0.0432,
+ "step": 7890
+ },
+ {
+ "epoch": 17.63,
+ "learning_rate": 1.2378980200706376e-05,
+ "loss": 0.0534,
+ "step": 7900
+ },
+ {
+ "epoch": 17.66,
+ "learning_rate": 1.236142139447002e-05,
+ "loss": 0.0555,
+ "step": 7910
+ },
+ {
+ "epoch": 17.68,
+ "learning_rate": 1.2343854874340324e-05,
+ "loss": 0.0456,
+ "step": 7920
+ },
+ {
+ "epoch": 17.7,
+ "learning_rate": 1.2326280697700632e-05,
+ "loss": 0.0528,
+ "step": 7930
+ },
+ {
+ "epoch": 17.72,
+ "learning_rate": 1.2308698921959306e-05,
+ "loss": 0.0617,
+ "step": 7940
+ },
+ {
+ "epoch": 17.75,
+ "learning_rate": 1.2291109604549525e-05,
+ "loss": 0.0721,
+ "step": 7950
+ },
+ {
+ "epoch": 17.77,
+ "learning_rate": 1.2273512802929107e-05,
+ "loss": 0.0667,
+ "step": 7960
+ },
+ {
+ "epoch": 17.79,
+ "learning_rate": 1.2255908574580311e-05,
+ "loss": 0.0533,
+ "step": 7970
+ },
+ {
+ "epoch": 17.81,
+ "learning_rate": 1.2238296977009672e-05,
+ "loss": 0.0549,
+ "step": 7980
+ },
+ {
+ "epoch": 17.83,
+ "learning_rate": 1.2220678067747785e-05,
+ "loss": 0.0521,
+ "step": 7990
+ },
+ {
+ "epoch": 17.86,
+ "learning_rate": 1.2203051904349128e-05,
+ "loss": 0.0473,
+ "step": 8000
+ },
+ {
+ "epoch": 17.88,
+ "learning_rate": 1.2185418544391885e-05,
+ "loss": 0.0619,
+ "step": 8010
+ },
+ {
+ "epoch": 17.9,
+ "learning_rate": 1.2167778045477743e-05,
+ "loss": 0.0554,
+ "step": 8020
+ },
+ {
+ "epoch": 17.92,
+ "learning_rate": 1.215013046523171e-05,
+ "loss": 0.0544,
+ "step": 8030
+ },
+ {
+ "epoch": 17.95,
+ "learning_rate": 1.2132475861301928e-05,
+ "loss": 0.0525,
+ "step": 8040
+ },
+ {
+ "epoch": 17.97,
+ "learning_rate": 1.2114814291359476e-05,
+ "loss": 0.0522,
+ "step": 8050
+ },
+ {
+ "epoch": 17.99,
+ "learning_rate": 1.20971458130982e-05,
+ "loss": 0.0428,
+ "step": 8060
+ },
+ {
+ "epoch": 18.01,
+ "learning_rate": 1.20794704842345e-05,
+ "loss": 0.0498,
+ "step": 8070
+ },
+ {
+ "epoch": 18.04,
+ "learning_rate": 1.2061788362507168e-05,
+ "loss": 0.036,
+ "step": 8080
+ },
+ {
+ "epoch": 18.06,
+ "learning_rate": 1.204409950567717e-05,
+ "loss": 0.0364,
+ "step": 8090
+ },
+ {
+ "epoch": 18.08,
+ "learning_rate": 1.2026403971527487e-05,
+ "loss": 0.0404,
+ "step": 8100
+ },
+ {
+ "epoch": 18.1,
+ "learning_rate": 1.2008701817862906e-05,
+ "loss": 0.046,
+ "step": 8110
+ },
+ {
+ "epoch": 18.12,
+ "learning_rate": 1.1990993102509838e-05,
+ "loss": 0.0505,
+ "step": 8120
+ },
+ {
+ "epoch": 18.15,
+ "learning_rate": 1.1973277883316128e-05,
+ "loss": 0.0412,
+ "step": 8130
+ },
+ {
+ "epoch": 18.17,
+ "learning_rate": 1.1955556218150872e-05,
+ "loss": 0.0494,
+ "step": 8140
+ },
+ {
+ "epoch": 18.19,
+ "learning_rate": 1.1937828164904216e-05,
+ "loss": 0.045,
+ "step": 8150
+ },
+ {
+ "epoch": 18.21,
+ "learning_rate": 1.1920093781487175e-05,
+ "loss": 0.0433,
+ "step": 8160
+ },
+ {
+ "epoch": 18.24,
+ "learning_rate": 1.1902353125831441e-05,
+ "loss": 0.0395,
+ "step": 8170
+ },
+ {
+ "epoch": 18.26,
+ "learning_rate": 1.1884606255889203e-05,
+ "loss": 0.0387,
+ "step": 8180
+ },
+ {
+ "epoch": 18.28,
+ "learning_rate": 1.1866853229632942e-05,
+ "loss": 0.0528,
+ "step": 8190
+ },
+ {
+ "epoch": 18.3,
+ "learning_rate": 1.1849094105055248e-05,
+ "loss": 0.0441,
+ "step": 8200
+ },
+ {
+ "epoch": 18.33,
+ "learning_rate": 1.1831328940168638e-05,
+ "loss": 0.0459,
+ "step": 8210
+ },
+ {
+ "epoch": 18.35,
+ "learning_rate": 1.181355779300536e-05,
+ "loss": 0.0452,
+ "step": 8220
+ },
+ {
+ "epoch": 18.37,
+ "learning_rate": 1.1795780721617199e-05,
+ "loss": 0.0502,
+ "step": 8230
+ },
+ {
+ "epoch": 18.39,
+ "learning_rate": 1.1777997784075294e-05,
+ "loss": 0.0615,
+ "step": 8240
+ },
+ {
+ "epoch": 18.42,
+ "learning_rate": 1.176020903846995e-05,
+ "loss": 0.0523,
+ "step": 8250
+ },
+ {
+ "epoch": 18.44,
+ "learning_rate": 1.1742414542910444e-05,
+ "loss": 0.0456,
+ "step": 8260
+ },
+ {
+ "epoch": 18.46,
+ "learning_rate": 1.1724614355524832e-05,
+ "loss": 0.0528,
+ "step": 8270
+ },
+ {
+ "epoch": 18.48,
+ "learning_rate": 1.1706808534459768e-05,
+ "loss": 0.0466,
+ "step": 8280
+ },
+ {
+ "epoch": 18.5,
+ "learning_rate": 1.16889971378803e-05,
+ "loss": 0.0449,
+ "step": 8290
+ },
+ {
+ "epoch": 18.53,
+ "learning_rate": 1.1671180223969705e-05,
+ "loss": 0.0452,
+ "step": 8300
+ },
+ {
+ "epoch": 18.55,
+ "learning_rate": 1.1653357850929268e-05,
+ "loss": 0.0436,
+ "step": 8310
+ },
+ {
+ "epoch": 18.57,
+ "learning_rate": 1.1635530076978115e-05,
+ "loss": 0.0513,
+ "step": 8320
+ },
+ {
+ "epoch": 18.59,
+ "learning_rate": 1.161769696035301e-05,
+ "loss": 0.0385,
+ "step": 8330
+ },
+ {
+ "epoch": 18.62,
+ "learning_rate": 1.1599858559308175e-05,
+ "loss": 0.0472,
+ "step": 8340
+ },
+ {
+ "epoch": 18.64,
+ "learning_rate": 1.158201493211509e-05,
+ "loss": 0.054,
+ "step": 8350
+ },
+ {
+ "epoch": 18.66,
+ "learning_rate": 1.156416613706231e-05,
+ "loss": 0.0434,
+ "step": 8360
+ },
+ {
+ "epoch": 18.68,
+ "learning_rate": 1.1546312232455266e-05,
+ "loss": 0.0476,
+ "step": 8370
+ },
+ {
+ "epoch": 18.71,
+ "learning_rate": 1.152845327661609e-05,
+ "loss": 0.0477,
+ "step": 8380
+ },
+ {
+ "epoch": 18.73,
+ "learning_rate": 1.1510589327883406e-05,
+ "loss": 0.0511,
+ "step": 8390
+ },
+ {
+ "epoch": 18.75,
+ "learning_rate": 1.1492720444612148e-05,
+ "loss": 0.0511,
+ "step": 8400
+ },
+ {
+ "epoch": 18.77,
+ "learning_rate": 1.1474846685173374e-05,
+ "loss": 0.0495,
+ "step": 8410
+ },
+ {
+ "epoch": 18.79,
+ "learning_rate": 1.1456968107954066e-05,
+ "loss": 0.0524,
+ "step": 8420
+ },
+ {
+ "epoch": 18.82,
+ "learning_rate": 1.143908477135695e-05,
+ "loss": 0.0521,
+ "step": 8430
+ },
+ {
+ "epoch": 18.84,
+ "learning_rate": 1.1421196733800291e-05,
+ "loss": 0.0543,
+ "step": 8440
+ },
+ {
+ "epoch": 18.86,
+ "learning_rate": 1.1403304053717719e-05,
+ "loss": 0.05,
+ "step": 8450
+ },
+ {
+ "epoch": 18.88,
+ "learning_rate": 1.138540678955802e-05,
+ "loss": 0.0791,
+ "step": 8460
+ },
+ {
+ "epoch": 18.91,
+ "learning_rate": 1.1367504999784963e-05,
+ "loss": 0.0589,
+ "step": 8470
+ },
+ {
+ "epoch": 18.93,
+ "learning_rate": 1.1349598742877097e-05,
+ "loss": 0.0631,
+ "step": 8480
+ },
+ {
+ "epoch": 18.95,
+ "learning_rate": 1.1331688077327563e-05,
+ "loss": 0.0663,
+ "step": 8490
+ },
+ {
+ "epoch": 18.97,
+ "learning_rate": 1.1313773061643905e-05,
+ "loss": 0.0773,
+ "step": 8500
+ },
+ {
+ "epoch": 19.0,
+ "learning_rate": 1.1295853754347876e-05,
+ "loss": 0.0568,
+ "step": 8510
+ },
+ {
+ "epoch": 19.02,
+ "learning_rate": 1.1277930213975249e-05,
+ "loss": 0.0648,
+ "step": 8520
+ },
+ {
+ "epoch": 19.04,
+ "learning_rate": 1.1260002499075617e-05,
+ "loss": 0.0415,
+ "step": 8530
+ },
+ {
+ "epoch": 19.06,
+ "learning_rate": 1.1242070668212227e-05,
+ "loss": 0.0494,
+ "step": 8540
+ },
+ {
+ "epoch": 19.08,
+ "learning_rate": 1.1224134779961758e-05,
+ "loss": 0.0527,
+ "step": 8550
+ },
+ {
+ "epoch": 19.11,
+ "learning_rate": 1.1206194892914142e-05,
+ "loss": 0.0555,
+ "step": 8560
+ },
+ {
+ "epoch": 19.13,
+ "learning_rate": 1.1188251065672382e-05,
+ "loss": 0.0511,
+ "step": 8570
+ },
+ {
+ "epoch": 19.15,
+ "learning_rate": 1.117030335685235e-05,
+ "loss": 0.0421,
+ "step": 8580
+ },
+ {
+ "epoch": 19.17,
+ "learning_rate": 1.1152351825082588e-05,
+ "loss": 0.0443,
+ "step": 8590
+ },
+ {
+ "epoch": 19.2,
+ "learning_rate": 1.1134396529004143e-05,
+ "loss": 0.0497,
+ "step": 8600
+ },
+ {
+ "epoch": 19.22,
+ "learning_rate": 1.1116437527270343e-05,
+ "loss": 0.0441,
+ "step": 8610
+ },
+ {
+ "epoch": 19.24,
+ "learning_rate": 1.109847487854663e-05,
+ "loss": 0.0526,
+ "step": 8620
+ },
+ {
+ "epoch": 19.26,
+ "learning_rate": 1.1080508641510357e-05,
+ "loss": 0.0505,
+ "step": 8630
+ },
+ {
+ "epoch": 19.29,
+ "learning_rate": 1.1062538874850597e-05,
+ "loss": 0.0482,
+ "step": 8640
+ },
+ {
+ "epoch": 19.31,
+ "learning_rate": 1.1044565637267957e-05,
+ "loss": 0.0455,
+ "step": 8650
+ },
+ {
+ "epoch": 19.33,
+ "learning_rate": 1.1026588987474379e-05,
+ "loss": 0.0505,
+ "step": 8660
+ },
+ {
+ "epoch": 19.35,
+ "learning_rate": 1.100860898419295e-05,
+ "loss": 0.0438,
+ "step": 8670
+ },
+ {
+ "epoch": 19.38,
+ "learning_rate": 1.0990625686157714e-05,
+ "loss": 0.0418,
+ "step": 8680
+ },
+ {
+ "epoch": 19.4,
+ "learning_rate": 1.097263915211348e-05,
+ "loss": 0.047,
+ "step": 8690
+ },
+ {
+ "epoch": 19.42,
+ "learning_rate": 1.0954649440815625e-05,
+ "loss": 0.0493,
+ "step": 8700
+ },
+ {
+ "epoch": 19.44,
+ "learning_rate": 1.0936656611029901e-05,
+ "loss": 0.0391,
+ "step": 8710
+ },
+ {
+ "epoch": 19.46,
+ "learning_rate": 1.091866072153226e-05,
+ "loss": 0.0415,
+ "step": 8720
+ },
+ {
+ "epoch": 19.49,
+ "learning_rate": 1.090066183110863e-05,
+ "loss": 0.0452,
+ "step": 8730
+ },
+ {
+ "epoch": 19.51,
+ "learning_rate": 1.0882659998554759e-05,
+ "loss": 0.0438,
+ "step": 8740
+ },
+ {
+ "epoch": 19.53,
+ "learning_rate": 1.0864655282675997e-05,
+ "loss": 0.0401,
+ "step": 8750
+ },
+ {
+ "epoch": 19.55,
+ "learning_rate": 1.0846647742287116e-05,
+ "loss": 0.0457,
+ "step": 8760
+ },
+ {
+ "epoch": 19.58,
+ "learning_rate": 1.0828637436212111e-05,
+ "loss": 0.0429,
+ "step": 8770
+ },
+ {
+ "epoch": 19.6,
+ "learning_rate": 1.0810624423284012e-05,
+ "loss": 0.0528,
+ "step": 8780
+ },
+ {
+ "epoch": 19.62,
+ "learning_rate": 1.07926087623447e-05,
+ "loss": 0.0554,
+ "step": 8790
+ },
+ {
+ "epoch": 19.64,
+ "learning_rate": 1.0774590512244694e-05,
+ "loss": 0.0639,
+ "step": 8800
+ },
+ {
+ "epoch": 19.67,
+ "learning_rate": 1.0756569731842978e-05,
+ "loss": 0.0418,
+ "step": 8810
+ },
+ {
+ "epoch": 19.69,
+ "learning_rate": 1.07385464800068e-05,
+ "loss": 0.0422,
+ "step": 8820
+ },
+ {
+ "epoch": 19.71,
+ "learning_rate": 1.0720520815611476e-05,
+ "loss": 0.0413,
+ "step": 8830
+ },
+ {
+ "epoch": 19.73,
+ "learning_rate": 1.0702492797540214e-05,
+ "loss": 0.0504,
+ "step": 8840
+ },
+ {
+ "epoch": 19.75,
+ "learning_rate": 1.06844624846839e-05,
+ "loss": 0.0407,
+ "step": 8850
+ },
+ {
+ "epoch": 19.78,
+ "learning_rate": 1.0666429935940925e-05,
+ "loss": 0.0462,
+ "step": 8860
+ },
+ {
+ "epoch": 19.8,
+ "learning_rate": 1.0648395210216975e-05,
+ "loss": 0.0525,
+ "step": 8870
+ },
+ {
+ "epoch": 19.82,
+ "learning_rate": 1.0630358366424856e-05,
+ "loss": 0.0482,
+ "step": 8880
+ },
+ {
+ "epoch": 19.84,
+ "learning_rate": 1.0612319463484286e-05,
+ "loss": 0.0465,
+ "step": 8890
+ },
+ {
+ "epoch": 19.87,
+ "learning_rate": 1.0594278560321713e-05,
+ "loss": 0.0651,
+ "step": 8900
+ },
+ {
+ "epoch": 19.89,
+ "learning_rate": 1.0576235715870119e-05,
+ "loss": 0.0517,
+ "step": 8910
+ },
+ {
+ "epoch": 19.91,
+ "learning_rate": 1.0558190989068822e-05,
+ "loss": 0.0474,
+ "step": 8920
+ },
+ {
+ "epoch": 19.93,
+ "learning_rate": 1.0540144438863302e-05,
+ "loss": 0.0551,
+ "step": 8930
+ },
+ {
+ "epoch": 19.96,
+ "learning_rate": 1.052209612420498e-05,
+ "loss": 0.0493,
+ "step": 8940
+ },
+ {
+ "epoch": 19.98,
+ "learning_rate": 1.050404610405105e-05,
+ "loss": 0.0529,
+ "step": 8950
+ },
+ {
+ "epoch": 20.0,
+ "learning_rate": 1.0485994437364278e-05,
+ "loss": 0.0474,
+ "step": 8960
+ },
+ {
+ "epoch": 20.02,
+ "learning_rate": 1.0467941183112801e-05,
+ "loss": 0.0373,
+ "step": 8970
+ },
+ {
+ "epoch": 20.04,
+ "learning_rate": 1.0449886400269952e-05,
+ "loss": 0.032,
+ "step": 8980
+ },
+ {
+ "epoch": 20.07,
+ "learning_rate": 1.0431830147814049e-05,
+ "loss": 0.033,
+ "step": 8990
+ },
+ {
+ "epoch": 20.09,
+ "learning_rate": 1.0413772484728211e-05,
+ "loss": 0.0379,
+ "step": 9000
+ },
+ {
+ "epoch": 20.11,
+ "learning_rate": 1.0395713470000173e-05,
+ "loss": 0.0394,
+ "step": 9010
+ },
+ {
+ "epoch": 20.13,
+ "learning_rate": 1.0377653162622076e-05,
+ "loss": 0.0396,
+ "step": 9020
+ },
+ {
+ "epoch": 20.16,
+ "learning_rate": 1.0359591621590292e-05,
+ "loss": 0.0401,
+ "step": 9030
+ },
+ {
+ "epoch": 20.18,
+ "learning_rate": 1.034152890590521e-05,
+ "loss": 0.036,
+ "step": 9040
+ },
+ {
+ "epoch": 20.2,
+ "learning_rate": 1.0323465074571078e-05,
+ "loss": 0.0426,
+ "step": 9050
+ },
+ {
+ "epoch": 20.22,
+ "learning_rate": 1.0305400186595764e-05,
+ "loss": 0.0414,
+ "step": 9060
+ },
+ {
+ "epoch": 20.25,
+ "learning_rate": 1.0287334300990602e-05,
+ "loss": 0.0409,
+ "step": 9070
+ },
+ {
+ "epoch": 20.27,
+ "learning_rate": 1.026926747677018e-05,
+ "loss": 0.0415,
+ "step": 9080
+ },
+ {
+ "epoch": 20.29,
+ "learning_rate": 1.025119977295216e-05,
+ "loss": 0.0424,
+ "step": 9090
+ },
+ {
+ "epoch": 20.31,
+ "learning_rate": 1.0233131248557067e-05,
+ "loss": 0.0339,
+ "step": 9100
+ },
+ {
+ "epoch": 20.33,
+ "learning_rate": 1.0215061962608111e-05,
+ "loss": 0.0533,
+ "step": 9110
+ },
+ {
+ "epoch": 20.36,
+ "learning_rate": 1.0196991974130986e-05,
+ "loss": 0.0475,
+ "step": 9120
+ },
+ {
+ "epoch": 20.38,
+ "learning_rate": 1.017892134215369e-05,
+ "loss": 0.0421,
+ "step": 9130
+ },
+ {
+ "epoch": 20.4,
+ "learning_rate": 1.0160850125706314e-05,
+ "loss": 0.0398,
+ "step": 9140
+ },
+ {
+ "epoch": 20.42,
+ "learning_rate": 1.0142778383820861e-05,
+ "loss": 0.0401,
+ "step": 9150
+ },
+ {
+ "epoch": 20.45,
+ "learning_rate": 1.0124706175531054e-05,
+ "loss": 0.0384,
+ "step": 9160
+ },
+ {
+ "epoch": 20.47,
+ "learning_rate": 1.0106633559872135e-05,
+ "loss": 0.0378,
+ "step": 9170
+ },
+ {
+ "epoch": 20.49,
+ "learning_rate": 1.0088560595880676e-05,
+ "loss": 0.0411,
+ "step": 9180
+ },
+ {
+ "epoch": 20.51,
+ "learning_rate": 1.0070487342594392e-05,
+ "loss": 0.0373,
+ "step": 9190
+ },
+ {
+ "epoch": 20.54,
+ "learning_rate": 1.005241385905194e-05,
+ "loss": 0.0416,
+ "step": 9200
+ },
+ {
+ "epoch": 20.56,
+ "learning_rate": 1.0034340204292728e-05,
+ "loss": 0.0448,
+ "step": 9210
+ },
+ {
+ "epoch": 20.58,
+ "learning_rate": 1.0016266437356727e-05,
+ "loss": 0.0369,
+ "step": 9220
+ },
+ {
+ "epoch": 20.6,
+ "learning_rate": 9.998192617284271e-06,
+ "loss": 0.0471,
+ "step": 9230
+ },
+ {
+ "epoch": 20.62,
+ "learning_rate": 9.980118803115867e-06,
+ "loss": 0.0492,
+ "step": 9240
+ },
+ {
+ "epoch": 20.65,
+ "learning_rate": 9.962045053892004e-06,
+ "loss": 0.0498,
+ "step": 9250
+ },
+ {
+ "epoch": 20.67,
+ "learning_rate": 9.94397142865296e-06,
+ "loss": 0.0477,
+ "step": 9260
+ },
+ {
+ "epoch": 20.69,
+ "learning_rate": 9.925897986438613e-06,
+ "loss": 0.0397,
+ "step": 9270
+ },
+ {
+ "epoch": 20.71,
+ "learning_rate": 9.907824786288226e-06,
+ "loss": 0.0556,
+ "step": 9280
+ },
+ {
+ "epoch": 20.74,
+ "learning_rate": 9.889751887240296e-06,
+ "loss": 0.043,
+ "step": 9290
+ },
+ {
+ "epoch": 20.76,
+ "learning_rate": 9.87167934833231e-06,
+ "loss": 0.0459,
+ "step": 9300
+ },
+ {
+ "epoch": 20.78,
+ "learning_rate": 9.853607228600602e-06,
+ "loss": 0.0442,
+ "step": 9310
+ },
+ {
+ "epoch": 20.8,
+ "learning_rate": 9.835535587080118e-06,
+ "loss": 0.0457,
+ "step": 9320
+ },
+ {
+ "epoch": 20.83,
+ "learning_rate": 9.817464482804257e-06,
+ "loss": 0.045,
+ "step": 9330
+ },
+ {
+ "epoch": 20.85,
+ "learning_rate": 9.799393974804651e-06,
+ "loss": 0.047,
+ "step": 9340
+ },
+ {
+ "epoch": 20.87,
+ "learning_rate": 9.781324122110993e-06,
+ "loss": 0.0425,
+ "step": 9350
+ },
+ {
+ "epoch": 20.89,
+ "learning_rate": 9.763254983750829e-06,
+ "loss": 0.047,
+ "step": 9360
+ },
+ {
+ "epoch": 20.92,
+ "learning_rate": 9.745186618749373e-06,
+ "loss": 0.0526,
+ "step": 9370
+ },
+ {
+ "epoch": 20.94,
+ "learning_rate": 9.727119086129321e-06,
+ "loss": 0.0396,
+ "step": 9380
+ },
+ {
+ "epoch": 20.96,
+ "learning_rate": 9.709052444910636e-06,
+ "loss": 0.0483,
+ "step": 9390
+ },
+ {
+ "epoch": 20.98,
+ "learning_rate": 9.690986754110378e-06,
+ "loss": 0.042,
+ "step": 9400
+ },
+ {
+ "epoch": 21.0,
+ "learning_rate": 9.6729220727425e-06,
+ "loss": 0.0428,
+ "step": 9410
+ },
+ {
+ "epoch": 21.03,
+ "learning_rate": 9.654858459817663e-06,
+ "loss": 0.0497,
+ "step": 9420
+ },
+ {
+ "epoch": 21.05,
+ "learning_rate": 9.636795974343023e-06,
+ "loss": 0.0413,
+ "step": 9430
+ },
+ {
+ "epoch": 21.07,
+ "learning_rate": 9.61873467532207e-06,
+ "loss": 0.0302,
+ "step": 9440
+ },
+ {
+ "epoch": 21.09,
+ "learning_rate": 9.600674621754406e-06,
+ "loss": 0.0392,
+ "step": 9450
+ },
+ {
+ "epoch": 21.12,
+ "learning_rate": 9.582615872635578e-06,
+ "loss": 0.0378,
+ "step": 9460
+ },
+ {
+ "epoch": 21.14,
+ "learning_rate": 9.564558486956853e-06,
+ "loss": 0.043,
+ "step": 9470
+ },
+ {
+ "epoch": 21.16,
+ "learning_rate": 9.546502523705057e-06,
+ "loss": 0.038,
+ "step": 9480
+ },
+ {
+ "epoch": 21.18,
+ "learning_rate": 9.528448041862375e-06,
+ "loss": 0.0377,
+ "step": 9490
+ },
+ {
+ "epoch": 21.21,
+ "learning_rate": 9.510395100406136e-06,
+ "loss": 0.0493,
+ "step": 9500
+ },
+ {
+ "epoch": 21.23,
+ "learning_rate": 9.492343758308651e-06,
+ "loss": 0.0415,
+ "step": 9510
+ },
+ {
+ "epoch": 21.25,
+ "learning_rate": 9.474294074536996e-06,
+ "loss": 0.0471,
+ "step": 9520
+ },
+ {
+ "epoch": 21.27,
+ "learning_rate": 9.456246108052844e-06,
+ "loss": 0.0391,
+ "step": 9530
+ },
+ {
+ "epoch": 21.29,
+ "learning_rate": 9.438199917812241e-06,
+ "loss": 0.0317,
+ "step": 9540
+ },
+ {
+ "epoch": 21.32,
+ "learning_rate": 9.420155562765443e-06,
+ "loss": 0.0481,
+ "step": 9550
+ },
+ {
+ "epoch": 21.34,
+ "learning_rate": 9.402113101856705e-06,
+ "loss": 0.0521,
+ "step": 9560
+ },
+ {
+ "epoch": 21.36,
+ "learning_rate": 9.384072594024103e-06,
+ "loss": 0.0405,
+ "step": 9570
+ },
+ {
+ "epoch": 21.38,
+ "learning_rate": 9.366034098199317e-06,
+ "loss": 0.0359,
+ "step": 9580
+ },
+ {
+ "epoch": 21.41,
+ "learning_rate": 9.347997673307473e-06,
+ "loss": 0.0484,
+ "step": 9590
+ },
+ {
+ "epoch": 21.43,
+ "learning_rate": 9.329963378266919e-06,
+ "loss": 0.0439,
+ "step": 9600
+ },
+ {
+ "epoch": 21.45,
+ "learning_rate": 9.31193127198905e-06,
+ "loss": 0.0392,
+ "step": 9610
+ },
+ {
+ "epoch": 21.47,
+ "learning_rate": 9.293901413378116e-06,
+ "loss": 0.0426,
+ "step": 9620
+ },
+ {
+ "epoch": 21.5,
+ "learning_rate": 9.275873861331012e-06,
+ "loss": 0.044,
+ "step": 9630
+ },
+ {
+ "epoch": 21.52,
+ "learning_rate": 9.257848674737112e-06,
+ "loss": 0.0385,
+ "step": 9640
+ },
+ {
+ "epoch": 21.54,
+ "learning_rate": 9.239825912478054e-06,
+ "loss": 0.0375,
+ "step": 9650
+ },
+ {
+ "epoch": 21.56,
+ "learning_rate": 9.221805633427564e-06,
+ "loss": 0.0381,
+ "step": 9660
+ },
+ {
+ "epoch": 21.58,
+ "learning_rate": 9.203787896451246e-06,
+ "loss": 0.0356,
+ "step": 9670
+ },
+ {
+ "epoch": 21.61,
+ "learning_rate": 9.185772760406408e-06,
+ "loss": 0.0375,
+ "step": 9680
+ },
+ {
+ "epoch": 21.63,
+ "learning_rate": 9.167760284141859e-06,
+ "loss": 0.0341,
+ "step": 9690
+ },
+ {
+ "epoch": 21.65,
+ "learning_rate": 9.149750526497725e-06,
+ "loss": 0.047,
+ "step": 9700
+ },
+ {
+ "epoch": 21.67,
+ "learning_rate": 9.131743546305235e-06,
+ "loss": 0.0417,
+ "step": 9710
+ },
+ {
+ "epoch": 21.7,
+ "learning_rate": 9.113739402386566e-06,
+ "loss": 0.0428,
+ "step": 9720
+ },
+ {
+ "epoch": 21.72,
+ "learning_rate": 9.095738153554624e-06,
+ "loss": 0.0432,
+ "step": 9730
+ },
+ {
+ "epoch": 21.74,
+ "learning_rate": 9.077739858612843e-06,
+ "loss": 0.0467,
+ "step": 9740
+ },
+ {
+ "epoch": 21.76,
+ "learning_rate": 9.059744576355027e-06,
+ "loss": 0.0467,
+ "step": 9750
+ },
+ {
+ "epoch": 21.79,
+ "learning_rate": 9.041752365565125e-06,
+ "loss": 0.0396,
+ "step": 9760
+ },
+ {
+ "epoch": 21.81,
+ "learning_rate": 9.023763285017065e-06,
+ "loss": 0.0365,
+ "step": 9770
+ },
+ {
+ "epoch": 21.83,
+ "learning_rate": 9.005777393474534e-06,
+ "loss": 0.0468,
+ "step": 9780
+ },
+ {
+ "epoch": 21.85,
+ "learning_rate": 8.987794749690819e-06,
+ "loss": 0.0415,
+ "step": 9790
+ },
+ {
+ "epoch": 21.88,
+ "learning_rate": 8.969815412408583e-06,
+ "loss": 0.04,
+ "step": 9800
+ },
+ {
+ "epoch": 21.9,
+ "learning_rate": 8.951839440359701e-06,
+ "loss": 0.0483,
+ "step": 9810
+ },
+ {
+ "epoch": 21.92,
+ "learning_rate": 8.93386689226504e-06,
+ "loss": 0.0421,
+ "step": 9820
+ },
+ {
+ "epoch": 21.94,
+ "learning_rate": 8.915897826834295e-06,
+ "loss": 0.0382,
+ "step": 9830
+ },
+ {
+ "epoch": 21.96,
+ "learning_rate": 8.89793230276578e-06,
+ "loss": 0.051,
+ "step": 9840
+ },
+ {
+ "epoch": 21.99,
+ "learning_rate": 8.879970378746238e-06,
+ "loss": 0.0404,
+ "step": 9850
+ },
+ {
+ "epoch": 22.01,
+ "learning_rate": 8.862012113450662e-06,
+ "loss": 0.0403,
+ "step": 9860
+ },
+ {
+ "epoch": 22.03,
+ "learning_rate": 8.844057565542074e-06,
+ "loss": 0.0302,
+ "step": 9870
+ },
+ {
+ "epoch": 22.05,
+ "learning_rate": 8.826106793671376e-06,
+ "loss": 0.045,
+ "step": 9880
+ },
+ {
+ "epoch": 22.08,
+ "learning_rate": 8.808159856477115e-06,
+ "loss": 0.0349,
+ "step": 9890
+ },
+ {
+ "epoch": 22.1,
+ "learning_rate": 8.790216812585327e-06,
+ "loss": 0.0402,
+ "step": 9900
+ },
+ {
+ "epoch": 22.12,
+ "learning_rate": 8.772277720609312e-06,
+ "loss": 0.0367,
+ "step": 9910
+ },
+ {
+ "epoch": 22.14,
+ "learning_rate": 8.754342639149486e-06,
+ "loss": 0.0359,
+ "step": 9920
+ },
+ {
+ "epoch": 22.17,
+ "learning_rate": 8.736411626793139e-06,
+ "loss": 0.0368,
+ "step": 9930
+ },
+ {
+ "epoch": 22.19,
+ "learning_rate": 8.718484742114285e-06,
+ "loss": 0.0356,
+ "step": 9940
+ },
+ {
+ "epoch": 22.21,
+ "learning_rate": 8.700562043673448e-06,
+ "loss": 0.0402,
+ "step": 9950
+ },
+ {
+ "epoch": 22.23,
+ "learning_rate": 8.682643590017474e-06,
+ "loss": 0.0334,
+ "step": 9960
+ },
+ {
+ "epoch": 22.25,
+ "learning_rate": 8.664729439679354e-06,
+ "loss": 0.0423,
+ "step": 9970
+ },
+ {
+ "epoch": 22.28,
+ "learning_rate": 8.646819651178008e-06,
+ "loss": 0.04,
+ "step": 9980
+ },
+ {
+ "epoch": 22.3,
+ "learning_rate": 8.628914283018119e-06,
+ "loss": 0.0373,
+ "step": 9990
+ },
+ {
+ "epoch": 22.32,
+ "learning_rate": 8.61101339368992e-06,
+ "loss": 0.0422,
+ "step": 10000
+ },
+ {
+ "epoch": 22.34,
+ "learning_rate": 8.593117041669024e-06,
+ "loss": 0.0443,
+ "step": 10010
+ },
+ {
+ "epoch": 22.37,
+ "learning_rate": 8.57522528541621e-06,
+ "loss": 0.0465,
+ "step": 10020
+ },
+ {
+ "epoch": 22.39,
+ "learning_rate": 8.55733818337726e-06,
+ "loss": 0.0408,
+ "step": 10030
+ },
+ {
+ "epoch": 22.41,
+ "learning_rate": 8.539455793982737e-06,
+ "loss": 0.0372,
+ "step": 10040
+ },
+ {
+ "epoch": 22.43,
+ "learning_rate": 8.521578175647823e-06,
+ "loss": 0.0375,
+ "step": 10050
+ },
+ {
+ "epoch": 22.46,
+ "learning_rate": 8.503705386772098e-06,
+ "loss": 0.0377,
+ "step": 10060
+ },
+ {
+ "epoch": 22.48,
+ "learning_rate": 8.485837485739384e-06,
+ "loss": 0.0302,
+ "step": 10070
+ },
+ {
+ "epoch": 22.5,
+ "learning_rate": 8.467974530917524e-06,
+ "loss": 0.0419,
+ "step": 10080
+ },
+ {
+ "epoch": 22.52,
+ "learning_rate": 8.450116580658208e-06,
+ "loss": 0.0366,
+ "step": 10090
+ },
+ {
+ "epoch": 22.54,
+ "learning_rate": 8.432263693296783e-06,
+ "loss": 0.0402,
+ "step": 10100
+ },
+ {
+ "epoch": 22.57,
+ "learning_rate": 8.414415927152042e-06,
+ "loss": 0.0395,
+ "step": 10110
+ },
+ {
+ "epoch": 22.59,
+ "learning_rate": 8.396573340526069e-06,
+ "loss": 0.0405,
+ "step": 10120
+ },
+ {
+ "epoch": 22.61,
+ "learning_rate": 8.37873599170401e-06,
+ "loss": 0.0394,
+ "step": 10130
+ },
+ {
+ "epoch": 22.63,
+ "learning_rate": 8.360903938953914e-06,
+ "loss": 0.0415,
+ "step": 10140
+ },
+ {
+ "epoch": 22.66,
+ "learning_rate": 8.343077240526522e-06,
+ "loss": 0.0483,
+ "step": 10150
+ },
+ {
+ "epoch": 22.68,
+ "learning_rate": 8.325255954655093e-06,
+ "loss": 0.0462,
+ "step": 10160
+ },
+ {
+ "epoch": 22.7,
+ "learning_rate": 8.307440139555192e-06,
+ "loss": 0.0346,
+ "step": 10170
+ },
+ {
+ "epoch": 22.72,
+ "learning_rate": 8.289629853424526e-06,
+ "loss": 0.0436,
+ "step": 10180
+ },
+ {
+ "epoch": 22.75,
+ "learning_rate": 8.271825154442732e-06,
+ "loss": 0.0371,
+ "step": 10190
+ },
+ {
+ "epoch": 22.77,
+ "learning_rate": 8.2540261007712e-06,
+ "loss": 0.0606,
+ "step": 10200
+ },
+ {
+ "epoch": 22.79,
+ "learning_rate": 8.236232750552881e-06,
+ "loss": 0.0432,
+ "step": 10210
+ },
+ {
+ "epoch": 22.81,
+ "learning_rate": 8.218445161912088e-06,
+ "loss": 0.0533,
+ "step": 10220
+ },
+ {
+ "epoch": 22.83,
+ "learning_rate": 8.20066339295432e-06,
+ "loss": 0.051,
+ "step": 10230
+ },
+ {
+ "epoch": 22.86,
+ "learning_rate": 8.182887501766059e-06,
+ "loss": 0.044,
+ "step": 10240
+ },
+ {
+ "epoch": 22.88,
+ "learning_rate": 8.165117546414595e-06,
+ "loss": 0.0346,
+ "step": 10250
+ },
+ {
+ "epoch": 22.9,
+ "learning_rate": 8.147353584947818e-06,
+ "loss": 0.0443,
+ "step": 10260
+ },
+ {
+ "epoch": 22.92,
+ "learning_rate": 8.129595675394045e-06,
+ "loss": 0.0449,
+ "step": 10270
+ },
+ {
+ "epoch": 22.95,
+ "learning_rate": 8.11184387576182e-06,
+ "loss": 0.0381,
+ "step": 10280
+ },
+ {
+ "epoch": 22.97,
+ "learning_rate": 8.094098244039734e-06,
+ "loss": 0.0416,
+ "step": 10290
+ },
+ {
+ "epoch": 22.99,
+ "learning_rate": 8.076358838196216e-06,
+ "loss": 0.0416,
+ "step": 10300
+ },
+ {
+ "epoch": 23.01,
+ "learning_rate": 8.058625716179375e-06,
+ "loss": 0.0363,
+ "step": 10310
+ },
+ {
+ "epoch": 23.04,
+ "learning_rate": 8.04089893591678e-06,
+ "loss": 0.0351,
+ "step": 10320
+ },
+ {
+ "epoch": 23.06,
+ "learning_rate": 8.023178555315291e-06,
+ "loss": 0.0286,
+ "step": 10330
+ },
+ {
+ "epoch": 23.08,
+ "learning_rate": 8.005464632260862e-06,
+ "loss": 0.0358,
+ "step": 10340
+ },
+ {
+ "epoch": 23.1,
+ "learning_rate": 7.987757224618346e-06,
+ "loss": 0.0335,
+ "step": 10350
+ },
+ {
+ "epoch": 23.12,
+ "learning_rate": 7.970056390231323e-06,
+ "loss": 0.0351,
+ "step": 10360
+ },
+ {
+ "epoch": 23.15,
+ "learning_rate": 7.952362186921889e-06,
+ "loss": 0.0324,
+ "step": 10370
+ },
+ {
+ "epoch": 23.17,
+ "learning_rate": 7.934674672490488e-06,
+ "loss": 0.0304,
+ "step": 10380
+ },
+ {
+ "epoch": 23.19,
+ "learning_rate": 7.916993904715708e-06,
+ "loss": 0.0406,
+ "step": 10390
+ },
+ {
+ "epoch": 23.21,
+ "learning_rate": 7.899319941354107e-06,
+ "loss": 0.0298,
+ "step": 10400
+ },
+ {
+ "epoch": 23.24,
+ "learning_rate": 7.881652840140001e-06,
+ "loss": 0.0359,
+ "step": 10410
+ },
+ {
+ "epoch": 23.26,
+ "learning_rate": 7.863992658785302e-06,
+ "loss": 0.0431,
+ "step": 10420
+ },
+ {
+ "epoch": 23.28,
+ "learning_rate": 7.846339454979312e-06,
+ "loss": 0.0265,
+ "step": 10430
+ },
+ {
+ "epoch": 23.3,
+ "learning_rate": 7.828693286388542e-06,
+ "loss": 0.0373,
+ "step": 10440
+ },
+ {
+ "epoch": 23.33,
+ "learning_rate": 7.811054210656526e-06,
+ "loss": 0.0345,
+ "step": 10450
+ },
+ {
+ "epoch": 23.35,
+ "learning_rate": 7.793422285403614e-06,
+ "loss": 0.0254,
+ "step": 10460
+ },
+ {
+ "epoch": 23.37,
+ "learning_rate": 7.775797568226816e-06,
+ "loss": 0.0354,
+ "step": 10470
+ },
+ {
+ "epoch": 23.39,
+ "learning_rate": 7.758180116699578e-06,
+ "loss": 0.0399,
+ "step": 10480
+ },
+ {
+ "epoch": 23.42,
+ "learning_rate": 7.74056998837163e-06,
+ "loss": 0.0427,
+ "step": 10490
+ },
+ {
+ "epoch": 23.44,
+ "learning_rate": 7.722967240768761e-06,
+ "loss": 0.041,
+ "step": 10500
+ },
+ {
+ "epoch": 23.46,
+ "learning_rate": 7.705371931392668e-06,
+ "loss": 0.0354,
+ "step": 10510
+ },
+ {
+ "epoch": 23.48,
+ "learning_rate": 7.687784117720736e-06,
+ "loss": 0.0403,
+ "step": 10520
+ },
+ {
+ "epoch": 23.5,
+ "learning_rate": 7.670203857205877e-06,
+ "loss": 0.0343,
+ "step": 10530
+ },
+ {
+ "epoch": 23.53,
+ "learning_rate": 7.652631207276311e-06,
+ "loss": 0.0367,
+ "step": 10540
+ },
+ {
+ "epoch": 23.55,
+ "learning_rate": 7.635066225335417e-06,
+ "loss": 0.0395,
+ "step": 10550
+ },
+ {
+ "epoch": 23.57,
+ "learning_rate": 7.617508968761519e-06,
+ "loss": 0.0335,
+ "step": 10560
+ },
+ {
+ "epoch": 23.59,
+ "learning_rate": 7.599959494907695e-06,
+ "loss": 0.0444,
+ "step": 10570
+ },
+ {
+ "epoch": 23.62,
+ "learning_rate": 7.582417861101614e-06,
+ "loss": 0.0348,
+ "step": 10580
+ },
+ {
+ "epoch": 23.64,
+ "learning_rate": 7.564884124645325e-06,
+ "loss": 0.0341,
+ "step": 10590
+ },
+ {
+ "epoch": 23.66,
+ "learning_rate": 7.547358342815089e-06,
+ "loss": 0.0422,
+ "step": 10600
+ },
+ {
+ "epoch": 23.68,
+ "learning_rate": 7.5298405728611645e-06,
+ "loss": 0.0344,
+ "step": 10610
+ },
+ {
+ "epoch": 23.71,
+ "learning_rate": 7.512330872007659e-06,
+ "loss": 0.0412,
+ "step": 10620
+ },
+ {
+ "epoch": 23.73,
+ "learning_rate": 7.494829297452306e-06,
+ "loss": 0.0455,
+ "step": 10630
+ },
+ {
+ "epoch": 23.75,
+ "learning_rate": 7.4773359063663045e-06,
+ "loss": 0.0448,
+ "step": 10640
+ },
+ {
+ "epoch": 23.77,
+ "learning_rate": 7.459850755894108e-06,
+ "loss": 0.0504,
+ "step": 10650
+ },
+ {
+ "epoch": 23.79,
+ "learning_rate": 7.442373903153266e-06,
+ "loss": 0.0466,
+ "step": 10660
+ },
+ {
+ "epoch": 23.82,
+ "learning_rate": 7.424905405234209e-06,
+ "loss": 0.0388,
+ "step": 10670
+ },
+ {
+ "epoch": 23.84,
+ "learning_rate": 7.407445319200083e-06,
+ "loss": 0.0395,
+ "step": 10680
+ },
+ {
+ "epoch": 23.86,
+ "learning_rate": 7.38999370208656e-06,
+ "loss": 0.0421,
+ "step": 10690
+ },
+ {
+ "epoch": 23.88,
+ "learning_rate": 7.37255061090163e-06,
+ "loss": 0.0367,
+ "step": 10700
+ },
+ {
+ "epoch": 23.91,
+ "learning_rate": 7.355116102625451e-06,
+ "loss": 0.0387,
+ "step": 10710
+ },
+ {
+ "epoch": 23.93,
+ "learning_rate": 7.337690234210132e-06,
+ "loss": 0.0308,
+ "step": 10720
+ },
+ {
+ "epoch": 23.95,
+ "learning_rate": 7.320273062579568e-06,
+ "loss": 0.032,
+ "step": 10730
+ },
+ {
+ "epoch": 23.97,
+ "learning_rate": 7.3028646446292295e-06,
+ "loss": 0.0475,
+ "step": 10740
+ },
+ {
+ "epoch": 24.0,
+ "learning_rate": 7.28546503722601e-06,
+ "loss": 0.039,
+ "step": 10750
+ },
+ {
+ "epoch": 24.02,
+ "learning_rate": 7.268074297208008e-06,
+ "loss": 0.0334,
+ "step": 10760
+ },
+ {
+ "epoch": 24.04,
+ "learning_rate": 7.250692481384366e-06,
+ "loss": 0.0297,
+ "step": 10770
+ },
+ {
+ "epoch": 24.06,
+ "learning_rate": 7.233319646535067e-06,
+ "loss": 0.0311,
+ "step": 10780
+ },
+ {
+ "epoch": 24.08,
+ "learning_rate": 7.21595584941076e-06,
+ "loss": 0.0267,
+ "step": 10790
+ },
+ {
+ "epoch": 24.11,
+ "learning_rate": 7.198601146732573e-06,
+ "loss": 0.0323,
+ "step": 10800
+ },
+ {
+ "epoch": 24.13,
+ "learning_rate": 7.181255595191919e-06,
+ "loss": 0.0391,
+ "step": 10810
+ },
+ {
+ "epoch": 24.15,
+ "learning_rate": 7.1639192514503265e-06,
+ "loss": 0.0284,
+ "step": 10820
+ },
+ {
+ "epoch": 24.17,
+ "learning_rate": 7.146592172139234e-06,
+ "loss": 0.036,
+ "step": 10830
+ },
+ {
+ "epoch": 24.2,
+ "learning_rate": 7.129274413859832e-06,
+ "loss": 0.0376,
+ "step": 10840
+ },
+ {
+ "epoch": 24.22,
+ "learning_rate": 7.111966033182845e-06,
+ "loss": 0.0391,
+ "step": 10850
+ },
+ {
+ "epoch": 24.24,
+ "learning_rate": 7.094667086648381e-06,
+ "loss": 0.0337,
+ "step": 10860
+ },
+ {
+ "epoch": 24.26,
+ "learning_rate": 7.077377630765716e-06,
+ "loss": 0.0278,
+ "step": 10870
+ },
+ {
+ "epoch": 24.29,
+ "learning_rate": 7.060097722013137e-06,
+ "loss": 0.0392,
+ "step": 10880
+ },
+ {
+ "epoch": 24.31,
+ "learning_rate": 7.042827416837728e-06,
+ "loss": 0.032,
+ "step": 10890
+ },
+ {
+ "epoch": 24.33,
+ "learning_rate": 7.025566771655219e-06,
+ "loss": 0.0306,
+ "step": 10900
+ },
+ {
+ "epoch": 24.35,
+ "learning_rate": 7.00831584284977e-06,
+ "loss": 0.0293,
+ "step": 10910
+ },
+ {
+ "epoch": 24.38,
+ "learning_rate": 6.991074686773809e-06,
+ "loss": 0.03,
+ "step": 10920
+ },
+ {
+ "epoch": 24.4,
+ "learning_rate": 6.973843359747845e-06,
+ "loss": 0.0322,
+ "step": 10930
+ },
+ {
+ "epoch": 24.42,
+ "learning_rate": 6.95662191806026e-06,
+ "loss": 0.0331,
+ "step": 10940
+ },
+ {
+ "epoch": 24.44,
+ "learning_rate": 6.939410417967168e-06,
+ "loss": 0.0354,
+ "step": 10950
+ },
+ {
+ "epoch": 24.46,
+ "learning_rate": 6.922208915692186e-06,
+ "loss": 0.0333,
+ "step": 10960
+ },
+ {
+ "epoch": 24.49,
+ "learning_rate": 6.905017467426291e-06,
+ "loss": 0.0343,
+ "step": 10970
+ },
+ {
+ "epoch": 24.51,
+ "learning_rate": 6.887836129327602e-06,
+ "loss": 0.0336,
+ "step": 10980
+ },
+ {
+ "epoch": 24.53,
+ "learning_rate": 6.870664957521225e-06,
+ "loss": 0.036,
+ "step": 10990
+ },
+ {
+ "epoch": 24.55,
+ "learning_rate": 6.85350400809904e-06,
+ "loss": 0.0358,
+ "step": 11000
+ },
+ {
+ "epoch": 24.58,
+ "learning_rate": 6.83635333711955e-06,
+ "loss": 0.0406,
+ "step": 11010
+ },
+ {
+ "epoch": 24.6,
+ "learning_rate": 6.819213000607674e-06,
+ "loss": 0.0319,
+ "step": 11020
+ },
+ {
+ "epoch": 24.62,
+ "learning_rate": 6.802083054554572e-06,
+ "loss": 0.0305,
+ "step": 11030
+ },
+ {
+ "epoch": 24.64,
+ "learning_rate": 6.784963554917472e-06,
+ "loss": 0.0366,
+ "step": 11040
+ },
+ {
+ "epoch": 24.67,
+ "learning_rate": 6.76785455761946e-06,
+ "loss": 0.0455,
+ "step": 11050
+ },
+ {
+ "epoch": 24.69,
+ "learning_rate": 6.75075611854933e-06,
+ "loss": 0.0358,
+ "step": 11060
+ },
+ {
+ "epoch": 24.71,
+ "learning_rate": 6.733668293561376e-06,
+ "loss": 0.0421,
+ "step": 11070
+ },
+ {
+ "epoch": 24.73,
+ "learning_rate": 6.716591138475231e-06,
+ "loss": 0.0305,
+ "step": 11080
+ },
+ {
+ "epoch": 24.75,
+ "learning_rate": 6.699524709075658e-06,
+ "loss": 0.0382,
+ "step": 11090
+ },
+ {
+ "epoch": 24.78,
+ "learning_rate": 6.6824690611124e-06,
+ "loss": 0.0303,
+ "step": 11100
+ },
+ {
+ "epoch": 24.8,
+ "learning_rate": 6.665424250299963e-06,
+ "loss": 0.0375,
+ "step": 11110
+ },
+ {
+ "epoch": 24.82,
+ "learning_rate": 6.648390332317474e-06,
+ "loss": 0.0416,
+ "step": 11120
+ },
+ {
+ "epoch": 24.84,
+ "learning_rate": 6.631367362808453e-06,
+ "loss": 0.0356,
+ "step": 11130
+ },
+ {
+ "epoch": 24.87,
+ "learning_rate": 6.614355397380674e-06,
+ "loss": 0.0375,
+ "step": 11140
+ },
+ {
+ "epoch": 24.89,
+ "learning_rate": 6.597354491605954e-06,
+ "loss": 0.0381,
+ "step": 11150
+ },
+ {
+ "epoch": 24.91,
+ "learning_rate": 6.580364701019989e-06,
+ "loss": 0.0324,
+ "step": 11160
+ },
+ {
+ "epoch": 24.93,
+ "learning_rate": 6.563386081122166e-06,
+ "loss": 0.0394,
+ "step": 11170
+ },
+ {
+ "epoch": 24.96,
+ "learning_rate": 6.546418687375368e-06,
+ "loss": 0.0351,
+ "step": 11180
+ },
+ {
+ "epoch": 24.98,
+ "learning_rate": 6.529462575205829e-06,
+ "loss": 0.0346,
+ "step": 11190
+ },
+ {
+ "epoch": 25.0,
+ "learning_rate": 6.512517800002909e-06,
+ "loss": 0.0362,
+ "step": 11200
+ },
+ {
+ "epoch": 25.02,
+ "learning_rate": 6.495584417118948e-06,
+ "loss": 0.0323,
+ "step": 11210
+ },
+ {
+ "epoch": 25.04,
+ "learning_rate": 6.47866248186906e-06,
+ "loss": 0.0262,
+ "step": 11220
+ },
+ {
+ "epoch": 25.07,
+ "learning_rate": 6.461752049530983e-06,
+ "loss": 0.034,
+ "step": 11230
+ },
+ {
+ "epoch": 25.09,
+ "learning_rate": 6.444853175344854e-06,
+ "loss": 0.0272,
+ "step": 11240
+ },
+ {
+ "epoch": 25.11,
+ "learning_rate": 6.427965914513072e-06,
+ "loss": 0.0281,
+ "step": 11250
+ },
+ {
+ "epoch": 25.13,
+ "learning_rate": 6.411090322200094e-06,
+ "loss": 0.0329,
+ "step": 11260
+ },
+ {
+ "epoch": 25.16,
+ "learning_rate": 6.394226453532259e-06,
+ "loss": 0.0316,
+ "step": 11270
+ },
+ {
+ "epoch": 25.18,
+ "learning_rate": 6.377374363597615e-06,
+ "loss": 0.0358,
+ "step": 11280
+ },
+ {
+ "epoch": 25.2,
+ "learning_rate": 6.360534107445722e-06,
+ "loss": 0.0363,
+ "step": 11290
+ },
+ {
+ "epoch": 25.22,
+ "learning_rate": 6.343705740087493e-06,
+ "loss": 0.0243,
+ "step": 11300
+ },
+ {
+ "epoch": 25.25,
+ "learning_rate": 6.326889316494999e-06,
+ "loss": 0.0347,
+ "step": 11310
+ },
+ {
+ "epoch": 25.27,
+ "learning_rate": 6.310084891601307e-06,
+ "loss": 0.0255,
+ "step": 11320
+ },
+ {
+ "epoch": 25.29,
+ "learning_rate": 6.293292520300267e-06,
+ "loss": 0.0329,
+ "step": 11330
+ },
+ {
+ "epoch": 25.31,
+ "learning_rate": 6.276512257446374e-06,
+ "loss": 0.0297,
+ "step": 11340
+ },
+ {
+ "epoch": 25.33,
+ "learning_rate": 6.259744157854559e-06,
+ "loss": 0.0316,
+ "step": 11350
+ },
+ {
+ "epoch": 25.36,
+ "learning_rate": 6.24298827630003e-06,
+ "loss": 0.0369,
+ "step": 11360
+ },
+ {
+ "epoch": 25.38,
+ "learning_rate": 6.226244667518064e-06,
+ "loss": 0.026,
+ "step": 11370
+ },
+ {
+ "epoch": 25.4,
+ "learning_rate": 6.209513386203871e-06,
+ "loss": 0.0304,
+ "step": 11380
+ },
+ {
+ "epoch": 25.42,
+ "learning_rate": 6.1927944870123746e-06,
+ "loss": 0.0283,
+ "step": 11390
+ },
+ {
+ "epoch": 25.45,
+ "learning_rate": 6.176088024558056e-06,
+ "loss": 0.0279,
+ "step": 11400
+ },
+ {
+ "epoch": 25.47,
+ "learning_rate": 6.159394053414775e-06,
+ "loss": 0.0254,
+ "step": 11410
+ },
+ {
+ "epoch": 25.49,
+ "learning_rate": 6.142712628115577e-06,
+ "loss": 0.0303,
+ "step": 11420
+ },
+ {
+ "epoch": 25.51,
+ "learning_rate": 6.126043803152537e-06,
+ "loss": 0.0288,
+ "step": 11430
+ },
+ {
+ "epoch": 25.54,
+ "learning_rate": 6.109387632976556e-06,
+ "loss": 0.026,
+ "step": 11440
+ },
+ {
+ "epoch": 25.56,
+ "learning_rate": 6.092744171997205e-06,
+ "loss": 0.0304,
+ "step": 11450
+ },
+ {
+ "epoch": 25.58,
+ "learning_rate": 6.076113474582535e-06,
+ "loss": 0.0272,
+ "step": 11460
+ },
+ {
+ "epoch": 25.6,
+ "learning_rate": 6.059495595058911e-06,
+ "loss": 0.0369,
+ "step": 11470
+ },
+ {
+ "epoch": 25.62,
+ "learning_rate": 6.042890587710812e-06,
+ "loss": 0.0355,
+ "step": 11480
+ },
+ {
+ "epoch": 25.65,
+ "learning_rate": 6.026298506780684e-06,
+ "loss": 0.032,
+ "step": 11490
+ },
+ {
+ "epoch": 25.67,
+ "learning_rate": 6.009719406468735e-06,
+ "loss": 0.0296,
+ "step": 11500
+ },
+ {
+ "epoch": 25.69,
+ "learning_rate": 5.993153340932776e-06,
+ "loss": 0.0354,
+ "step": 11510
+ },
+ {
+ "epoch": 25.71,
+ "learning_rate": 5.9766003642880434e-06,
+ "loss": 0.0356,
+ "step": 11520
+ },
+ {
+ "epoch": 25.74,
+ "learning_rate": 5.960060530607001e-06,
+ "loss": 0.0308,
+ "step": 11530
+ },
+ {
+ "epoch": 25.76,
+ "learning_rate": 5.9435338939191976e-06,
+ "loss": 0.035,
+ "step": 11540
+ },
+ {
+ "epoch": 25.78,
+ "learning_rate": 5.927020508211059e-06,
+ "loss": 0.0301,
+ "step": 11550
+ },
+ {
+ "epoch": 25.8,
+ "learning_rate": 5.910520427425734e-06,
+ "loss": 0.0309,
+ "step": 11560
+ },
+ {
+ "epoch": 25.83,
+ "learning_rate": 5.8940337054629005e-06,
+ "loss": 0.038,
+ "step": 11570
+ },
+ {
+ "epoch": 25.85,
+ "learning_rate": 5.8775603961786055e-06,
+ "loss": 0.0264,
+ "step": 11580
+ },
+ {
+ "epoch": 25.87,
+ "learning_rate": 5.861100553385076e-06,
+ "loss": 0.0411,
+ "step": 11590
+ },
+ {
+ "epoch": 25.89,
+ "learning_rate": 5.8446542308505595e-06,
+ "loss": 0.034,
+ "step": 11600
+ },
+ {
+ "epoch": 25.92,
+ "learning_rate": 5.828221482299119e-06,
+ "loss": 0.0286,
+ "step": 11610
+ },
+ {
+ "epoch": 25.94,
+ "learning_rate": 5.811802361410492e-06,
+ "loss": 0.0353,
+ "step": 11620
+ },
+ {
+ "epoch": 25.96,
+ "learning_rate": 5.795396921819898e-06,
+ "loss": 0.0355,
+ "step": 11630
+ },
+ {
+ "epoch": 25.98,
+ "learning_rate": 5.779005217117858e-06,
+ "loss": 0.0249,
+ "step": 11640
+ },
+ {
+ "epoch": 26.0,
+ "learning_rate": 5.762627300850034e-06,
+ "loss": 0.0313,
+ "step": 11650
+ },
+ {
+ "epoch": 26.03,
+ "learning_rate": 5.746263226517037e-06,
+ "loss": 0.0232,
+ "step": 11660
+ },
+ {
+ "epoch": 26.05,
+ "learning_rate": 5.729913047574272e-06,
+ "loss": 0.0268,
+ "step": 11670
+ },
+ {
+ "epoch": 26.07,
+ "learning_rate": 5.7135768174317385e-06,
+ "loss": 0.0288,
+ "step": 11680
+ },
+ {
+ "epoch": 26.09,
+ "learning_rate": 5.6972545894538885e-06,
+ "loss": 0.0228,
+ "step": 11690
+ },
+ {
+ "epoch": 26.12,
+ "learning_rate": 5.680946416959417e-06,
+ "loss": 0.0308,
+ "step": 11700
+ },
+ {
+ "epoch": 26.14,
+ "learning_rate": 5.664652353221118e-06,
+ "loss": 0.0234,
+ "step": 11710
+ },
+ {
+ "epoch": 26.16,
+ "learning_rate": 5.64837245146569e-06,
+ "loss": 0.0221,
+ "step": 11720
+ },
+ {
+ "epoch": 26.18,
+ "learning_rate": 5.6321067648735775e-06,
+ "loss": 0.0258,
+ "step": 11730
+ },
+ {
+ "epoch": 26.21,
+ "learning_rate": 5.615855346578774e-06,
+ "loss": 0.0308,
+ "step": 11740
+ },
+ {
+ "epoch": 26.23,
+ "learning_rate": 5.599618249668677e-06,
+ "loss": 0.022,
+ "step": 11750
+ },
+ {
+ "epoch": 26.25,
+ "learning_rate": 5.583395527183904e-06,
+ "loss": 0.0275,
+ "step": 11760
+ },
+ {
+ "epoch": 26.27,
+ "learning_rate": 5.5671872321180985e-06,
+ "loss": 0.0295,
+ "step": 11770
+ },
+ {
+ "epoch": 26.29,
+ "learning_rate": 5.550993417417798e-06,
+ "loss": 0.0204,
+ "step": 11780
+ },
+ {
+ "epoch": 26.32,
+ "learning_rate": 5.534814135982217e-06,
+ "loss": 0.0318,
+ "step": 11790
+ },
+ {
+ "epoch": 26.34,
+ "learning_rate": 5.518649440663109e-06,
+ "loss": 0.0246,
+ "step": 11800
+ },
+ {
+ "epoch": 26.36,
+ "learning_rate": 5.502499384264575e-06,
+ "loss": 0.0256,
+ "step": 11810
+ },
+ {
+ "epoch": 26.38,
+ "learning_rate": 5.486364019542902e-06,
+ "loss": 0.0304,
+ "step": 11820
+ },
+ {
+ "epoch": 26.41,
+ "learning_rate": 5.47024339920637e-06,
+ "loss": 0.0307,
+ "step": 11830
+ },
+ {
+ "epoch": 26.43,
+ "learning_rate": 5.4541375759151145e-06,
+ "loss": 0.0312,
+ "step": 11840
+ },
+ {
+ "epoch": 26.45,
+ "learning_rate": 5.438046602280913e-06,
+ "loss": 0.0293,
+ "step": 11850
+ },
+ {
+ "epoch": 26.47,
+ "learning_rate": 5.421970530867051e-06,
+ "loss": 0.0302,
+ "step": 11860
+ },
+ {
+ "epoch": 26.5,
+ "learning_rate": 5.405909414188131e-06,
+ "loss": 0.0296,
+ "step": 11870
+ },
+ {
+ "epoch": 26.52,
+ "learning_rate": 5.389863304709897e-06,
+ "loss": 0.0288,
+ "step": 11880
+ },
+ {
+ "epoch": 26.54,
+ "learning_rate": 5.373832254849081e-06,
+ "loss": 0.0291,
+ "step": 11890
+ },
+ {
+ "epoch": 26.56,
+ "learning_rate": 5.357816316973205e-06,
+ "loss": 0.0225,
+ "step": 11900
+ },
+ {
+ "epoch": 26.58,
+ "learning_rate": 5.341815543400446e-06,
+ "loss": 0.0264,
+ "step": 11910
+ },
+ {
+ "epoch": 26.61,
+ "learning_rate": 5.325829986399423e-06,
+ "loss": 0.0267,
+ "step": 11920
+ },
+ {
+ "epoch": 26.63,
+ "learning_rate": 5.309859698189067e-06,
+ "loss": 0.027,
+ "step": 11930
+ },
+ {
+ "epoch": 26.65,
+ "learning_rate": 5.293904730938417e-06,
+ "loss": 0.0288,
+ "step": 11940
+ },
+ {
+ "epoch": 26.67,
+ "learning_rate": 5.277965136766473e-06,
+ "loss": 0.0303,
+ "step": 11950
+ },
+ {
+ "epoch": 26.7,
+ "learning_rate": 5.262040967742015e-06,
+ "loss": 0.0207,
+ "step": 11960
+ },
+ {
+ "epoch": 26.72,
+ "learning_rate": 5.2461322758834375e-06,
+ "loss": 0.0354,
+ "step": 11970
+ },
+ {
+ "epoch": 26.74,
+ "learning_rate": 5.2302391131585665e-06,
+ "loss": 0.0251,
+ "step": 11980
+ },
+ {
+ "epoch": 26.76,
+ "learning_rate": 5.214361531484509e-06,
+ "loss": 0.0221,
+ "step": 11990
+ },
+ {
+ "epoch": 26.79,
+ "learning_rate": 5.1984995827274775e-06,
+ "loss": 0.0391,
+ "step": 12000
+ },
+ {
+ "epoch": 26.81,
+ "learning_rate": 5.182653318702604e-06,
+ "loss": 0.0326,
+ "step": 12010
+ },
+ {
+ "epoch": 26.83,
+ "learning_rate": 5.166822791173796e-06,
+ "loss": 0.0264,
+ "step": 12020
+ },
+ {
+ "epoch": 26.85,
+ "learning_rate": 5.151008051853554e-06,
+ "loss": 0.0286,
+ "step": 12030
+ },
+ {
+ "epoch": 26.88,
+ "learning_rate": 5.1352091524028045e-06,
+ "loss": 0.0256,
+ "step": 12040
+ },
+ {
+ "epoch": 26.9,
+ "learning_rate": 5.119426144430722e-06,
+ "loss": 0.0257,
+ "step": 12050
+ },
+ {
+ "epoch": 26.92,
+ "learning_rate": 5.103659079494584e-06,
+ "loss": 0.0321,
+ "step": 12060
+ },
+ {
+ "epoch": 26.94,
+ "learning_rate": 5.087908009099573e-06,
+ "loss": 0.0289,
+ "step": 12070
+ },
+ {
+ "epoch": 26.96,
+ "learning_rate": 5.072172984698638e-06,
+ "loss": 0.028,
+ "step": 12080
+ },
+ {
+ "epoch": 26.99,
+ "learning_rate": 5.056454057692295e-06,
+ "loss": 0.0357,
+ "step": 12090
+ },
+ {
+ "epoch": 27.01,
+ "learning_rate": 5.040751279428491e-06,
+ "loss": 0.025,
+ "step": 12100
+ },
+ {
+ "epoch": 27.03,
+ "learning_rate": 5.025064701202413e-06,
+ "loss": 0.0286,
+ "step": 12110
+ },
+ {
+ "epoch": 27.05,
+ "learning_rate": 5.009394374256333e-06,
+ "loss": 0.0251,
+ "step": 12120
+ },
+ {
+ "epoch": 27.08,
+ "learning_rate": 4.993740349779437e-06,
+ "loss": 0.024,
+ "step": 12130
+ },
+ {
+ "epoch": 27.1,
+ "learning_rate": 4.978102678907644e-06,
+ "loss": 0.0304,
+ "step": 12140
+ },
+ {
+ "epoch": 27.12,
+ "learning_rate": 4.962481412723468e-06,
+ "loss": 0.0269,
+ "step": 12150
+ },
+ {
+ "epoch": 27.14,
+ "learning_rate": 4.946876602255822e-06,
+ "loss": 0.0166,
+ "step": 12160
+ },
+ {
+ "epoch": 27.17,
+ "learning_rate": 4.931288298479879e-06,
+ "loss": 0.0257,
+ "step": 12170
+ },
+ {
+ "epoch": 27.19,
+ "learning_rate": 4.915716552316873e-06,
+ "loss": 0.0244,
+ "step": 12180
+ },
+ {
+ "epoch": 27.21,
+ "learning_rate": 4.900161414633962e-06,
+ "loss": 0.0314,
+ "step": 12190
+ },
+ {
+ "epoch": 27.23,
+ "learning_rate": 4.88462293624405e-06,
+ "loss": 0.0194,
+ "step": 12200
+ },
+ {
+ "epoch": 27.25,
+ "learning_rate": 4.869101167905621e-06,
+ "loss": 0.032,
+ "step": 12210
+ },
+ {
+ "epoch": 27.28,
+ "learning_rate": 4.853596160322565e-06,
+ "loss": 0.0269,
+ "step": 12220
+ },
+ {
+ "epoch": 27.3,
+ "learning_rate": 4.838107964144029e-06,
+ "loss": 0.0228,
+ "step": 12230
+ },
+ {
+ "epoch": 27.32,
+ "learning_rate": 4.822636629964245e-06,
+ "loss": 0.0217,
+ "step": 12240
+ },
+ {
+ "epoch": 27.34,
+ "learning_rate": 4.807182208322356e-06,
+ "loss": 0.0254,
+ "step": 12250
+ },
+ {
+ "epoch": 27.37,
+ "learning_rate": 4.7917447497022604e-06,
+ "loss": 0.0208,
+ "step": 12260
+ },
+ {
+ "epoch": 27.39,
+ "learning_rate": 4.7763243045324495e-06,
+ "loss": 0.0374,
+ "step": 12270
+ },
+ {
+ "epoch": 27.41,
+ "learning_rate": 4.760920923185834e-06,
+ "loss": 0.0338,
+ "step": 12280
+ },
+ {
+ "epoch": 27.43,
+ "learning_rate": 4.745534655979579e-06,
+ "loss": 0.0353,
+ "step": 12290
+ },
+ {
+ "epoch": 27.46,
+ "learning_rate": 4.730165553174955e-06,
+ "loss": 0.0285,
+ "step": 12300
+ },
+ {
+ "epoch": 27.48,
+ "learning_rate": 4.714813664977149e-06,
+ "loss": 0.0264,
+ "step": 12310
+ },
+ {
+ "epoch": 27.5,
+ "learning_rate": 4.6994790415351285e-06,
+ "loss": 0.0255,
+ "step": 12320
+ },
+ {
+ "epoch": 27.52,
+ "learning_rate": 4.684161732941449e-06,
+ "loss": 0.0284,
+ "step": 12330
+ },
+ {
+ "epoch": 27.54,
+ "learning_rate": 4.6688617892321155e-06,
+ "loss": 0.0255,
+ "step": 12340
+ },
+ {
+ "epoch": 27.57,
+ "learning_rate": 4.653579260386406e-06,
+ "loss": 0.0293,
+ "step": 12350
+ },
+ {
+ "epoch": 27.59,
+ "learning_rate": 4.638314196326709e-06,
+ "loss": 0.0217,
+ "step": 12360
+ },
+ {
+ "epoch": 27.61,
+ "learning_rate": 4.623066646918366e-06,
+ "loss": 0.0325,
+ "step": 12370
+ },
+ {
+ "epoch": 27.63,
+ "learning_rate": 4.607836661969494e-06,
+ "loss": 0.0256,
+ "step": 12380
+ },
+ {
+ "epoch": 27.66,
+ "learning_rate": 4.592624291230847e-06,
+ "loss": 0.0262,
+ "step": 12390
+ },
+ {
+ "epoch": 27.68,
+ "learning_rate": 4.577429584395626e-06,
+ "loss": 0.0324,
+ "step": 12400
+ },
+ {
+ "epoch": 27.7,
+ "learning_rate": 4.562252591099348e-06,
+ "loss": 0.023,
+ "step": 12410
+ },
+ {
+ "epoch": 27.72,
+ "learning_rate": 4.547093360919645e-06,
+ "loss": 0.0289,
+ "step": 12420
+ },
+ {
+ "epoch": 27.75,
+ "learning_rate": 4.531951943376142e-06,
+ "loss": 0.0338,
+ "step": 12430
+ },
+ {
+ "epoch": 27.77,
+ "learning_rate": 4.516828387930265e-06,
+ "loss": 0.0258,
+ "step": 12440
+ },
+ {
+ "epoch": 27.79,
+ "learning_rate": 4.501722743985104e-06,
+ "loss": 0.028,
+ "step": 12450
+ },
+ {
+ "epoch": 27.81,
+ "learning_rate": 4.486635060885221e-06,
+ "loss": 0.0341,
+ "step": 12460
+ },
+ {
+ "epoch": 27.83,
+ "learning_rate": 4.471565387916518e-06,
+ "loss": 0.0249,
+ "step": 12470
+ },
+ {
+ "epoch": 27.86,
+ "learning_rate": 4.4565137743060675e-06,
+ "loss": 0.0203,
+ "step": 12480
+ },
+ {
+ "epoch": 27.88,
+ "learning_rate": 4.441480269221935e-06,
+ "loss": 0.0249,
+ "step": 12490
+ },
+ {
+ "epoch": 27.9,
+ "learning_rate": 4.426464921773046e-06,
+ "loss": 0.0314,
+ "step": 12500
+ },
+ {
+ "epoch": 27.92,
+ "learning_rate": 4.411467781009006e-06,
+ "loss": 0.0303,
+ "step": 12510
+ },
+ {
+ "epoch": 27.95,
+ "learning_rate": 4.396488895919948e-06,
+ "loss": 0.0254,
+ "step": 12520
+ },
+ {
+ "epoch": 27.97,
+ "learning_rate": 4.381528315436364e-06,
+ "loss": 0.0324,
+ "step": 12530
+ },
+ {
+ "epoch": 27.99,
+ "learning_rate": 4.366586088428963e-06,
+ "loss": 0.0286,
+ "step": 12540
+ },
+ {
+ "epoch": 28.01,
+ "learning_rate": 4.351662263708486e-06,
+ "loss": 0.0249,
+ "step": 12550
+ },
+ {
+ "epoch": 28.04,
+ "learning_rate": 4.336756890025574e-06,
+ "loss": 0.0224,
+ "step": 12560
+ },
+ {
+ "epoch": 28.06,
+ "learning_rate": 4.321870016070583e-06,
+ "loss": 0.0173,
+ "step": 12570
+ },
+ {
+ "epoch": 28.08,
+ "learning_rate": 4.307001690473447e-06,
+ "loss": 0.0208,
+ "step": 12580
+ },
+ {
+ "epoch": 28.1,
+ "learning_rate": 4.2921519618035055e-06,
+ "loss": 0.0235,
+ "step": 12590
+ },
+ {
+ "epoch": 28.12,
+ "learning_rate": 4.277320878569349e-06,
+ "loss": 0.0191,
+ "step": 12600
+ },
+ {
+ "epoch": 28.15,
+ "learning_rate": 4.262508489218662e-06,
+ "loss": 0.0208,
+ "step": 12610
+ },
+ {
+ "epoch": 28.17,
+ "learning_rate": 4.2477148421380565e-06,
+ "loss": 0.0194,
+ "step": 12620
+ },
+ {
+ "epoch": 28.19,
+ "learning_rate": 4.2329399856529305e-06,
+ "loss": 0.0201,
+ "step": 12630
+ },
+ {
+ "epoch": 28.21,
+ "learning_rate": 4.218183968027287e-06,
+ "loss": 0.0228,
+ "step": 12640
+ },
+ {
+ "epoch": 28.24,
+ "learning_rate": 4.203446837463606e-06,
+ "loss": 0.0209,
+ "step": 12650
+ },
+ {
+ "epoch": 28.26,
+ "learning_rate": 4.18872864210265e-06,
+ "loss": 0.0239,
+ "step": 12660
+ },
+ {
+ "epoch": 28.28,
+ "learning_rate": 4.174029430023352e-06,
+ "loss": 0.0277,
+ "step": 12670
+ },
+ {
+ "epoch": 28.3,
+ "learning_rate": 4.159349249242609e-06,
+ "loss": 0.0238,
+ "step": 12680
+ },
+ {
+ "epoch": 28.33,
+ "learning_rate": 4.144688147715166e-06,
+ "loss": 0.0219,
+ "step": 12690
+ },
+ {
+ "epoch": 28.35,
+ "learning_rate": 4.1300461733334395e-06,
+ "loss": 0.0327,
+ "step": 12700
+ },
+ {
+ "epoch": 28.37,
+ "learning_rate": 4.115423373927358e-06,
+ "loss": 0.028,
+ "step": 12710
+ },
+ {
+ "epoch": 28.39,
+ "learning_rate": 4.100819797264225e-06,
+ "loss": 0.0184,
+ "step": 12720
+ },
+ {
+ "epoch": 28.42,
+ "learning_rate": 4.086235491048535e-06,
+ "loss": 0.0212,
+ "step": 12730
+ },
+ {
+ "epoch": 28.44,
+ "learning_rate": 4.071670502921843e-06,
+ "loss": 0.0306,
+ "step": 12740
+ },
+ {
+ "epoch": 28.46,
+ "learning_rate": 4.0571248804625995e-06,
+ "loss": 0.0272,
+ "step": 12750
+ },
+ {
+ "epoch": 28.48,
+ "learning_rate": 4.042598671185994e-06,
+ "loss": 0.0224,
+ "step": 12760
+ },
+ {
+ "epoch": 28.5,
+ "learning_rate": 4.028091922543792e-06,
+ "loss": 0.0294,
+ "step": 12770
+ },
+ {
+ "epoch": 28.53,
+ "learning_rate": 4.013604681924201e-06,
+ "loss": 0.0292,
+ "step": 12780
+ },
+ {
+ "epoch": 28.55,
+ "learning_rate": 3.9991369966516905e-06,
+ "loss": 0.0224,
+ "step": 12790
+ },
+ {
+ "epoch": 28.57,
+ "learning_rate": 3.98468891398686e-06,
+ "loss": 0.0211,
+ "step": 12800
+ },
+ {
+ "epoch": 28.59,
+ "learning_rate": 3.970260481126268e-06,
+ "loss": 0.0237,
+ "step": 12810
+ },
+ {
+ "epoch": 28.62,
+ "learning_rate": 3.955851745202286e-06,
+ "loss": 0.0259,
+ "step": 12820
+ },
+ {
+ "epoch": 28.64,
+ "learning_rate": 3.9414627532829465e-06,
+ "loss": 0.0201,
+ "step": 12830
+ },
+ {
+ "epoch": 28.66,
+ "learning_rate": 3.927093552371781e-06,
+ "loss": 0.0279,
+ "step": 12840
+ },
+ {
+ "epoch": 28.68,
+ "learning_rate": 3.912744189407677e-06,
+ "loss": 0.0254,
+ "step": 12850
+ },
+ {
+ "epoch": 28.71,
+ "learning_rate": 3.898414711264707e-06,
+ "loss": 0.0221,
+ "step": 12860
+ },
+ {
+ "epoch": 28.73,
+ "learning_rate": 3.884105164752002e-06,
+ "loss": 0.0201,
+ "step": 12870
+ },
+ {
+ "epoch": 28.75,
+ "learning_rate": 3.8698155966135695e-06,
+ "loss": 0.0199,
+ "step": 12880
+ },
+ {
+ "epoch": 28.77,
+ "learning_rate": 3.8555460535281655e-06,
+ "loss": 0.027,
+ "step": 12890
+ },
+ {
+ "epoch": 28.79,
+ "learning_rate": 3.841296582109127e-06,
+ "loss": 0.0199,
+ "step": 12900
+ },
+ {
+ "epoch": 28.82,
+ "learning_rate": 3.82706722890423e-06,
+ "loss": 0.0155,
+ "step": 12910
+ },
+ {
+ "epoch": 28.84,
+ "learning_rate": 3.8128580403955184e-06,
+ "loss": 0.0238,
+ "step": 12920
+ },
+ {
+ "epoch": 28.86,
+ "learning_rate": 3.7986690629991786e-06,
+ "loss": 0.0253,
+ "step": 12930
+ },
+ {
+ "epoch": 28.88,
+ "learning_rate": 3.7845003430653727e-06,
+ "loss": 0.0222,
+ "step": 12940
+ },
+ {
+ "epoch": 28.91,
+ "learning_rate": 3.7703519268780808e-06,
+ "loss": 0.0279,
+ "step": 12950
+ },
+ {
+ "epoch": 28.93,
+ "learning_rate": 3.7562238606549694e-06,
+ "loss": 0.0345,
+ "step": 12960
+ },
+ {
+ "epoch": 28.95,
+ "learning_rate": 3.7421161905472182e-06,
+ "loss": 0.0244,
+ "step": 12970
+ },
+ {
+ "epoch": 28.97,
+ "learning_rate": 3.7280289626393874e-06,
+ "loss": 0.0228,
+ "step": 12980
+ },
+ {
+ "epoch": 29.0,
+ "learning_rate": 3.7139622229492587e-06,
+ "loss": 0.0302,
+ "step": 12990
+ },
+ {
+ "epoch": 29.02,
+ "learning_rate": 3.6999160174276914e-06,
+ "loss": 0.0213,
+ "step": 13000
+ },
+ {
+ "epoch": 29.04,
+ "learning_rate": 3.685890391958453e-06,
+ "loss": 0.0195,
+ "step": 13010
+ },
+ {
+ "epoch": 29.06,
+ "learning_rate": 3.671885392358101e-06,
+ "loss": 0.0201,
+ "step": 13020
+ },
+ {
+ "epoch": 29.08,
+ "learning_rate": 3.6579010643757982e-06,
+ "loss": 0.0197,
+ "step": 13030
+ },
+ {
+ "epoch": 29.11,
+ "learning_rate": 3.6439374536931992e-06,
+ "loss": 0.0246,
+ "step": 13040
+ },
+ {
+ "epoch": 29.13,
+ "learning_rate": 3.629994605924265e-06,
+ "loss": 0.0239,
+ "step": 13050
+ },
+ {
+ "epoch": 29.15,
+ "learning_rate": 3.616072566615141e-06,
+ "loss": 0.0203,
+ "step": 13060
+ },
+ {
+ "epoch": 29.17,
+ "learning_rate": 3.602171381244007e-06,
+ "loss": 0.0179,
+ "step": 13070
+ },
+ {
+ "epoch": 29.2,
+ "learning_rate": 3.588291095220905e-06,
+ "loss": 0.0229,
+ "step": 13080
+ },
+ {
+ "epoch": 29.22,
+ "learning_rate": 3.574431753887617e-06,
+ "loss": 0.0214,
+ "step": 13090
+ },
+ {
+ "epoch": 29.24,
+ "learning_rate": 3.560593402517498e-06,
+ "loss": 0.0199,
+ "step": 13100
+ },
+ {
+ "epoch": 29.26,
+ "learning_rate": 3.546776086315349e-06,
+ "loss": 0.0151,
+ "step": 13110
+ },
+ {
+ "epoch": 29.29,
+ "learning_rate": 3.53297985041724e-06,
+ "loss": 0.0196,
+ "step": 13120
+ },
+ {
+ "epoch": 29.31,
+ "learning_rate": 3.519204739890394e-06,
+ "loss": 0.023,
+ "step": 13130
+ },
+ {
+ "epoch": 29.33,
+ "learning_rate": 3.50545079973302e-06,
+ "loss": 0.0148,
+ "step": 13140
+ },
+ {
+ "epoch": 29.35,
+ "learning_rate": 3.491718074874174e-06,
+ "loss": 0.0121,
+ "step": 13150
+ },
+ {
+ "epoch": 29.38,
+ "learning_rate": 3.4780066101736e-06,
+ "loss": 0.0172,
+ "step": 13160
+ },
+ {
+ "epoch": 29.4,
+ "learning_rate": 3.464316450421602e-06,
+ "loss": 0.0205,
+ "step": 13170
+ },
+ {
+ "epoch": 29.42,
+ "learning_rate": 3.450647640338891e-06,
+ "loss": 0.0186,
+ "step": 13180
+ },
+ {
+ "epoch": 29.44,
+ "learning_rate": 3.4370002245764244e-06,
+ "loss": 0.0254,
+ "step": 13190
+ },
+ {
+ "epoch": 29.46,
+ "learning_rate": 3.4233742477152854e-06,
+ "loss": 0.0232,
+ "step": 13200
+ },
+ {
+ "epoch": 29.49,
+ "learning_rate": 3.409769754266513e-06,
+ "loss": 0.0204,
+ "step": 13210
+ },
+ {
+ "epoch": 29.51,
+ "learning_rate": 3.3961867886709756e-06,
+ "loss": 0.0183,
+ "step": 13220
+ },
+ {
+ "epoch": 29.53,
+ "learning_rate": 3.382625395299214e-06,
+ "loss": 0.0173,
+ "step": 13230
+ },
+ {
+ "epoch": 29.55,
+ "learning_rate": 3.369085618451308e-06,
+ "loss": 0.0216,
+ "step": 13240
+ },
+ {
+ "epoch": 29.58,
+ "learning_rate": 3.355567502356709e-06,
+ "loss": 0.0276,
+ "step": 13250
+ },
+ {
+ "epoch": 29.6,
+ "learning_rate": 3.3420710911741284e-06,
+ "loss": 0.0336,
+ "step": 13260
+ },
+ {
+ "epoch": 29.62,
+ "learning_rate": 3.3285964289913597e-06,
+ "loss": 0.0195,
+ "step": 13270
+ },
+ {
+ "epoch": 29.64,
+ "learning_rate": 3.3151435598251646e-06,
+ "loss": 0.027,
+ "step": 13280
+ },
+ {
+ "epoch": 29.67,
+ "learning_rate": 3.301712527621099e-06,
+ "loss": 0.016,
+ "step": 13290
+ },
+ {
+ "epoch": 29.69,
+ "learning_rate": 3.2883033762534033e-06,
+ "loss": 0.0237,
+ "step": 13300
+ },
+ {
+ "epoch": 29.71,
+ "learning_rate": 3.2749161495248337e-06,
+ "loss": 0.0202,
+ "step": 13310
+ },
+ {
+ "epoch": 29.73,
+ "learning_rate": 3.2615508911665196e-06,
+ "loss": 0.0238,
+ "step": 13320
+ },
+ {
+ "epoch": 29.75,
+ "learning_rate": 3.2482076448378395e-06,
+ "loss": 0.0176,
+ "step": 13330
+ },
+ {
+ "epoch": 29.78,
+ "learning_rate": 3.2348864541262557e-06,
+ "loss": 0.0209,
+ "step": 13340
+ },
+ {
+ "epoch": 29.8,
+ "learning_rate": 3.2215873625471938e-06,
+ "loss": 0.0242,
+ "step": 13350
+ },
+ {
+ "epoch": 29.82,
+ "learning_rate": 3.2083104135438792e-06,
+ "loss": 0.0224,
+ "step": 13360
+ },
+ {
+ "epoch": 29.84,
+ "learning_rate": 3.1950556504872133e-06,
+ "loss": 0.0249,
+ "step": 13370
+ },
+ {
+ "epoch": 29.87,
+ "learning_rate": 3.1818231166756208e-06,
+ "loss": 0.0223,
+ "step": 13380
+ },
+ {
+ "epoch": 29.89,
+ "learning_rate": 3.168612855334916e-06,
+ "loss": 0.0262,
+ "step": 13390
+ },
+ {
+ "epoch": 29.91,
+ "learning_rate": 3.155424909618149e-06,
+ "loss": 0.0229,
+ "step": 13400
+ },
+ {
+ "epoch": 29.93,
+ "learning_rate": 3.14225932260548e-06,
+ "loss": 0.0161,
+ "step": 13410
+ },
+ {
+ "epoch": 29.96,
+ "learning_rate": 3.1291161373040313e-06,
+ "loss": 0.0252,
+ "step": 13420
+ },
+ {
+ "epoch": 29.98,
+ "learning_rate": 3.1159953966477406e-06,
+ "loss": 0.0208,
+ "step": 13430
+ },
+ {
+ "epoch": 30.0,
+ "learning_rate": 3.1028971434972366e-06,
+ "loss": 0.028,
+ "step": 13440
+ },
+ {
+ "epoch": 30.02,
+ "learning_rate": 3.089821420639679e-06,
+ "loss": 0.0184,
+ "step": 13450
+ },
+ {
+ "epoch": 30.04,
+ "learning_rate": 3.0767682707886383e-06,
+ "loss": 0.0215,
+ "step": 13460
+ },
+ {
+ "epoch": 30.07,
+ "learning_rate": 3.0637377365839437e-06,
+ "loss": 0.017,
+ "step": 13470
+ },
+ {
+ "epoch": 30.09,
+ "learning_rate": 3.0507298605915515e-06,
+ "loss": 0.0201,
+ "step": 13480
+ },
+ {
+ "epoch": 30.11,
+ "learning_rate": 3.037744685303391e-06,
+ "loss": 0.0251,
+ "step": 13490
+ },
+ {
+ "epoch": 30.13,
+ "learning_rate": 3.0247822531372495e-06,
+ "loss": 0.0183,
+ "step": 13500
+ },
+ {
+ "epoch": 30.16,
+ "learning_rate": 3.0118426064366115e-06,
+ "loss": 0.021,
+ "step": 13510
+ },
+ {
+ "epoch": 30.18,
+ "learning_rate": 2.9989257874705347e-06,
+ "loss": 0.0187,
+ "step": 13520
+ },
+ {
+ "epoch": 30.2,
+ "learning_rate": 2.9860318384335076e-06,
+ "loss": 0.0114,
+ "step": 13530
+ },
+ {
+ "epoch": 30.22,
+ "learning_rate": 2.9731608014453074e-06,
+ "loss": 0.0221,
+ "step": 13540
+ },
+ {
+ "epoch": 30.25,
+ "learning_rate": 2.9603127185508717e-06,
+ "loss": 0.0186,
+ "step": 13550
+ },
+ {
+ "epoch": 30.27,
+ "learning_rate": 2.947487631720146e-06,
+ "loss": 0.0179,
+ "step": 13560
+ },
+ {
+ "epoch": 30.29,
+ "learning_rate": 2.934685582847968e-06,
+ "loss": 0.0211,
+ "step": 13570
+ },
+ {
+ "epoch": 30.31,
+ "learning_rate": 2.9219066137539077e-06,
+ "loss": 0.0171,
+ "step": 13580
+ },
+ {
+ "epoch": 30.33,
+ "learning_rate": 2.909150766182152e-06,
+ "loss": 0.014,
+ "step": 13590
+ },
+ {
+ "epoch": 30.36,
+ "learning_rate": 2.8964180818013476e-06,
+ "loss": 0.0265,
+ "step": 13600
+ },
+ {
+ "epoch": 30.38,
+ "learning_rate": 2.883708602204485e-06,
+ "loss": 0.0223,
+ "step": 13610
+ },
+ {
+ "epoch": 30.4,
+ "learning_rate": 2.871022368908749e-06,
+ "loss": 0.0152,
+ "step": 13620
+ },
+ {
+ "epoch": 30.42,
+ "learning_rate": 2.8583594233553925e-06,
+ "loss": 0.0196,
+ "step": 13630
+ },
+ {
+ "epoch": 30.45,
+ "learning_rate": 2.8457198069095827e-06,
+ "loss": 0.0141,
+ "step": 13640
+ },
+ {
+ "epoch": 30.47,
+ "learning_rate": 2.833103560860293e-06,
+ "loss": 0.024,
+ "step": 13650
+ },
+ {
+ "epoch": 30.49,
+ "learning_rate": 2.8205107264201526e-06,
+ "loss": 0.0183,
+ "step": 13660
+ },
+ {
+ "epoch": 30.51,
+ "learning_rate": 2.8079413447253023e-06,
+ "loss": 0.0154,
+ "step": 13670
+ },
+ {
+ "epoch": 30.54,
+ "learning_rate": 2.795395456835287e-06,
+ "loss": 0.0144,
+ "step": 13680
+ },
+ {
+ "epoch": 30.56,
+ "learning_rate": 2.7828731037328915e-06,
+ "loss": 0.0157,
+ "step": 13690
+ },
+ {
+ "epoch": 30.58,
+ "learning_rate": 2.770374326324031e-06,
+ "loss": 0.0206,
+ "step": 13700
+ },
+ {
+ "epoch": 30.6,
+ "learning_rate": 2.7578991654376053e-06,
+ "loss": 0.0207,
+ "step": 13710
+ },
+ {
+ "epoch": 30.62,
+ "learning_rate": 2.74544766182537e-06,
+ "loss": 0.0196,
+ "step": 13720
+ },
+ {
+ "epoch": 30.65,
+ "learning_rate": 2.7330198561617917e-06,
+ "loss": 0.0231,
+ "step": 13730
+ },
+ {
+ "epoch": 30.67,
+ "learning_rate": 2.7206157890439367e-06,
+ "loss": 0.0176,
+ "step": 13740
+ },
+ {
+ "epoch": 30.69,
+ "learning_rate": 2.708235500991314e-06,
+ "loss": 0.0255,
+ "step": 13750
+ },
+ {
+ "epoch": 30.71,
+ "learning_rate": 2.6958790324457664e-06,
+ "loss": 0.0269,
+ "step": 13760
+ },
+ {
+ "epoch": 30.74,
+ "learning_rate": 2.6835464237713207e-06,
+ "loss": 0.0182,
+ "step": 13770
+ },
+ {
+ "epoch": 30.76,
+ "learning_rate": 2.6712377152540646e-06,
+ "loss": 0.0177,
+ "step": 13780
+ },
+ {
+ "epoch": 30.78,
+ "learning_rate": 2.6589529471020158e-06,
+ "loss": 0.0168,
+ "step": 13790
+ },
+ {
+ "epoch": 30.8,
+ "learning_rate": 2.646692159444977e-06,
+ "loss": 0.0181,
+ "step": 13800
+ },
+ {
+ "epoch": 30.83,
+ "learning_rate": 2.6344553923344284e-06,
+ "loss": 0.0241,
+ "step": 13810
+ },
+ {
+ "epoch": 30.85,
+ "learning_rate": 2.6222426857433737e-06,
+ "loss": 0.0189,
+ "step": 13820
+ },
+ {
+ "epoch": 30.87,
+ "learning_rate": 2.610054079566229e-06,
+ "loss": 0.0217,
+ "step": 13830
+ },
+ {
+ "epoch": 30.89,
+ "learning_rate": 2.5978896136186704e-06,
+ "loss": 0.0124,
+ "step": 13840
+ },
+ {
+ "epoch": 30.92,
+ "learning_rate": 2.585749327637531e-06,
+ "loss": 0.0246,
+ "step": 13850
+ },
+ {
+ "epoch": 30.94,
+ "learning_rate": 2.573633261280649e-06,
+ "loss": 0.0154,
+ "step": 13860
+ },
+ {
+ "epoch": 30.96,
+ "learning_rate": 2.56154145412675e-06,
+ "loss": 0.019,
+ "step": 13870
+ },
+ {
+ "epoch": 30.98,
+ "learning_rate": 2.549473945675306e-06,
+ "loss": 0.0301,
+ "step": 13880
+ },
+ {
+ "epoch": 31.0,
+ "learning_rate": 2.537430775346422e-06,
+ "loss": 0.011,
+ "step": 13890
+ },
+ {
+ "epoch": 31.03,
+ "learning_rate": 2.5254119824807e-06,
+ "loss": 0.0167,
+ "step": 13900
+ },
+ {
+ "epoch": 31.05,
+ "learning_rate": 2.5134176063391004e-06,
+ "loss": 0.0123,
+ "step": 13910
+ },
+ {
+ "epoch": 31.07,
+ "learning_rate": 2.5014476861028314e-06,
+ "loss": 0.0208,
+ "step": 13920
+ },
+ {
+ "epoch": 31.09,
+ "learning_rate": 2.4895022608732124e-06,
+ "loss": 0.0137,
+ "step": 13930
+ },
+ {
+ "epoch": 31.12,
+ "learning_rate": 2.477581369671546e-06,
+ "loss": 0.0111,
+ "step": 13940
+ },
+ {
+ "epoch": 31.14,
+ "learning_rate": 2.465685051438985e-06,
+ "loss": 0.0155,
+ "step": 13950
+ },
+ {
+ "epoch": 31.16,
+ "learning_rate": 2.4538133450364234e-06,
+ "loss": 0.0166,
+ "step": 13960
+ },
+ {
+ "epoch": 31.18,
+ "learning_rate": 2.441966289244342e-06,
+ "loss": 0.0116,
+ "step": 13970
+ },
+ {
+ "epoch": 31.21,
+ "learning_rate": 2.4301439227627135e-06,
+ "loss": 0.0193,
+ "step": 13980
+ },
+ {
+ "epoch": 31.23,
+ "learning_rate": 2.418346284210845e-06,
+ "loss": 0.0142,
+ "step": 13990
+ },
+ {
+ "epoch": 31.25,
+ "learning_rate": 2.406573412127278e-06,
+ "loss": 0.0192,
+ "step": 14000
+ },
+ {
+ "epoch": 31.27,
+ "learning_rate": 2.3948253449696435e-06,
+ "loss": 0.0213,
+ "step": 14010
+ },
+ {
+ "epoch": 31.29,
+ "learning_rate": 2.383102121114549e-06,
+ "loss": 0.0149,
+ "step": 14020
+ },
+ {
+ "epoch": 31.32,
+ "learning_rate": 2.3714037788574483e-06,
+ "loss": 0.0209,
+ "step": 14030
+ },
+ {
+ "epoch": 31.34,
+ "learning_rate": 2.359730356412507e-06,
+ "loss": 0.021,
+ "step": 14040
+ },
+ {
+ "epoch": 31.36,
+ "learning_rate": 2.3480818919125027e-06,
+ "loss": 0.0207,
+ "step": 14050
+ },
+ {
+ "epoch": 31.38,
+ "learning_rate": 2.3364584234086663e-06,
+ "loss": 0.0154,
+ "step": 14060
+ },
+ {
+ "epoch": 31.41,
+ "learning_rate": 2.3248599888705924e-06,
+ "loss": 0.0218,
+ "step": 14070
+ },
+ {
+ "epoch": 31.43,
+ "learning_rate": 2.3132866261860863e-06,
+ "loss": 0.0141,
+ "step": 14080
+ },
+ {
+ "epoch": 31.45,
+ "learning_rate": 2.301738373161061e-06,
+ "loss": 0.0176,
+ "step": 14090
+ },
+ {
+ "epoch": 31.47,
+ "learning_rate": 2.2902152675194033e-06,
+ "loss": 0.0148,
+ "step": 14100
+ },
+ {
+ "epoch": 31.5,
+ "learning_rate": 2.2787173469028536e-06,
+ "loss": 0.0127,
+ "step": 14110
+ },
+ {
+ "epoch": 31.52,
+ "learning_rate": 2.2672446488708768e-06,
+ "loss": 0.016,
+ "step": 14120
+ },
+ {
+ "epoch": 31.54,
+ "learning_rate": 2.25579721090055e-06,
+ "loss": 0.0173,
+ "step": 14130
+ },
+ {
+ "epoch": 31.56,
+ "learning_rate": 2.2443750703864377e-06,
+ "loss": 0.0127,
+ "step": 14140
+ },
+ {
+ "epoch": 31.58,
+ "learning_rate": 2.2329782646404573e-06,
+ "loss": 0.0178,
+ "step": 14150
+ },
+ {
+ "epoch": 31.61,
+ "learning_rate": 2.221606830891776e-06,
+ "loss": 0.0251,
+ "step": 14160
+ },
+ {
+ "epoch": 31.63,
+ "learning_rate": 2.210260806286676e-06,
+ "loss": 0.0198,
+ "step": 14170
+ },
+ {
+ "epoch": 31.65,
+ "learning_rate": 2.19894022788844e-06,
+ "loss": 0.0121,
+ "step": 14180
+ },
+ {
+ "epoch": 31.67,
+ "learning_rate": 2.1876451326772196e-06,
+ "loss": 0.0117,
+ "step": 14190
+ },
+ {
+ "epoch": 31.7,
+ "learning_rate": 2.1763755575499355e-06,
+ "loss": 0.0237,
+ "step": 14200
+ },
+ {
+ "epoch": 31.72,
+ "learning_rate": 2.1651315393201276e-06,
+ "loss": 0.0135,
+ "step": 14210
+ },
+ {
+ "epoch": 31.74,
+ "learning_rate": 2.153913114717867e-06,
+ "loss": 0.0146,
+ "step": 14220
+ },
+ {
+ "epoch": 31.76,
+ "learning_rate": 2.1427203203896054e-06,
+ "loss": 0.0149,
+ "step": 14230
+ },
+ {
+ "epoch": 31.79,
+ "learning_rate": 2.1315531928980803e-06,
+ "loss": 0.0204,
+ "step": 14240
+ },
+ {
+ "epoch": 31.81,
+ "learning_rate": 2.1204117687221794e-06,
+ "loss": 0.0164,
+ "step": 14250
+ },
+ {
+ "epoch": 31.83,
+ "learning_rate": 2.109296084256831e-06,
+ "loss": 0.0229,
+ "step": 14260
+ },
+ {
+ "epoch": 31.85,
+ "learning_rate": 2.0982061758128803e-06,
+ "loss": 0.0161,
+ "step": 14270
+ },
+ {
+ "epoch": 31.88,
+ "learning_rate": 2.087142079616967e-06,
+ "loss": 0.0202,
+ "step": 14280
+ },
+ {
+ "epoch": 31.9,
+ "learning_rate": 2.0761038318114202e-06,
+ "loss": 0.0169,
+ "step": 14290
+ },
+ {
+ "epoch": 31.92,
+ "learning_rate": 2.0650914684541214e-06,
+ "loss": 0.0165,
+ "step": 14300
+ },
+ {
+ "epoch": 31.94,
+ "learning_rate": 2.054105025518408e-06,
+ "loss": 0.019,
+ "step": 14310
+ },
+ {
+ "epoch": 31.96,
+ "learning_rate": 2.043144538892936e-06,
+ "loss": 0.0246,
+ "step": 14320
+ },
+ {
+ "epoch": 31.99,
+ "learning_rate": 2.0322100443815763e-06,
+ "loss": 0.0245,
+ "step": 14330
+ },
+ {
+ "epoch": 32.01,
+ "learning_rate": 2.021301577703293e-06,
+ "loss": 0.0187,
+ "step": 14340
+ },
+ {
+ "epoch": 32.03,
+ "learning_rate": 2.010419174492029e-06,
+ "loss": 0.0127,
+ "step": 14350
+ },
+ {
+ "epoch": 32.05,
+ "learning_rate": 1.9995628702965785e-06,
+ "loss": 0.0109,
+ "step": 14360
+ },
+ {
+ "epoch": 32.08,
+ "learning_rate": 1.988732700580489e-06,
+ "loss": 0.0156,
+ "step": 14370
+ },
+ {
+ "epoch": 32.1,
+ "learning_rate": 1.9779287007219373e-06,
+ "loss": 0.0168,
+ "step": 14380
+ },
+ {
+ "epoch": 32.12,
+ "learning_rate": 1.9671509060136016e-06,
+ "loss": 0.0134,
+ "step": 14390
+ },
+ {
+ "epoch": 32.14,
+ "learning_rate": 1.956399351662569e-06,
+ "loss": 0.0114,
+ "step": 14400
+ },
+ {
+ "epoch": 32.17,
+ "learning_rate": 1.9456740727902045e-06,
+ "loss": 0.014,
+ "step": 14410
+ },
+ {
+ "epoch": 32.19,
+ "learning_rate": 1.934975104432043e-06,
+ "loss": 0.0182,
+ "step": 14420
+ },
+ {
+ "epoch": 32.21,
+ "learning_rate": 1.9243024815376655e-06,
+ "loss": 0.0124,
+ "step": 14430
+ },
+ {
+ "epoch": 32.23,
+ "learning_rate": 1.913656238970604e-06,
+ "loss": 0.0152,
+ "step": 14440
+ },
+ {
+ "epoch": 32.25,
+ "learning_rate": 1.9030364115082023e-06,
+ "loss": 0.0106,
+ "step": 14450
+ },
+ {
+ "epoch": 32.28,
+ "learning_rate": 1.8924430338415279e-06,
+ "loss": 0.0162,
+ "step": 14460
+ },
+ {
+ "epoch": 32.3,
+ "learning_rate": 1.8818761405752372e-06,
+ "loss": 0.0107,
+ "step": 14470
+ },
+ {
+ "epoch": 32.32,
+ "learning_rate": 1.8713357662274778e-06,
+ "loss": 0.0161,
+ "step": 14480
+ },
+ {
+ "epoch": 32.34,
+ "learning_rate": 1.8608219452297671e-06,
+ "loss": 0.0143,
+ "step": 14490
+ },
+ {
+ "epoch": 32.37,
+ "learning_rate": 1.8503347119268833e-06,
+ "loss": 0.0163,
+ "step": 14500
+ },
+ {
+ "epoch": 32.39,
+ "learning_rate": 1.8398741005767562e-06,
+ "loss": 0.0196,
+ "step": 14510
+ },
+ {
+ "epoch": 32.41,
+ "learning_rate": 1.8294401453503418e-06,
+ "loss": 0.0193,
+ "step": 14520
+ },
+ {
+ "epoch": 32.43,
+ "learning_rate": 1.8190328803315305e-06,
+ "loss": 0.0193,
+ "step": 14530
+ },
+ {
+ "epoch": 32.46,
+ "learning_rate": 1.8086523395170174e-06,
+ "loss": 0.0146,
+ "step": 14540
+ },
+ {
+ "epoch": 32.48,
+ "learning_rate": 1.798298556816207e-06,
+ "loss": 0.0148,
+ "step": 14550
+ },
+ {
+ "epoch": 32.5,
+ "learning_rate": 1.7879715660510844e-06,
+ "loss": 0.0177,
+ "step": 14560
+ },
+ {
+ "epoch": 32.52,
+ "learning_rate": 1.7776714009561335e-06,
+ "loss": 0.0139,
+ "step": 14570
+ },
+ {
+ "epoch": 32.54,
+ "learning_rate": 1.7673980951781878e-06,
+ "loss": 0.0114,
+ "step": 14580
+ },
+ {
+ "epoch": 32.57,
+ "learning_rate": 1.7571516822763567e-06,
+ "loss": 0.0109,
+ "step": 14590
+ },
+ {
+ "epoch": 32.59,
+ "learning_rate": 1.7469321957218889e-06,
+ "loss": 0.0064,
+ "step": 14600
+ },
+ {
+ "epoch": 32.61,
+ "learning_rate": 1.7367396688980842e-06,
+ "loss": 0.0155,
+ "step": 14610
+ },
+ {
+ "epoch": 32.63,
+ "learning_rate": 1.726574135100173e-06,
+ "loss": 0.014,
+ "step": 14620
+ },
+ {
+ "epoch": 32.66,
+ "learning_rate": 1.7164356275352024e-06,
+ "loss": 0.0143,
+ "step": 14630
+ },
+ {
+ "epoch": 32.68,
+ "learning_rate": 1.706324179321941e-06,
+ "loss": 0.0118,
+ "step": 14640
+ },
+ {
+ "epoch": 32.7,
+ "learning_rate": 1.6962398234907661e-06,
+ "loss": 0.0091,
+ "step": 14650
+ },
+ {
+ "epoch": 32.72,
+ "learning_rate": 1.6861825929835518e-06,
+ "loss": 0.0143,
+ "step": 14660
+ },
+ {
+ "epoch": 32.75,
+ "learning_rate": 1.6761525206535588e-06,
+ "loss": 0.0104,
+ "step": 14670
+ },
+ {
+ "epoch": 32.77,
+ "learning_rate": 1.6661496392653409e-06,
+ "loss": 0.013,
+ "step": 14680
+ },
+ {
+ "epoch": 32.79,
+ "learning_rate": 1.656173981494622e-06,
+ "loss": 0.0109,
+ "step": 14690
+ },
+ {
+ "epoch": 32.81,
+ "learning_rate": 1.6462255799282024e-06,
+ "loss": 0.0137,
+ "step": 14700
+ },
+ {
+ "epoch": 32.83,
+ "learning_rate": 1.6363044670638385e-06,
+ "loss": 0.0165,
+ "step": 14710
+ },
+ {
+ "epoch": 32.86,
+ "learning_rate": 1.6264106753101526e-06,
+ "loss": 0.0239,
+ "step": 14720
+ },
+ {
+ "epoch": 32.88,
+ "learning_rate": 1.6165442369865148e-06,
+ "loss": 0.0123,
+ "step": 14730
+ },
+ {
+ "epoch": 32.9,
+ "learning_rate": 1.6067051843229431e-06,
+ "loss": 0.0144,
+ "step": 14740
+ },
+ {
+ "epoch": 32.92,
+ "learning_rate": 1.5968935494599991e-06,
+ "loss": 0.0158,
+ "step": 14750
+ },
+ {
+ "epoch": 32.95,
+ "learning_rate": 1.5871093644486724e-06,
+ "loss": 0.0186,
+ "step": 14760
+ },
+ {
+ "epoch": 32.97,
+ "learning_rate": 1.5773526612502943e-06,
+ "loss": 0.0176,
+ "step": 14770
+ },
+ {
+ "epoch": 32.99,
+ "learning_rate": 1.5676234717364114e-06,
+ "loss": 0.0111,
+ "step": 14780
+ },
+ {
+ "epoch": 33.01,
+ "learning_rate": 1.5579218276887054e-06,
+ "loss": 0.0172,
+ "step": 14790
+ },
+ {
+ "epoch": 33.04,
+ "learning_rate": 1.54824776079887e-06,
+ "loss": 0.0109,
+ "step": 14800
+ },
+ {
+ "epoch": 33.06,
+ "learning_rate": 1.5386013026685176e-06,
+ "loss": 0.0144,
+ "step": 14810
+ },
+ {
+ "epoch": 33.08,
+ "learning_rate": 1.5289824848090684e-06,
+ "loss": 0.0084,
+ "step": 14820
+ },
+ {
+ "epoch": 33.1,
+ "learning_rate": 1.519391338641657e-06,
+ "loss": 0.0099,
+ "step": 14830
+ },
+ {
+ "epoch": 33.12,
+ "learning_rate": 1.5098278954970247e-06,
+ "loss": 0.0185,
+ "step": 14840
+ },
+ {
+ "epoch": 33.15,
+ "learning_rate": 1.5002921866154119e-06,
+ "loss": 0.0071,
+ "step": 14850
+ },
+ {
+ "epoch": 33.17,
+ "learning_rate": 1.490784243146468e-06,
+ "loss": 0.0115,
+ "step": 14860
+ },
+ {
+ "epoch": 33.19,
+ "learning_rate": 1.4813040961491366e-06,
+ "loss": 0.0159,
+ "step": 14870
+ },
+ {
+ "epoch": 33.21,
+ "learning_rate": 1.4718517765915653e-06,
+ "loss": 0.0112,
+ "step": 14880
+ },
+ {
+ "epoch": 33.24,
+ "learning_rate": 1.4624273153509982e-06,
+ "loss": 0.0105,
+ "step": 14890
+ },
+ {
+ "epoch": 33.26,
+ "learning_rate": 1.4530307432136803e-06,
+ "loss": 0.0131,
+ "step": 14900
+ },
+ {
+ "epoch": 33.28,
+ "learning_rate": 1.4436620908747434e-06,
+ "loss": 0.015,
+ "step": 14910
+ },
+ {
+ "epoch": 33.3,
+ "learning_rate": 1.434321388938127e-06,
+ "loss": 0.0166,
+ "step": 14920
+ },
+ {
+ "epoch": 33.33,
+ "learning_rate": 1.4250086679164565e-06,
+ "loss": 0.0147,
+ "step": 14930
+ },
+ {
+ "epoch": 33.35,
+ "learning_rate": 1.4157239582309634e-06,
+ "loss": 0.0134,
+ "step": 14940
+ },
+ {
+ "epoch": 33.37,
+ "learning_rate": 1.4064672902113685e-06,
+ "loss": 0.0199,
+ "step": 14950
+ },
+ {
+ "epoch": 33.39,
+ "learning_rate": 1.3972386940957938e-06,
+ "loss": 0.0137,
+ "step": 14960
+ },
+ {
+ "epoch": 33.42,
+ "learning_rate": 1.388038200030668e-06,
+ "loss": 0.0142,
+ "step": 14970
+ },
+ {
+ "epoch": 33.44,
+ "learning_rate": 1.3788658380706089e-06,
+ "loss": 0.0124,
+ "step": 14980
+ },
+ {
+ "epoch": 33.46,
+ "learning_rate": 1.369721638178345e-06,
+ "loss": 0.0155,
+ "step": 14990
+ },
+ {
+ "epoch": 33.48,
+ "learning_rate": 1.3606056302246029e-06,
+ "loss": 0.0099,
+ "step": 15000
+ },
+ {
+ "epoch": 33.5,
+ "learning_rate": 1.3515178439880262e-06,
+ "loss": 0.0134,
+ "step": 15010
+ },
+ {
+ "epoch": 33.53,
+ "learning_rate": 1.342458309155058e-06,
+ "loss": 0.0133,
+ "step": 15020
+ },
+ {
+ "epoch": 33.55,
+ "learning_rate": 1.3334270553198615e-06,
+ "loss": 0.0151,
+ "step": 15030
+ },
+ {
+ "epoch": 33.57,
+ "learning_rate": 1.3244241119842138e-06,
+ "loss": 0.009,
+ "step": 15040
+ },
+ {
+ "epoch": 33.59,
+ "learning_rate": 1.315449508557416e-06,
+ "loss": 0.0158,
+ "step": 15050
+ },
+ {
+ "epoch": 33.62,
+ "learning_rate": 1.3065032743561845e-06,
+ "loss": 0.0093,
+ "step": 15060
+ },
+ {
+ "epoch": 33.64,
+ "learning_rate": 1.2975854386045717e-06,
+ "loss": 0.0119,
+ "step": 15070
+ },
+ {
+ "epoch": 33.66,
+ "learning_rate": 1.2886960304338613e-06,
+ "loss": 0.013,
+ "step": 15080
+ },
+ {
+ "epoch": 33.68,
+ "learning_rate": 1.2798350788824709e-06,
+ "loss": 0.0173,
+ "step": 15090
+ },
+ {
+ "epoch": 33.71,
+ "learning_rate": 1.2710026128958652e-06,
+ "loss": 0.0184,
+ "step": 15100
+ },
+ {
+ "epoch": 33.73,
+ "learning_rate": 1.2621986613264536e-06,
+ "loss": 0.016,
+ "step": 15110
+ },
+ {
+ "epoch": 33.75,
+ "learning_rate": 1.253423252933501e-06,
+ "loss": 0.0088,
+ "step": 15120
+ },
+ {
+ "epoch": 33.77,
+ "learning_rate": 1.2446764163830337e-06,
+ "loss": 0.0169,
+ "step": 15130
+ },
+ {
+ "epoch": 33.79,
+ "learning_rate": 1.2359581802477438e-06,
+ "loss": 0.008,
+ "step": 15140
+ },
+ {
+ "epoch": 33.82,
+ "learning_rate": 1.2272685730068922e-06,
+ "loss": 0.0154,
+ "step": 15150
+ },
+ {
+ "epoch": 33.84,
+ "learning_rate": 1.2186076230462263e-06,
+ "loss": 0.0102,
+ "step": 15160
+ },
+ {
+ "epoch": 33.86,
+ "learning_rate": 1.2099753586578744e-06,
+ "loss": 0.0146,
+ "step": 15170
+ },
+ {
+ "epoch": 33.88,
+ "learning_rate": 1.2013718080402659e-06,
+ "loss": 0.0116,
+ "step": 15180
+ },
+ {
+ "epoch": 33.91,
+ "learning_rate": 1.1927969992980227e-06,
+ "loss": 0.0138,
+ "step": 15190
+ },
+ {
+ "epoch": 33.93,
+ "learning_rate": 1.1842509604418929e-06,
+ "loss": 0.0155,
+ "step": 15200
+ },
+ {
+ "epoch": 33.95,
+ "learning_rate": 1.1757337193886332e-06,
+ "loss": 0.0088,
+ "step": 15210
+ },
+ {
+ "epoch": 33.97,
+ "learning_rate": 1.1672453039609287e-06,
+ "loss": 0.0176,
+ "step": 15220
+ },
+ {
+ "epoch": 34.0,
+ "learning_rate": 1.1587857418873071e-06,
+ "loss": 0.0107,
+ "step": 15230
+ },
+ {
+ "epoch": 34.02,
+ "learning_rate": 1.1503550608020352e-06,
+ "loss": 0.0115,
+ "step": 15240
+ },
+ {
+ "epoch": 34.04,
+ "learning_rate": 1.1419532882450468e-06,
+ "loss": 0.0124,
+ "step": 15250
+ },
+ {
+ "epoch": 34.06,
+ "learning_rate": 1.1335804516618298e-06,
+ "loss": 0.011,
+ "step": 15260
+ },
+ {
+ "epoch": 34.08,
+ "learning_rate": 1.1252365784033592e-06,
+ "loss": 0.0112,
+ "step": 15270
+ },
+ {
+ "epoch": 34.11,
+ "learning_rate": 1.116921695725992e-06,
+ "loss": 0.0172,
+ "step": 15280
+ },
+ {
+ "epoch": 34.13,
+ "learning_rate": 1.108635830791389e-06,
+ "loss": 0.0132,
+ "step": 15290
+ },
+ {
+ "epoch": 34.15,
+ "learning_rate": 1.1003790106664125e-06,
+ "loss": 0.0079,
+ "step": 15300
+ },
+ {
+ "epoch": 34.17,
+ "learning_rate": 1.0921512623230513e-06,
+ "loss": 0.0136,
+ "step": 15310
+ },
+ {
+ "epoch": 34.2,
+ "learning_rate": 1.0839526126383293e-06,
+ "loss": 0.0151,
+ "step": 15320
+ },
+ {
+ "epoch": 34.22,
+ "learning_rate": 1.07578308839421e-06,
+ "loss": 0.0061,
+ "step": 15330
+ },
+ {
+ "epoch": 34.24,
+ "learning_rate": 1.0676427162775216e-06,
+ "loss": 0.0134,
+ "step": 15340
+ },
+ {
+ "epoch": 34.26,
+ "learning_rate": 1.0595315228798563e-06,
+ "loss": 0.0089,
+ "step": 15350
+ },
+ {
+ "epoch": 34.29,
+ "learning_rate": 1.0514495346974928e-06,
+ "loss": 0.0122,
+ "step": 15360
+ },
+ {
+ "epoch": 34.31,
+ "learning_rate": 1.0433967781313115e-06,
+ "loss": 0.0106,
+ "step": 15370
+ },
+ {
+ "epoch": 34.33,
+ "learning_rate": 1.0353732794866988e-06,
+ "loss": 0.0062,
+ "step": 15380
+ },
+ {
+ "epoch": 34.35,
+ "learning_rate": 1.0273790649734649e-06,
+ "loss": 0.0143,
+ "step": 15390
+ },
+ {
+ "epoch": 34.38,
+ "learning_rate": 1.019414160705765e-06,
+ "loss": 0.0094,
+ "step": 15400
+ },
+ {
+ "epoch": 34.4,
+ "learning_rate": 1.0114785927020044e-06,
+ "loss": 0.0112,
+ "step": 15410
+ },
+ {
+ "epoch": 34.42,
+ "learning_rate": 1.003572386884758e-06,
+ "loss": 0.0118,
+ "step": 15420
+ },
+ {
+ "epoch": 34.44,
+ "learning_rate": 9.95695569080688e-07,
+ "loss": 0.0089,
+ "step": 15430
+ },
+ {
+ "epoch": 34.46,
+ "learning_rate": 9.878481650204552e-07,
+ "loss": 0.0053,
+ "step": 15440
+ },
+ {
+ "epoch": 34.49,
+ "learning_rate": 9.800302003386364e-07,
+ "loss": 0.0134,
+ "step": 15450
+ },
+ {
+ "epoch": 34.51,
+ "learning_rate": 9.722417005736385e-07,
+ "loss": 0.0095,
+ "step": 15460
+ },
+ {
+ "epoch": 34.53,
+ "learning_rate": 9.64482691167622e-07,
+ "loss": 0.0135,
+ "step": 15470
+ },
+ {
+ "epoch": 34.55,
+ "learning_rate": 9.56753197466409e-07,
+ "loss": 0.0076,
+ "step": 15480
+ },
+ {
+ "epoch": 34.58,
+ "learning_rate": 9.49053244719409e-07,
+ "loss": 0.013,
+ "step": 15490
+ },
+ {
+ "epoch": 34.6,
+ "learning_rate": 9.413828580795259e-07,
+ "loss": 0.0077,
+ "step": 15500
+ },
+ {
+ "epoch": 34.62,
+ "learning_rate": 9.337420626030879e-07,
+ "loss": 0.0119,
+ "step": 15510
+ },
+ {
+ "epoch": 34.64,
+ "learning_rate": 9.261308832497584e-07,
+ "loss": 0.0133,
+ "step": 15520
+ },
+ {
+ "epoch": 34.67,
+ "learning_rate": 9.185493448824556e-07,
+ "loss": 0.0084,
+ "step": 15530
+ },
+ {
+ "epoch": 34.69,
+ "learning_rate": 9.109974722672677e-07,
+ "loss": 0.0137,
+ "step": 15540
+ },
+ {
+ "epoch": 34.71,
+ "learning_rate": 9.034752900733812e-07,
+ "loss": 0.0158,
+ "step": 15550
+ },
+ {
+ "epoch": 34.73,
+ "learning_rate": 8.95982822872995e-07,
+ "loss": 0.013,
+ "step": 15560
+ },
+ {
+ "epoch": 34.75,
+ "learning_rate": 8.885200951412332e-07,
+ "loss": 0.0107,
+ "step": 15570
+ },
+ {
+ "epoch": 34.78,
+ "learning_rate": 8.81087131256082e-07,
+ "loss": 0.0119,
+ "step": 15580
+ },
+ {
+ "epoch": 34.8,
+ "learning_rate": 8.73683955498289e-07,
+ "loss": 0.0101,
+ "step": 15590
+ },
+ {
+ "epoch": 34.82,
+ "learning_rate": 8.663105920513048e-07,
+ "loss": 0.0154,
+ "step": 15600
+ },
+ {
+ "epoch": 34.84,
+ "learning_rate": 8.589670650011895e-07,
+ "loss": 0.0143,
+ "step": 15610
+ },
+ {
+ "epoch": 34.87,
+ "learning_rate": 8.516533983365394e-07,
+ "loss": 0.0114,
+ "step": 15620
+ },
+ {
+ "epoch": 34.89,
+ "learning_rate": 8.443696159484071e-07,
+ "loss": 0.0078,
+ "step": 15630
+ },
+ {
+ "epoch": 34.91,
+ "learning_rate": 8.371157416302266e-07,
+ "loss": 0.0136,
+ "step": 15640
+ },
+ {
+ "epoch": 34.93,
+ "learning_rate": 8.2989179907773e-07,
+ "loss": 0.0139,
+ "step": 15650
+ },
+ {
+ "epoch": 34.96,
+ "learning_rate": 8.226978118888751e-07,
+ "loss": 0.0153,
+ "step": 15660
+ },
+ {
+ "epoch": 34.98,
+ "learning_rate": 8.155338035637662e-07,
+ "loss": 0.0132,
+ "step": 15670
+ },
+ {
+ "epoch": 35.0,
+ "learning_rate": 8.083997975045798e-07,
+ "loss": 0.0148,
+ "step": 15680
+ },
+ {
+ "epoch": 35.02,
+ "learning_rate": 8.012958170154861e-07,
+ "loss": 0.009,
+ "step": 15690
+ },
+ {
+ "epoch": 35.04,
+ "learning_rate": 7.942218853025652e-07,
+ "loss": 0.0092,
+ "step": 15700
+ },
+ {
+ "epoch": 35.07,
+ "learning_rate": 7.871780254737505e-07,
+ "loss": 0.0097,
+ "step": 15710
+ },
+ {
+ "epoch": 35.09,
+ "learning_rate": 7.801642605387316e-07,
+ "loss": 0.0108,
+ "step": 15720
+ },
+ {
+ "epoch": 35.11,
+ "learning_rate": 7.73180613408896e-07,
+ "loss": 0.0058,
+ "step": 15730
+ },
+ {
+ "epoch": 35.13,
+ "learning_rate": 7.662271068972438e-07,
+ "loss": 0.011,
+ "step": 15740
+ },
+ {
+ "epoch": 35.16,
+ "learning_rate": 7.593037637183176e-07,
+ "loss": 0.0116,
+ "step": 15750
+ },
+ {
+ "epoch": 35.18,
+ "learning_rate": 7.524106064881275e-07,
+ "loss": 0.0081,
+ "step": 15760
+ },
+ {
+ "epoch": 35.2,
+ "learning_rate": 7.455476577240816e-07,
+ "loss": 0.0122,
+ "step": 15770
+ },
+ {
+ "epoch": 35.22,
+ "learning_rate": 7.38714939844899e-07,
+ "loss": 0.0096,
+ "step": 15780
+ },
+ {
+ "epoch": 35.25,
+ "learning_rate": 7.31912475170552e-07,
+ "loss": 0.0133,
+ "step": 15790
+ },
+ {
+ "epoch": 35.27,
+ "learning_rate": 7.251402859221879e-07,
+ "loss": 0.0148,
+ "step": 15800
+ },
+ {
+ "epoch": 35.29,
+ "learning_rate": 7.183983942220485e-07,
+ "loss": 0.0093,
+ "step": 15810
+ },
+ {
+ "epoch": 35.31,
+ "learning_rate": 7.116868220934125e-07,
+ "loss": 0.0171,
+ "step": 15820
+ },
+ {
+ "epoch": 35.33,
+ "learning_rate": 7.050055914605125e-07,
+ "loss": 0.0081,
+ "step": 15830
+ },
+ {
+ "epoch": 35.36,
+ "learning_rate": 6.983547241484656e-07,
+ "loss": 0.0088,
+ "step": 15840
+ },
+ {
+ "epoch": 35.38,
+ "learning_rate": 6.917342418832029e-07,
+ "loss": 0.0082,
+ "step": 15850
+ },
+ {
+ "epoch": 35.4,
+ "learning_rate": 6.851441662914027e-07,
+ "loss": 0.0075,
+ "step": 15860
+ },
+ {
+ "epoch": 35.42,
+ "learning_rate": 6.785845189004081e-07,
+ "loss": 0.009,
+ "step": 15870
+ },
+ {
+ "epoch": 35.45,
+ "learning_rate": 6.720553211381753e-07,
+ "loss": 0.0076,
+ "step": 15880
+ },
+ {
+ "epoch": 35.47,
+ "learning_rate": 6.655565943331821e-07,
+ "loss": 0.0152,
+ "step": 15890
+ },
+ {
+ "epoch": 35.49,
+ "learning_rate": 6.59088359714376e-07,
+ "loss": 0.0159,
+ "step": 15900
+ },
+ {
+ "epoch": 35.51,
+ "learning_rate": 6.526506384110954e-07,
+ "loss": 0.01,
+ "step": 15910
+ },
+ {
+ "epoch": 35.54,
+ "learning_rate": 6.462434514530014e-07,
+ "loss": 0.0088,
+ "step": 15920
+ },
+ {
+ "epoch": 35.56,
+ "learning_rate": 6.398668197700164e-07,
+ "loss": 0.011,
+ "step": 15930
+ },
+ {
+ "epoch": 35.58,
+ "learning_rate": 6.335207641922414e-07,
+ "loss": 0.0086,
+ "step": 15940
+ },
+ {
+ "epoch": 35.6,
+ "learning_rate": 6.272053054499039e-07,
+ "loss": 0.0102,
+ "step": 15950
+ },
+ {
+ "epoch": 35.62,
+ "learning_rate": 6.20920464173278e-07,
+ "loss": 0.0077,
+ "step": 15960
+ },
+ {
+ "epoch": 35.65,
+ "learning_rate": 6.146662608926246e-07,
+ "loss": 0.0117,
+ "step": 15970
+ },
+ {
+ "epoch": 35.67,
+ "learning_rate": 6.084427160381201e-07,
+ "loss": 0.0076,
+ "step": 15980
+ },
+ {
+ "epoch": 35.69,
+ "learning_rate": 6.022498499397922e-07,
+ "loss": 0.0146,
+ "step": 15990
+ },
+ {
+ "epoch": 35.71,
+ "learning_rate": 5.960876828274509e-07,
+ "loss": 0.0057,
+ "step": 16000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 1.4211082860101632e+17,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s1/training_args.bin b/s1/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..7cf6d3ea4694a33ea80e34b1b74ec07dbe160f0a
--- /dev/null
+++ b/s1/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87f1ffd8be10f016d5b0f1117c5a9d55c37250231e131f685a241781c4bad5e5
+size 5819
diff --git a/s1/zero_to_fp32.py b/s1/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..c5246ff52274e1d6142001ccf085186d3545ce57
--- /dev/null
+++ b/s1/zero_to_fp32.py
@@ -0,0 +1,578 @@
+#!/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 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_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_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 _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()
+ 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("-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)
diff --git a/s1_en/README.md b/s1_en/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s1_en/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s1_en/adapter_config.json b/s1_en/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..a771d7433ea3273b0090a5135567909c7b03f365
--- /dev/null
+++ b/s1_en/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "o_proj",
+ "gate_proj",
+ "k_proj",
+ "up_proj",
+ "v_proj",
+ "down_proj",
+ "q_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s1_en/adapter_model.bin b/s1_en/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..e723dace96c0522b745fd75a0cb7aeef24f7fa02
--- /dev/null
+++ b/s1_en/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc9db79395c61ca721098c42fa0604509f6486b854b170f74f5f3f094450850d
+size 639787082
diff --git a/s1_en/config.json b/s1_en/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s1_en/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s1_en/global_step9000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s1_en/global_step9000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..2b2f1935ed92e18c90e5c51327428c390a4c9f1c
--- /dev/null
+++ b/s1_en/global_step9000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:834a66d3e6038595959d043acd94e8307f2cc3d99b1bd7c086e3ea2b0c2db7f2
+size 4089600080
diff --git a/s1_en/global_step9000/mp_rank_00_model_states.pt b/s1_en/global_step9000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e06b38c957b2c0d9bd02b095239b0a6f885a7ea6
--- /dev/null
+++ b/s1_en/global_step9000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b2df990e43d817071629b6f3d8595491b9b28bd994e08cff95295e85abed0a5
+size 28850200603
diff --git a/s1_en/latest b/s1_en/latest
new file mode 100644
index 0000000000000000000000000000000000000000..99981cb573ef099d25110dd7d77225a8bc8a8f96
--- /dev/null
+++ b/s1_en/latest
@@ -0,0 +1 @@
+global_step9000
\ No newline at end of file
diff --git a/s1_en/non_lora_trainables.bin b/s1_en/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..65777999493d31a3ddb074c28bf71ee83c3438e8
--- /dev/null
+++ b/s1_en/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32b51a7f52ae8e9f73dcaa6749addaa2061b0e2c36c2d8808d0db708c5ff6571
+size 41961648
diff --git a/s1_en/rng_state.pth b/s1_en/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..05f556e84cf3edd20b601a6fe84e313bf094f20c
--- /dev/null
+++ b/s1_en/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de6235bea9da7d32a83efa288efc2ba3fb1faa6f936d34421a99a0ab1b4f0e31
+size 14244
diff --git a/s1_en/special_tokens_map.json b/s1_en/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s1_en/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s1_en/tokenizer.model b/s1_en/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s1_en/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s1_en/tokenizer_config.json b/s1_en/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s1_en/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s1_en/trainer_state.json b/s1_en/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b287643390e975855844ca325f214b5c1c54007d
--- /dev/null
+++ b/s1_en/trainer_state.json
@@ -0,0 +1,5416 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 20.089285714285715,
+ "global_step": 9000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 5.7781,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 5.6469,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.0344,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.9625,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 4.0125,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.5672,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 3.0547,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.8406,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.35,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 2.1641,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 2.0945,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 2.4211,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.8055,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 1.6703,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.7797,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.6438,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.4383,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 1.3055,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 1.4102,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 1.2016,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 1.0539,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 1.0406,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 1.4035,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 1.1852,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 1.1125,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 1.3109,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 0.8574,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 1.0813,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 0.9789,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 0.8387,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 1.091,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 0.8059,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 1.2641,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 1.0562,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 0.8744,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 1.383,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 1.1834,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 1.0166,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 0.8836,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 1.4254,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 0.9842,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 0.8867,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 0.8643,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 0.8075,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 0.6127,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 0.7235,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 1.2168,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.6662,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 0.9365,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 0.7451,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.6506,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.91,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.7818,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.7001,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 0.8992,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.8086,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 0.984,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.6067,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 1.0652,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.7799,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.6088,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 1.1023,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.6547,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 0.7667,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.6599,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.8306,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 0.8326,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.6851,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.7607,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.8871,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.7674,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.6289,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 1.0892,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.9307,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.7438,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.7782,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.8975,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.9773,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.7079,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.8084,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.8938,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.7381,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 0.6167,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.81,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.7041,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.6085,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.5408,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.875,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.7848,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.7604,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.4512,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.6634,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.5457,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.5067,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.4481,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.6794,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.7652,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.4444,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.531,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.5603,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.7034,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.6761,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.5227,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.5009,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.4775,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.6511,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.6324,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.5648,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.4833,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.4682,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.4768,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.4215,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.5746,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.4238,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.5196,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.4704,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.7146,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.5448,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.4067,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.8938,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.3698,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.5995,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.5395,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.6432,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.4604,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.4403,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.5813,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.3951,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.4842,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.3954,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.6324,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.6352,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.6021,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.4616,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.3593,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.365,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.4799,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.3185,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.3461,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.204,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.3084,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.5712,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.3108,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.4363,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.3979,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.3407,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.3246,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.4165,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.2841,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.5167,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.2964,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.357,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.39,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.4069,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.4074,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.5279,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.3686,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.4326,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.3332,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.3788,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.5455,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.4021,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.2666,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.3576,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.4862,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.4592,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.4761,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.2884,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.3702,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.3427,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.4193,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.3606,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.3713,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.3484,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.4606,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.4088,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.359,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.3509,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.3525,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.2705,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.26,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.3239,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.2359,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.3296,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.2959,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.2205,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.2575,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.221,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.2626,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.2777,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.3046,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.2604,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.241,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.3288,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.2794,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.3062,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.2662,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.3048,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.2866,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.282,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.2546,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.3228,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.2176,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.2861,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.3095,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.2786,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.2285,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.3668,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.2067,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.2231,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.2101,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.242,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.2307,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.3205,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.3045,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.2441,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.3314,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.247,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.2431,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.3088,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.3641,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.2542,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.2266,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.319,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.2805,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.2405,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.2299,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.1826,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.188,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.2561,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.212,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.1973,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.1916,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.1862,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.1809,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.201,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.2359,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.1496,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.1907,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.1897,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.2717,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.2991,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.1743,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.2042,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.2594,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.2075,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.2343,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.2372,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.1661,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.1943,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.2266,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.2885,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.1997,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.235,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.24,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.1678,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.2187,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.2059,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.2039,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.2337,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.1634,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.2182,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.2083,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.1979,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.2242,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.236,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.3181,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.2292,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.1742,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.1662,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.2049,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.1432,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.1616,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.1592,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.1533,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.2162,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.1873,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.1673,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.1589,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.2069,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.2219,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.1604,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.1625,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.2225,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.1596,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.1582,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.1895,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.1808,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.2021,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.1541,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.212,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.1632,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.1529,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.2059,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.178,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.1758,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.2051,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.1796,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.111,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.1963,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.1287,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.2088,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.1717,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.1738,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.1864,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.2027,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.2229,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.1975,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.1697,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.1808,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.1946,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.1575,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.2549,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.213,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.1685,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.1473,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.1121,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.1535,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.1159,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.1089,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.1588,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.1446,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.1467,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.1643,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.1634,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.1885,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.155,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.1241,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.1368,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.1465,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.1553,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.1294,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.1396,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.1491,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.1295,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.1701,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.1521,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.1751,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.1384,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.1721,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.1448,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.2168,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.1872,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.1604,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.1509,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.1358,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.1438,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.1731,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.1664,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.1609,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.1428,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.1784,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.2344,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.2212,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.1499,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.189,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.1502,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.2051,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.2146,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.1439,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.1508,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.1411,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.1167,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.1353,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.1328,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.1827,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.1448,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.1774,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.1335,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.1423,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.2239,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.1368,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.1408,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.1132,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.1127,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.1386,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.129,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.1468,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.1803,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.1962,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.1572,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.1654,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.1296,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.1212,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.1439,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.161,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.1405,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.1596,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.1206,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.12,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.1461,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.1433,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.1253,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.1434,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.1289,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.1622,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.1163,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.1972,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.1339,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.1652,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.1092,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.1229,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.1518,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.1236,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.1262,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.1774,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.1144,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.113,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.125,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.1167,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.106,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.1178,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.1332,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.1403,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.1629,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.1282,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.131,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.1292,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.1147,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.121,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.1197,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.1425,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.1883,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.1034,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.1188,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.1896,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.1494,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.1187,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.1171,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.126,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.1157,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.1234,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.1064,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.1747,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.1404,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.1148,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.1326,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.1171,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.1321,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.1352,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.1529,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.1549,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.1087,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.1483,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.1237,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.1411,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.1427,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.1107,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.1649,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.1027,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.1219,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.1058,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.103,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.0818,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.1301,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.13,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.0992,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.1254,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.1082,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.1119,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.1087,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.1112,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.0901,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.1059,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.1013,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.1139,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.1257,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.1386,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.1253,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.1114,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.1366,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.1061,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.1147,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.1363,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.1064,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.1186,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.1253,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.1344,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.1282,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.2301,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.1044,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.1143,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.1479,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.1251,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.1313,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.2274,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.1195,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.15,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.1329,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.1,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.1252,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.1236,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.1469,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.0912,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.0954,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.1121,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.0911,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.0932,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.1054,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.0926,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.1523,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.109,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.1197,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.1093,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.1058,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.1269,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.1155,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.1096,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.1692,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.1195,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.091,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.1119,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.1023,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.1058,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.1008,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.1003,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.1242,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.1395,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.1802,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.1082,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.1534,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.1146,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.0976,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.1024,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.1266,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.1389,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.1005,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.1213,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.1109,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.1064,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.1313,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.1261,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.1242,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.1235,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.1172,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.1093,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.1159,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.1043,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.1073,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.1092,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.0877,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.091,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.1076,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.1057,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.1169,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.1066,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.081,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.0977,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.0901,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.1125,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.1227,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.0932,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.0934,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.1105,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.0995,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0942,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.111,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0976,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.0893,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.1018,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.0932,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.126,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.0942,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.101,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.1157,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.1083,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.1428,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.1103,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.1652,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.0877,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.1173,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.1059,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.1242,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.1093,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.1234,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.0886,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.1058,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.1136,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.1305,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.1241,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.0994,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.1642,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.0758,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.0979,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.099,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.1167,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.1191,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.0949,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.1192,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.0887,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.0793,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.1081,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.0881,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.0893,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.0983,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.0893,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.0993,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.1001,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0838,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.0978,
+ "step": 6000
+ },
+ {
+ "epoch": 13.42,
+ "learning_rate": 1.5495257562515308e-05,
+ "loss": 0.0857,
+ "step": 6010
+ },
+ {
+ "epoch": 13.44,
+ "learning_rate": 1.5480148332320562e-05,
+ "loss": 0.1036,
+ "step": 6020
+ },
+ {
+ "epoch": 13.46,
+ "learning_rate": 1.5465021200501046e-05,
+ "loss": 0.1021,
+ "step": 6030
+ },
+ {
+ "epoch": 13.48,
+ "learning_rate": 1.5449876216471525e-05,
+ "loss": 0.1723,
+ "step": 6040
+ },
+ {
+ "epoch": 13.5,
+ "learning_rate": 1.5434713429705078e-05,
+ "loss": 0.1045,
+ "step": 6050
+ },
+ {
+ "epoch": 13.53,
+ "learning_rate": 1.5419532889732943e-05,
+ "loss": 0.0941,
+ "step": 6060
+ },
+ {
+ "epoch": 13.55,
+ "learning_rate": 1.540433464614435e-05,
+ "loss": 0.0917,
+ "step": 6070
+ },
+ {
+ "epoch": 13.57,
+ "learning_rate": 1.5389118748586357e-05,
+ "loss": 0.1004,
+ "step": 6080
+ },
+ {
+ "epoch": 13.59,
+ "learning_rate": 1.537388524676369e-05,
+ "loss": 0.1077,
+ "step": 6090
+ },
+ {
+ "epoch": 13.62,
+ "learning_rate": 1.5358634190438592e-05,
+ "loss": 0.1287,
+ "step": 6100
+ },
+ {
+ "epoch": 13.64,
+ "learning_rate": 1.5343365629430638e-05,
+ "loss": 0.1345,
+ "step": 6110
+ },
+ {
+ "epoch": 13.66,
+ "learning_rate": 1.5328079613616592e-05,
+ "loss": 0.1024,
+ "step": 6120
+ },
+ {
+ "epoch": 13.68,
+ "learning_rate": 1.531277619293023e-05,
+ "loss": 0.1595,
+ "step": 6130
+ },
+ {
+ "epoch": 13.71,
+ "learning_rate": 1.5297455417362194e-05,
+ "loss": 0.0976,
+ "step": 6140
+ },
+ {
+ "epoch": 13.73,
+ "learning_rate": 1.52821173369598e-05,
+ "loss": 0.1126,
+ "step": 6150
+ },
+ {
+ "epoch": 13.75,
+ "learning_rate": 1.526676200182691e-05,
+ "loss": 0.1151,
+ "step": 6160
+ },
+ {
+ "epoch": 13.77,
+ "learning_rate": 1.5251389462123748e-05,
+ "loss": 0.1039,
+ "step": 6170
+ },
+ {
+ "epoch": 13.79,
+ "learning_rate": 1.5235999768066729e-05,
+ "loss": 0.1066,
+ "step": 6180
+ },
+ {
+ "epoch": 13.82,
+ "learning_rate": 1.5220592969928313e-05,
+ "loss": 0.1056,
+ "step": 6190
+ },
+ {
+ "epoch": 13.84,
+ "learning_rate": 1.5205169118036831e-05,
+ "loss": 0.0994,
+ "step": 6200
+ },
+ {
+ "epoch": 13.86,
+ "learning_rate": 1.5189728262776325e-05,
+ "loss": 0.1004,
+ "step": 6210
+ },
+ {
+ "epoch": 13.88,
+ "learning_rate": 1.5174270454586375e-05,
+ "loss": 0.1311,
+ "step": 6220
+ },
+ {
+ "epoch": 13.91,
+ "learning_rate": 1.5158795743961942e-05,
+ "loss": 0.0896,
+ "step": 6230
+ },
+ {
+ "epoch": 13.93,
+ "learning_rate": 1.5143304181453204e-05,
+ "loss": 0.0995,
+ "step": 6240
+ },
+ {
+ "epoch": 13.95,
+ "learning_rate": 1.5127795817665389e-05,
+ "loss": 0.1024,
+ "step": 6250
+ },
+ {
+ "epoch": 13.97,
+ "learning_rate": 1.5112270703258602e-05,
+ "loss": 0.0808,
+ "step": 6260
+ },
+ {
+ "epoch": 14.0,
+ "learning_rate": 1.5096728888947669e-05,
+ "loss": 0.1142,
+ "step": 6270
+ },
+ {
+ "epoch": 14.02,
+ "learning_rate": 1.508117042550197e-05,
+ "loss": 0.1293,
+ "step": 6280
+ },
+ {
+ "epoch": 14.04,
+ "learning_rate": 1.5065595363745272e-05,
+ "loss": 0.1025,
+ "step": 6290
+ },
+ {
+ "epoch": 14.06,
+ "learning_rate": 1.505000375455556e-05,
+ "loss": 0.065,
+ "step": 6300
+ },
+ {
+ "epoch": 14.08,
+ "learning_rate": 1.503439564886487e-05,
+ "loss": 0.1233,
+ "step": 6310
+ },
+ {
+ "epoch": 14.11,
+ "learning_rate": 1.501877109765914e-05,
+ "loss": 0.0944,
+ "step": 6320
+ },
+ {
+ "epoch": 14.13,
+ "learning_rate": 1.5003130151978012e-05,
+ "loss": 0.1119,
+ "step": 6330
+ },
+ {
+ "epoch": 14.15,
+ "learning_rate": 1.4987472862914697e-05,
+ "loss": 0.0677,
+ "step": 6340
+ },
+ {
+ "epoch": 14.17,
+ "learning_rate": 1.4971799281615782e-05,
+ "loss": 0.1003,
+ "step": 6350
+ },
+ {
+ "epoch": 14.2,
+ "learning_rate": 1.4956109459281083e-05,
+ "loss": 0.0886,
+ "step": 6360
+ },
+ {
+ "epoch": 14.22,
+ "learning_rate": 1.4940403447163467e-05,
+ "loss": 0.0973,
+ "step": 6370
+ },
+ {
+ "epoch": 14.24,
+ "learning_rate": 1.4924681296568689e-05,
+ "loss": 0.1315,
+ "step": 6380
+ },
+ {
+ "epoch": 14.26,
+ "learning_rate": 1.4908943058855213e-05,
+ "loss": 0.106,
+ "step": 6390
+ },
+ {
+ "epoch": 14.29,
+ "learning_rate": 1.4893188785434067e-05,
+ "loss": 0.071,
+ "step": 6400
+ },
+ {
+ "epoch": 14.31,
+ "learning_rate": 1.4877418527768654e-05,
+ "loss": 0.0819,
+ "step": 6410
+ },
+ {
+ "epoch": 14.33,
+ "learning_rate": 1.4861632337374596e-05,
+ "loss": 0.1032,
+ "step": 6420
+ },
+ {
+ "epoch": 14.35,
+ "learning_rate": 1.4845830265819552e-05,
+ "loss": 0.084,
+ "step": 6430
+ },
+ {
+ "epoch": 14.38,
+ "learning_rate": 1.483001236472307e-05,
+ "loss": 0.0968,
+ "step": 6440
+ },
+ {
+ "epoch": 14.4,
+ "learning_rate": 1.4814178685756405e-05,
+ "loss": 0.0894,
+ "step": 6450
+ },
+ {
+ "epoch": 14.42,
+ "learning_rate": 1.4798329280642345e-05,
+ "loss": 0.0901,
+ "step": 6460
+ },
+ {
+ "epoch": 14.44,
+ "learning_rate": 1.4782464201155057e-05,
+ "loss": 0.0947,
+ "step": 6470
+ },
+ {
+ "epoch": 14.46,
+ "learning_rate": 1.476658349911991e-05,
+ "loss": 0.1024,
+ "step": 6480
+ },
+ {
+ "epoch": 14.49,
+ "learning_rate": 1.4750687226413305e-05,
+ "loss": 0.0825,
+ "step": 6490
+ },
+ {
+ "epoch": 14.51,
+ "learning_rate": 1.4734775434962504e-05,
+ "loss": 0.0959,
+ "step": 6500
+ },
+ {
+ "epoch": 14.53,
+ "learning_rate": 1.471884817674546e-05,
+ "loss": 0.0967,
+ "step": 6510
+ },
+ {
+ "epoch": 14.55,
+ "learning_rate": 1.4702905503790668e-05,
+ "loss": 0.0998,
+ "step": 6520
+ },
+ {
+ "epoch": 14.58,
+ "learning_rate": 1.4686947468176955e-05,
+ "loss": 0.0963,
+ "step": 6530
+ },
+ {
+ "epoch": 14.6,
+ "learning_rate": 1.467097412203334e-05,
+ "loss": 0.0815,
+ "step": 6540
+ },
+ {
+ "epoch": 14.62,
+ "learning_rate": 1.4654985517538864e-05,
+ "loss": 0.0865,
+ "step": 6550
+ },
+ {
+ "epoch": 14.64,
+ "learning_rate": 1.4638981706922401e-05,
+ "loss": 0.0966,
+ "step": 6560
+ },
+ {
+ "epoch": 14.67,
+ "learning_rate": 1.4622962742462503e-05,
+ "loss": 0.0961,
+ "step": 6570
+ },
+ {
+ "epoch": 14.69,
+ "learning_rate": 1.4606928676487223e-05,
+ "loss": 0.1126,
+ "step": 6580
+ },
+ {
+ "epoch": 14.71,
+ "learning_rate": 1.459087956137394e-05,
+ "loss": 0.0903,
+ "step": 6590
+ },
+ {
+ "epoch": 14.73,
+ "learning_rate": 1.4574815449549209e-05,
+ "loss": 0.0938,
+ "step": 6600
+ },
+ {
+ "epoch": 14.75,
+ "learning_rate": 1.4558736393488553e-05,
+ "loss": 0.0868,
+ "step": 6610
+ },
+ {
+ "epoch": 14.78,
+ "learning_rate": 1.4542642445716326e-05,
+ "loss": 0.1191,
+ "step": 6620
+ },
+ {
+ "epoch": 14.8,
+ "learning_rate": 1.4526533658805517e-05,
+ "loss": 0.0896,
+ "step": 6630
+ },
+ {
+ "epoch": 14.82,
+ "learning_rate": 1.4510410085377606e-05,
+ "loss": 0.0912,
+ "step": 6640
+ },
+ {
+ "epoch": 14.84,
+ "learning_rate": 1.4494271778102358e-05,
+ "loss": 0.1035,
+ "step": 6650
+ },
+ {
+ "epoch": 14.87,
+ "learning_rate": 1.4478118789697675e-05,
+ "loss": 0.1293,
+ "step": 6660
+ },
+ {
+ "epoch": 14.89,
+ "learning_rate": 1.4461951172929419e-05,
+ "loss": 0.104,
+ "step": 6670
+ },
+ {
+ "epoch": 14.91,
+ "learning_rate": 1.4445768980611233e-05,
+ "loss": 0.1208,
+ "step": 6680
+ },
+ {
+ "epoch": 14.93,
+ "learning_rate": 1.4429572265604375e-05,
+ "loss": 0.0944,
+ "step": 6690
+ },
+ {
+ "epoch": 14.96,
+ "learning_rate": 1.4413361080817545e-05,
+ "loss": 0.0806,
+ "step": 6700
+ },
+ {
+ "epoch": 14.98,
+ "learning_rate": 1.4397135479206705e-05,
+ "loss": 0.1948,
+ "step": 6710
+ },
+ {
+ "epoch": 15.0,
+ "learning_rate": 1.4380895513774922e-05,
+ "loss": 0.0773,
+ "step": 6720
+ },
+ {
+ "epoch": 15.02,
+ "learning_rate": 1.436464123757217e-05,
+ "loss": 0.0749,
+ "step": 6730
+ },
+ {
+ "epoch": 15.04,
+ "learning_rate": 1.4348372703695184e-05,
+ "loss": 0.0969,
+ "step": 6740
+ },
+ {
+ "epoch": 15.07,
+ "learning_rate": 1.4332089965287266e-05,
+ "loss": 0.0833,
+ "step": 6750
+ },
+ {
+ "epoch": 15.09,
+ "learning_rate": 1.431579307553812e-05,
+ "loss": 0.0856,
+ "step": 6760
+ },
+ {
+ "epoch": 15.11,
+ "learning_rate": 1.429948208768368e-05,
+ "loss": 0.0899,
+ "step": 6770
+ },
+ {
+ "epoch": 15.13,
+ "learning_rate": 1.4283157055005928e-05,
+ "loss": 0.0931,
+ "step": 6780
+ },
+ {
+ "epoch": 15.16,
+ "learning_rate": 1.4266818030832732e-05,
+ "loss": 0.0838,
+ "step": 6790
+ },
+ {
+ "epoch": 15.18,
+ "learning_rate": 1.4250465068537664e-05,
+ "loss": 0.1244,
+ "step": 6800
+ },
+ {
+ "epoch": 15.2,
+ "learning_rate": 1.4234098221539818e-05,
+ "loss": 0.1073,
+ "step": 6810
+ },
+ {
+ "epoch": 15.22,
+ "learning_rate": 1.4217717543303657e-05,
+ "loss": 0.1048,
+ "step": 6820
+ },
+ {
+ "epoch": 15.25,
+ "learning_rate": 1.4201323087338816e-05,
+ "loss": 0.083,
+ "step": 6830
+ },
+ {
+ "epoch": 15.27,
+ "learning_rate": 1.4184914907199942e-05,
+ "loss": 0.0983,
+ "step": 6840
+ },
+ {
+ "epoch": 15.29,
+ "learning_rate": 1.4168493056486512e-05,
+ "loss": 0.1123,
+ "step": 6850
+ },
+ {
+ "epoch": 15.31,
+ "learning_rate": 1.4152057588842657e-05,
+ "loss": 0.0909,
+ "step": 6860
+ },
+ {
+ "epoch": 15.33,
+ "learning_rate": 1.4135608557956992e-05,
+ "loss": 0.0812,
+ "step": 6870
+ },
+ {
+ "epoch": 15.36,
+ "learning_rate": 1.4119146017562441e-05,
+ "loss": 0.0793,
+ "step": 6880
+ },
+ {
+ "epoch": 15.38,
+ "learning_rate": 1.4102670021436059e-05,
+ "loss": 0.0972,
+ "step": 6890
+ },
+ {
+ "epoch": 15.4,
+ "learning_rate": 1.4086180623398842e-05,
+ "loss": 0.0958,
+ "step": 6900
+ },
+ {
+ "epoch": 15.42,
+ "learning_rate": 1.4069677877315587e-05,
+ "loss": 0.0921,
+ "step": 6910
+ },
+ {
+ "epoch": 15.45,
+ "learning_rate": 1.4053161837094675e-05,
+ "loss": 0.0978,
+ "step": 6920
+ },
+ {
+ "epoch": 15.47,
+ "learning_rate": 1.4036632556687927e-05,
+ "loss": 0.0961,
+ "step": 6930
+ },
+ {
+ "epoch": 15.49,
+ "learning_rate": 1.4020090090090408e-05,
+ "loss": 0.1013,
+ "step": 6940
+ },
+ {
+ "epoch": 15.51,
+ "learning_rate": 1.4003534491340259e-05,
+ "loss": 0.1192,
+ "step": 6950
+ },
+ {
+ "epoch": 15.54,
+ "learning_rate": 1.3986965814518521e-05,
+ "loss": 0.0743,
+ "step": 6960
+ },
+ {
+ "epoch": 15.56,
+ "learning_rate": 1.3970384113748951e-05,
+ "loss": 0.0887,
+ "step": 6970
+ },
+ {
+ "epoch": 15.58,
+ "learning_rate": 1.3953789443197857e-05,
+ "loss": 0.1247,
+ "step": 6980
+ },
+ {
+ "epoch": 15.6,
+ "learning_rate": 1.3937181857073912e-05,
+ "loss": 0.0988,
+ "step": 6990
+ },
+ {
+ "epoch": 15.62,
+ "learning_rate": 1.3920561409627974e-05,
+ "loss": 0.0991,
+ "step": 7000
+ },
+ {
+ "epoch": 15.65,
+ "learning_rate": 1.3903928155152926e-05,
+ "loss": 0.0875,
+ "step": 7010
+ },
+ {
+ "epoch": 15.67,
+ "learning_rate": 1.3887282147983472e-05,
+ "loss": 0.0827,
+ "step": 7020
+ },
+ {
+ "epoch": 15.69,
+ "learning_rate": 1.3870623442495987e-05,
+ "loss": 0.105,
+ "step": 7030
+ },
+ {
+ "epoch": 15.71,
+ "learning_rate": 1.3853952093108323e-05,
+ "loss": 0.0958,
+ "step": 7040
+ },
+ {
+ "epoch": 15.74,
+ "learning_rate": 1.3837268154279628e-05,
+ "loss": 0.0968,
+ "step": 7050
+ },
+ {
+ "epoch": 15.76,
+ "learning_rate": 1.3820571680510187e-05,
+ "loss": 0.1093,
+ "step": 7060
+ },
+ {
+ "epoch": 15.78,
+ "learning_rate": 1.3803862726341224e-05,
+ "loss": 0.0806,
+ "step": 7070
+ },
+ {
+ "epoch": 15.8,
+ "learning_rate": 1.3787141346354733e-05,
+ "loss": 0.0975,
+ "step": 7080
+ },
+ {
+ "epoch": 15.83,
+ "learning_rate": 1.3770407595173301e-05,
+ "loss": 0.091,
+ "step": 7090
+ },
+ {
+ "epoch": 15.85,
+ "learning_rate": 1.375366152745992e-05,
+ "loss": 0.0837,
+ "step": 7100
+ },
+ {
+ "epoch": 15.87,
+ "learning_rate": 1.373690319791783e-05,
+ "loss": 0.0949,
+ "step": 7110
+ },
+ {
+ "epoch": 15.89,
+ "learning_rate": 1.3720132661290311e-05,
+ "loss": 0.0833,
+ "step": 7120
+ },
+ {
+ "epoch": 15.92,
+ "learning_rate": 1.3703349972360527e-05,
+ "loss": 0.0864,
+ "step": 7130
+ },
+ {
+ "epoch": 15.94,
+ "learning_rate": 1.3686555185951334e-05,
+ "loss": 0.0732,
+ "step": 7140
+ },
+ {
+ "epoch": 15.96,
+ "learning_rate": 1.3669748356925112e-05,
+ "loss": 0.0988,
+ "step": 7150
+ },
+ {
+ "epoch": 15.98,
+ "learning_rate": 1.3652929540183578e-05,
+ "loss": 0.095,
+ "step": 7160
+ },
+ {
+ "epoch": 16.0,
+ "learning_rate": 1.3636098790667605e-05,
+ "loss": 0.1002,
+ "step": 7170
+ },
+ {
+ "epoch": 16.03,
+ "learning_rate": 1.3619256163357046e-05,
+ "loss": 0.0688,
+ "step": 7180
+ },
+ {
+ "epoch": 16.05,
+ "learning_rate": 1.3602401713270566e-05,
+ "loss": 0.0817,
+ "step": 7190
+ },
+ {
+ "epoch": 16.07,
+ "learning_rate": 1.3585535495465432e-05,
+ "loss": 0.0883,
+ "step": 7200
+ },
+ {
+ "epoch": 16.09,
+ "learning_rate": 1.3568657565037365e-05,
+ "loss": 0.0868,
+ "step": 7210
+ },
+ {
+ "epoch": 16.12,
+ "learning_rate": 1.3551767977120341e-05,
+ "loss": 0.077,
+ "step": 7220
+ },
+ {
+ "epoch": 16.14,
+ "learning_rate": 1.353486678688642e-05,
+ "loss": 0.077,
+ "step": 7230
+ },
+ {
+ "epoch": 16.16,
+ "learning_rate": 1.351795404954556e-05,
+ "loss": 0.0705,
+ "step": 7240
+ },
+ {
+ "epoch": 16.18,
+ "learning_rate": 1.3501029820345446e-05,
+ "loss": 0.1435,
+ "step": 7250
+ },
+ {
+ "epoch": 16.21,
+ "learning_rate": 1.3484094154571286e-05,
+ "loss": 0.091,
+ "step": 7260
+ },
+ {
+ "epoch": 16.23,
+ "learning_rate": 1.3467147107545668e-05,
+ "loss": 0.0915,
+ "step": 7270
+ },
+ {
+ "epoch": 16.25,
+ "learning_rate": 1.3450188734628344e-05,
+ "loss": 0.0868,
+ "step": 7280
+ },
+ {
+ "epoch": 16.27,
+ "learning_rate": 1.3433219091216069e-05,
+ "loss": 0.0936,
+ "step": 7290
+ },
+ {
+ "epoch": 16.29,
+ "learning_rate": 1.3416238232742414e-05,
+ "loss": 0.069,
+ "step": 7300
+ },
+ {
+ "epoch": 16.32,
+ "learning_rate": 1.3399246214677583e-05,
+ "loss": 0.0822,
+ "step": 7310
+ },
+ {
+ "epoch": 16.34,
+ "learning_rate": 1.338224309252824e-05,
+ "loss": 0.0862,
+ "step": 7320
+ },
+ {
+ "epoch": 16.36,
+ "learning_rate": 1.3365228921837314e-05,
+ "loss": 0.0821,
+ "step": 7330
+ },
+ {
+ "epoch": 16.38,
+ "learning_rate": 1.3348203758183831e-05,
+ "loss": 0.0655,
+ "step": 7340
+ },
+ {
+ "epoch": 16.41,
+ "learning_rate": 1.3331167657182726e-05,
+ "loss": 0.0721,
+ "step": 7350
+ },
+ {
+ "epoch": 16.43,
+ "learning_rate": 1.3314120674484663e-05,
+ "loss": 0.0745,
+ "step": 7360
+ },
+ {
+ "epoch": 16.45,
+ "learning_rate": 1.3297062865775851e-05,
+ "loss": 0.1002,
+ "step": 7370
+ },
+ {
+ "epoch": 16.47,
+ "learning_rate": 1.327999428677786e-05,
+ "loss": 0.0757,
+ "step": 7380
+ },
+ {
+ "epoch": 16.5,
+ "learning_rate": 1.3262914993247454e-05,
+ "loss": 0.0896,
+ "step": 7390
+ },
+ {
+ "epoch": 16.52,
+ "learning_rate": 1.324582504097638e-05,
+ "loss": 0.0974,
+ "step": 7400
+ },
+ {
+ "epoch": 16.54,
+ "learning_rate": 1.3228724485791225e-05,
+ "loss": 0.1337,
+ "step": 7410
+ },
+ {
+ "epoch": 16.56,
+ "learning_rate": 1.321161338355319e-05,
+ "loss": 0.0918,
+ "step": 7420
+ },
+ {
+ "epoch": 16.58,
+ "learning_rate": 1.3194491790157947e-05,
+ "loss": 0.0834,
+ "step": 7430
+ },
+ {
+ "epoch": 16.61,
+ "learning_rate": 1.3177359761535427e-05,
+ "loss": 0.1145,
+ "step": 7440
+ },
+ {
+ "epoch": 16.63,
+ "learning_rate": 1.3160217353649652e-05,
+ "loss": 0.0802,
+ "step": 7450
+ },
+ {
+ "epoch": 16.65,
+ "learning_rate": 1.3143064622498551e-05,
+ "loss": 0.0771,
+ "step": 7460
+ },
+ {
+ "epoch": 16.67,
+ "learning_rate": 1.312590162411378e-05,
+ "loss": 0.1083,
+ "step": 7470
+ },
+ {
+ "epoch": 16.7,
+ "learning_rate": 1.310872841456052e-05,
+ "loss": 0.0742,
+ "step": 7480
+ },
+ {
+ "epoch": 16.72,
+ "learning_rate": 1.3091545049937322e-05,
+ "loss": 0.0922,
+ "step": 7490
+ },
+ {
+ "epoch": 16.74,
+ "learning_rate": 1.3074351586375906e-05,
+ "loss": 0.0839,
+ "step": 7500
+ },
+ {
+ "epoch": 16.76,
+ "learning_rate": 1.305714808004098e-05,
+ "loss": 0.1032,
+ "step": 7510
+ },
+ {
+ "epoch": 16.79,
+ "learning_rate": 1.3039934587130056e-05,
+ "loss": 0.0823,
+ "step": 7520
+ },
+ {
+ "epoch": 16.81,
+ "learning_rate": 1.3022711163873272e-05,
+ "loss": 0.1234,
+ "step": 7530
+ },
+ {
+ "epoch": 16.83,
+ "learning_rate": 1.3005477866533202e-05,
+ "loss": 0.0739,
+ "step": 7540
+ },
+ {
+ "epoch": 16.85,
+ "learning_rate": 1.2988234751404683e-05,
+ "loss": 0.0952,
+ "step": 7550
+ },
+ {
+ "epoch": 16.88,
+ "learning_rate": 1.2970981874814613e-05,
+ "loss": 0.0809,
+ "step": 7560
+ },
+ {
+ "epoch": 16.9,
+ "learning_rate": 1.2953719293121775e-05,
+ "loss": 0.1305,
+ "step": 7570
+ },
+ {
+ "epoch": 16.92,
+ "learning_rate": 1.2936447062716668e-05,
+ "loss": 0.1082,
+ "step": 7580
+ },
+ {
+ "epoch": 16.94,
+ "learning_rate": 1.2919165240021303e-05,
+ "loss": 0.0805,
+ "step": 7590
+ },
+ {
+ "epoch": 16.96,
+ "learning_rate": 1.2901873881489021e-05,
+ "loss": 0.0813,
+ "step": 7600
+ },
+ {
+ "epoch": 16.99,
+ "learning_rate": 1.288457304360432e-05,
+ "loss": 0.0667,
+ "step": 7610
+ },
+ {
+ "epoch": 17.01,
+ "learning_rate": 1.2867262782882662e-05,
+ "loss": 0.0727,
+ "step": 7620
+ },
+ {
+ "epoch": 17.03,
+ "learning_rate": 1.2849943155870284e-05,
+ "loss": 0.081,
+ "step": 7630
+ },
+ {
+ "epoch": 17.05,
+ "learning_rate": 1.2832614219144027e-05,
+ "loss": 0.0627,
+ "step": 7640
+ },
+ {
+ "epoch": 17.08,
+ "learning_rate": 1.2815276029311138e-05,
+ "loss": 0.1201,
+ "step": 7650
+ },
+ {
+ "epoch": 17.1,
+ "learning_rate": 1.2797928643009097e-05,
+ "loss": 0.0793,
+ "step": 7660
+ },
+ {
+ "epoch": 17.12,
+ "learning_rate": 1.2780572116905418e-05,
+ "loss": 0.0972,
+ "step": 7670
+ },
+ {
+ "epoch": 17.14,
+ "learning_rate": 1.276320650769748e-05,
+ "loss": 0.0837,
+ "step": 7680
+ },
+ {
+ "epoch": 17.17,
+ "learning_rate": 1.2745831872112318e-05,
+ "loss": 0.0843,
+ "step": 7690
+ },
+ {
+ "epoch": 17.19,
+ "learning_rate": 1.2728448266906468e-05,
+ "loss": 0.0674,
+ "step": 7700
+ },
+ {
+ "epoch": 17.21,
+ "learning_rate": 1.2711055748865765e-05,
+ "loss": 0.0757,
+ "step": 7710
+ },
+ {
+ "epoch": 17.23,
+ "learning_rate": 1.2693654374805148e-05,
+ "loss": 0.0997,
+ "step": 7720
+ },
+ {
+ "epoch": 17.25,
+ "learning_rate": 1.2676244201568498e-05,
+ "loss": 0.0779,
+ "step": 7730
+ },
+ {
+ "epoch": 17.28,
+ "learning_rate": 1.2658825286028428e-05,
+ "loss": 0.0943,
+ "step": 7740
+ },
+ {
+ "epoch": 17.3,
+ "learning_rate": 1.2641397685086124e-05,
+ "loss": 0.069,
+ "step": 7750
+ },
+ {
+ "epoch": 17.32,
+ "learning_rate": 1.2623961455671125e-05,
+ "loss": 0.0745,
+ "step": 7760
+ },
+ {
+ "epoch": 17.34,
+ "learning_rate": 1.2606516654741172e-05,
+ "loss": 0.0602,
+ "step": 7770
+ },
+ {
+ "epoch": 17.37,
+ "learning_rate": 1.2589063339281995e-05,
+ "loss": 0.0584,
+ "step": 7780
+ },
+ {
+ "epoch": 17.39,
+ "learning_rate": 1.257160156630715e-05,
+ "loss": 0.2274,
+ "step": 7790
+ },
+ {
+ "epoch": 17.41,
+ "learning_rate": 1.2554131392857812e-05,
+ "loss": 0.0698,
+ "step": 7800
+ },
+ {
+ "epoch": 17.43,
+ "learning_rate": 1.253665287600259e-05,
+ "loss": 0.0643,
+ "step": 7810
+ },
+ {
+ "epoch": 17.46,
+ "learning_rate": 1.2519166072837368e-05,
+ "loss": 0.0783,
+ "step": 7820
+ },
+ {
+ "epoch": 17.48,
+ "learning_rate": 1.250167104048508e-05,
+ "loss": 0.08,
+ "step": 7830
+ },
+ {
+ "epoch": 17.5,
+ "learning_rate": 1.248416783609555e-05,
+ "loss": 0.0692,
+ "step": 7840
+ },
+ {
+ "epoch": 17.52,
+ "learning_rate": 1.2466656516845293e-05,
+ "loss": 0.1287,
+ "step": 7850
+ },
+ {
+ "epoch": 17.54,
+ "learning_rate": 1.244913713993734e-05,
+ "loss": 0.0834,
+ "step": 7860
+ },
+ {
+ "epoch": 17.57,
+ "learning_rate": 1.2431609762601036e-05,
+ "loss": 0.095,
+ "step": 7870
+ },
+ {
+ "epoch": 17.59,
+ "learning_rate": 1.241407444209186e-05,
+ "loss": 0.0925,
+ "step": 7880
+ },
+ {
+ "epoch": 17.61,
+ "learning_rate": 1.2396531235691245e-05,
+ "loss": 0.0875,
+ "step": 7890
+ },
+ {
+ "epoch": 17.63,
+ "learning_rate": 1.2378980200706376e-05,
+ "loss": 0.0761,
+ "step": 7900
+ },
+ {
+ "epoch": 17.66,
+ "learning_rate": 1.236142139447002e-05,
+ "loss": 0.0672,
+ "step": 7910
+ },
+ {
+ "epoch": 17.68,
+ "learning_rate": 1.2343854874340324e-05,
+ "loss": 0.0779,
+ "step": 7920
+ },
+ {
+ "epoch": 17.7,
+ "learning_rate": 1.2326280697700632e-05,
+ "loss": 0.0875,
+ "step": 7930
+ },
+ {
+ "epoch": 17.72,
+ "learning_rate": 1.2308698921959306e-05,
+ "loss": 0.0835,
+ "step": 7940
+ },
+ {
+ "epoch": 17.75,
+ "learning_rate": 1.2291109604549525e-05,
+ "loss": 0.0772,
+ "step": 7950
+ },
+ {
+ "epoch": 17.77,
+ "learning_rate": 1.2273512802929107e-05,
+ "loss": 0.0768,
+ "step": 7960
+ },
+ {
+ "epoch": 17.79,
+ "learning_rate": 1.2255908574580311e-05,
+ "loss": 0.0854,
+ "step": 7970
+ },
+ {
+ "epoch": 17.81,
+ "learning_rate": 1.2238296977009672e-05,
+ "loss": 0.0905,
+ "step": 7980
+ },
+ {
+ "epoch": 17.83,
+ "learning_rate": 1.2220678067747785e-05,
+ "loss": 0.0804,
+ "step": 7990
+ },
+ {
+ "epoch": 17.86,
+ "learning_rate": 1.2203051904349128e-05,
+ "loss": 0.085,
+ "step": 8000
+ },
+ {
+ "epoch": 17.88,
+ "learning_rate": 1.2185418544391885e-05,
+ "loss": 0.0767,
+ "step": 8010
+ },
+ {
+ "epoch": 17.9,
+ "learning_rate": 1.2167778045477743e-05,
+ "loss": 0.1049,
+ "step": 8020
+ },
+ {
+ "epoch": 17.92,
+ "learning_rate": 1.215013046523171e-05,
+ "loss": 0.0866,
+ "step": 8030
+ },
+ {
+ "epoch": 17.95,
+ "learning_rate": 1.2132475861301928e-05,
+ "loss": 0.0839,
+ "step": 8040
+ },
+ {
+ "epoch": 17.97,
+ "learning_rate": 1.2114814291359476e-05,
+ "loss": 0.0775,
+ "step": 8050
+ },
+ {
+ "epoch": 17.99,
+ "learning_rate": 1.20971458130982e-05,
+ "loss": 0.0958,
+ "step": 8060
+ },
+ {
+ "epoch": 18.01,
+ "learning_rate": 1.20794704842345e-05,
+ "loss": 0.0685,
+ "step": 8070
+ },
+ {
+ "epoch": 18.04,
+ "learning_rate": 1.2061788362507168e-05,
+ "loss": 0.0914,
+ "step": 8080
+ },
+ {
+ "epoch": 18.06,
+ "learning_rate": 1.204409950567717e-05,
+ "loss": 0.0608,
+ "step": 8090
+ },
+ {
+ "epoch": 18.08,
+ "learning_rate": 1.2026403971527487e-05,
+ "loss": 0.0619,
+ "step": 8100
+ },
+ {
+ "epoch": 18.1,
+ "learning_rate": 1.2008701817862906e-05,
+ "loss": 0.0703,
+ "step": 8110
+ },
+ {
+ "epoch": 18.12,
+ "learning_rate": 1.1990993102509838e-05,
+ "loss": 0.0743,
+ "step": 8120
+ },
+ {
+ "epoch": 18.15,
+ "learning_rate": 1.1973277883316128e-05,
+ "loss": 0.0661,
+ "step": 8130
+ },
+ {
+ "epoch": 18.17,
+ "learning_rate": 1.1955556218150872e-05,
+ "loss": 0.0994,
+ "step": 8140
+ },
+ {
+ "epoch": 18.19,
+ "learning_rate": 1.1937828164904216e-05,
+ "loss": 0.0508,
+ "step": 8150
+ },
+ {
+ "epoch": 18.21,
+ "learning_rate": 1.1920093781487175e-05,
+ "loss": 0.0715,
+ "step": 8160
+ },
+ {
+ "epoch": 18.24,
+ "learning_rate": 1.1902353125831441e-05,
+ "loss": 0.0693,
+ "step": 8170
+ },
+ {
+ "epoch": 18.26,
+ "learning_rate": 1.1884606255889203e-05,
+ "loss": 0.0618,
+ "step": 8180
+ },
+ {
+ "epoch": 18.28,
+ "learning_rate": 1.1866853229632942e-05,
+ "loss": 0.0942,
+ "step": 8190
+ },
+ {
+ "epoch": 18.3,
+ "learning_rate": 1.1849094105055248e-05,
+ "loss": 0.0713,
+ "step": 8200
+ },
+ {
+ "epoch": 18.33,
+ "learning_rate": 1.1831328940168638e-05,
+ "loss": 0.0668,
+ "step": 8210
+ },
+ {
+ "epoch": 18.35,
+ "learning_rate": 1.181355779300536e-05,
+ "loss": 0.0704,
+ "step": 8220
+ },
+ {
+ "epoch": 18.37,
+ "learning_rate": 1.1795780721617199e-05,
+ "loss": 0.085,
+ "step": 8230
+ },
+ {
+ "epoch": 18.39,
+ "learning_rate": 1.1777997784075294e-05,
+ "loss": 0.1076,
+ "step": 8240
+ },
+ {
+ "epoch": 18.42,
+ "learning_rate": 1.176020903846995e-05,
+ "loss": 0.0752,
+ "step": 8250
+ },
+ {
+ "epoch": 18.44,
+ "learning_rate": 1.1742414542910444e-05,
+ "loss": 0.0569,
+ "step": 8260
+ },
+ {
+ "epoch": 18.46,
+ "learning_rate": 1.1724614355524832e-05,
+ "loss": 0.0766,
+ "step": 8270
+ },
+ {
+ "epoch": 18.48,
+ "learning_rate": 1.1706808534459768e-05,
+ "loss": 0.0733,
+ "step": 8280
+ },
+ {
+ "epoch": 18.5,
+ "learning_rate": 1.16889971378803e-05,
+ "loss": 0.096,
+ "step": 8290
+ },
+ {
+ "epoch": 18.53,
+ "learning_rate": 1.1671180223969705e-05,
+ "loss": 0.0779,
+ "step": 8300
+ },
+ {
+ "epoch": 18.55,
+ "learning_rate": 1.1653357850929268e-05,
+ "loss": 0.0733,
+ "step": 8310
+ },
+ {
+ "epoch": 18.57,
+ "learning_rate": 1.1635530076978115e-05,
+ "loss": 0.1026,
+ "step": 8320
+ },
+ {
+ "epoch": 18.59,
+ "learning_rate": 1.161769696035301e-05,
+ "loss": 0.0777,
+ "step": 8330
+ },
+ {
+ "epoch": 18.62,
+ "learning_rate": 1.1599858559308175e-05,
+ "loss": 0.076,
+ "step": 8340
+ },
+ {
+ "epoch": 18.64,
+ "learning_rate": 1.158201493211509e-05,
+ "loss": 0.0722,
+ "step": 8350
+ },
+ {
+ "epoch": 18.66,
+ "learning_rate": 1.156416613706231e-05,
+ "loss": 0.0714,
+ "step": 8360
+ },
+ {
+ "epoch": 18.68,
+ "learning_rate": 1.1546312232455266e-05,
+ "loss": 0.0726,
+ "step": 8370
+ },
+ {
+ "epoch": 18.71,
+ "learning_rate": 1.152845327661609e-05,
+ "loss": 0.0811,
+ "step": 8380
+ },
+ {
+ "epoch": 18.73,
+ "learning_rate": 1.1510589327883406e-05,
+ "loss": 0.0793,
+ "step": 8390
+ },
+ {
+ "epoch": 18.75,
+ "learning_rate": 1.1492720444612148e-05,
+ "loss": 0.078,
+ "step": 8400
+ },
+ {
+ "epoch": 18.77,
+ "learning_rate": 1.1474846685173374e-05,
+ "loss": 0.0866,
+ "step": 8410
+ },
+ {
+ "epoch": 18.79,
+ "learning_rate": 1.1456968107954066e-05,
+ "loss": 0.1075,
+ "step": 8420
+ },
+ {
+ "epoch": 18.82,
+ "learning_rate": 1.143908477135695e-05,
+ "loss": 0.0777,
+ "step": 8430
+ },
+ {
+ "epoch": 18.84,
+ "learning_rate": 1.1421196733800291e-05,
+ "loss": 0.0881,
+ "step": 8440
+ },
+ {
+ "epoch": 18.86,
+ "learning_rate": 1.1403304053717719e-05,
+ "loss": 0.0817,
+ "step": 8450
+ },
+ {
+ "epoch": 18.88,
+ "learning_rate": 1.138540678955802e-05,
+ "loss": 0.0589,
+ "step": 8460
+ },
+ {
+ "epoch": 18.91,
+ "learning_rate": 1.1367504999784963e-05,
+ "loss": 0.0924,
+ "step": 8470
+ },
+ {
+ "epoch": 18.93,
+ "learning_rate": 1.1349598742877097e-05,
+ "loss": 0.0859,
+ "step": 8480
+ },
+ {
+ "epoch": 18.95,
+ "learning_rate": 1.1331688077327563e-05,
+ "loss": 0.0952,
+ "step": 8490
+ },
+ {
+ "epoch": 18.97,
+ "learning_rate": 1.1313773061643905e-05,
+ "loss": 0.0676,
+ "step": 8500
+ },
+ {
+ "epoch": 19.0,
+ "learning_rate": 1.1295853754347876e-05,
+ "loss": 0.0875,
+ "step": 8510
+ },
+ {
+ "epoch": 19.02,
+ "learning_rate": 1.1277930213975249e-05,
+ "loss": 0.0625,
+ "step": 8520
+ },
+ {
+ "epoch": 19.04,
+ "learning_rate": 1.1260002499075617e-05,
+ "loss": 0.0536,
+ "step": 8530
+ },
+ {
+ "epoch": 19.06,
+ "learning_rate": 1.1242070668212227e-05,
+ "loss": 0.0654,
+ "step": 8540
+ },
+ {
+ "epoch": 19.08,
+ "learning_rate": 1.1224134779961758e-05,
+ "loss": 0.0635,
+ "step": 8550
+ },
+ {
+ "epoch": 19.11,
+ "learning_rate": 1.1206194892914142e-05,
+ "loss": 0.0697,
+ "step": 8560
+ },
+ {
+ "epoch": 19.13,
+ "learning_rate": 1.1188251065672382e-05,
+ "loss": 0.0496,
+ "step": 8570
+ },
+ {
+ "epoch": 19.15,
+ "learning_rate": 1.117030335685235e-05,
+ "loss": 0.0623,
+ "step": 8580
+ },
+ {
+ "epoch": 19.17,
+ "learning_rate": 1.1152351825082588e-05,
+ "loss": 0.0614,
+ "step": 8590
+ },
+ {
+ "epoch": 19.2,
+ "learning_rate": 1.1134396529004143e-05,
+ "loss": 0.1199,
+ "step": 8600
+ },
+ {
+ "epoch": 19.22,
+ "learning_rate": 1.1116437527270343e-05,
+ "loss": 0.0521,
+ "step": 8610
+ },
+ {
+ "epoch": 19.24,
+ "learning_rate": 1.109847487854663e-05,
+ "loss": 0.1033,
+ "step": 8620
+ },
+ {
+ "epoch": 19.26,
+ "learning_rate": 1.1080508641510357e-05,
+ "loss": 0.0708,
+ "step": 8630
+ },
+ {
+ "epoch": 19.29,
+ "learning_rate": 1.1062538874850597e-05,
+ "loss": 0.0635,
+ "step": 8640
+ },
+ {
+ "epoch": 19.31,
+ "learning_rate": 1.1044565637267957e-05,
+ "loss": 0.0719,
+ "step": 8650
+ },
+ {
+ "epoch": 19.33,
+ "learning_rate": 1.1026588987474379e-05,
+ "loss": 0.0647,
+ "step": 8660
+ },
+ {
+ "epoch": 19.35,
+ "learning_rate": 1.100860898419295e-05,
+ "loss": 0.0678,
+ "step": 8670
+ },
+ {
+ "epoch": 19.38,
+ "learning_rate": 1.0990625686157714e-05,
+ "loss": 0.0698,
+ "step": 8680
+ },
+ {
+ "epoch": 19.4,
+ "learning_rate": 1.097263915211348e-05,
+ "loss": 0.0725,
+ "step": 8690
+ },
+ {
+ "epoch": 19.42,
+ "learning_rate": 1.0954649440815625e-05,
+ "loss": 0.1025,
+ "step": 8700
+ },
+ {
+ "epoch": 19.44,
+ "learning_rate": 1.0936656611029901e-05,
+ "loss": 0.0674,
+ "step": 8710
+ },
+ {
+ "epoch": 19.46,
+ "learning_rate": 1.091866072153226e-05,
+ "loss": 0.0844,
+ "step": 8720
+ },
+ {
+ "epoch": 19.49,
+ "learning_rate": 1.090066183110863e-05,
+ "loss": 0.0628,
+ "step": 8730
+ },
+ {
+ "epoch": 19.51,
+ "learning_rate": 1.0882659998554759e-05,
+ "loss": 0.0754,
+ "step": 8740
+ },
+ {
+ "epoch": 19.53,
+ "learning_rate": 1.0864655282675997e-05,
+ "loss": 0.0697,
+ "step": 8750
+ },
+ {
+ "epoch": 19.55,
+ "learning_rate": 1.0846647742287116e-05,
+ "loss": 0.0627,
+ "step": 8760
+ },
+ {
+ "epoch": 19.58,
+ "learning_rate": 1.0828637436212111e-05,
+ "loss": 0.0736,
+ "step": 8770
+ },
+ {
+ "epoch": 19.6,
+ "learning_rate": 1.0810624423284012e-05,
+ "loss": 0.0609,
+ "step": 8780
+ },
+ {
+ "epoch": 19.62,
+ "learning_rate": 1.07926087623447e-05,
+ "loss": 0.0724,
+ "step": 8790
+ },
+ {
+ "epoch": 19.64,
+ "learning_rate": 1.0774590512244694e-05,
+ "loss": 0.0716,
+ "step": 8800
+ },
+ {
+ "epoch": 19.67,
+ "learning_rate": 1.0756569731842978e-05,
+ "loss": 0.0834,
+ "step": 8810
+ },
+ {
+ "epoch": 19.69,
+ "learning_rate": 1.07385464800068e-05,
+ "loss": 0.065,
+ "step": 8820
+ },
+ {
+ "epoch": 19.71,
+ "learning_rate": 1.0720520815611476e-05,
+ "loss": 0.0746,
+ "step": 8830
+ },
+ {
+ "epoch": 19.73,
+ "learning_rate": 1.0702492797540214e-05,
+ "loss": 0.082,
+ "step": 8840
+ },
+ {
+ "epoch": 19.75,
+ "learning_rate": 1.06844624846839e-05,
+ "loss": 0.077,
+ "step": 8850
+ },
+ {
+ "epoch": 19.78,
+ "learning_rate": 1.0666429935940925e-05,
+ "loss": 0.0571,
+ "step": 8860
+ },
+ {
+ "epoch": 19.8,
+ "learning_rate": 1.0648395210216975e-05,
+ "loss": 0.0647,
+ "step": 8870
+ },
+ {
+ "epoch": 19.82,
+ "learning_rate": 1.0630358366424856e-05,
+ "loss": 0.0759,
+ "step": 8880
+ },
+ {
+ "epoch": 19.84,
+ "learning_rate": 1.0612319463484286e-05,
+ "loss": 0.0781,
+ "step": 8890
+ },
+ {
+ "epoch": 19.87,
+ "learning_rate": 1.0594278560321713e-05,
+ "loss": 0.0617,
+ "step": 8900
+ },
+ {
+ "epoch": 19.89,
+ "learning_rate": 1.0576235715870119e-05,
+ "loss": 0.0528,
+ "step": 8910
+ },
+ {
+ "epoch": 19.91,
+ "learning_rate": 1.0558190989068822e-05,
+ "loss": 0.0723,
+ "step": 8920
+ },
+ {
+ "epoch": 19.93,
+ "learning_rate": 1.0540144438863302e-05,
+ "loss": 0.0822,
+ "step": 8930
+ },
+ {
+ "epoch": 19.96,
+ "learning_rate": 1.052209612420498e-05,
+ "loss": 0.0633,
+ "step": 8940
+ },
+ {
+ "epoch": 19.98,
+ "learning_rate": 1.050404610405105e-05,
+ "loss": 0.0825,
+ "step": 8950
+ },
+ {
+ "epoch": 20.0,
+ "learning_rate": 1.0485994437364278e-05,
+ "loss": 0.0953,
+ "step": 8960
+ },
+ {
+ "epoch": 20.02,
+ "learning_rate": 1.0467941183112801e-05,
+ "loss": 0.0988,
+ "step": 8970
+ },
+ {
+ "epoch": 20.04,
+ "learning_rate": 1.0449886400269952e-05,
+ "loss": 0.0622,
+ "step": 8980
+ },
+ {
+ "epoch": 20.07,
+ "learning_rate": 1.0431830147814049e-05,
+ "loss": 0.0613,
+ "step": 8990
+ },
+ {
+ "epoch": 20.09,
+ "learning_rate": 1.0413772484728211e-05,
+ "loss": 0.0478,
+ "step": 9000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 5.264262028276531e+16,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s1_en/training_args.bin b/s1_en/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..963518f8cf6cc583b914262ec21d513207c0a246
--- /dev/null
+++ b/s1_en/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4786c5345d302e6b458af37541702ba130c2950bc2a2c1d2cdf2001843001b92
+size 6264
diff --git a/s1_en/zero_to_fp32.py b/s1_en/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..0e759146cadd92ddfefab3680146c2bd6a2b5c04
--- /dev/null
+++ b/s1_en/zero_to_fp32.py
@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+# python zero_to_fp32.py . output_dir/
+# or
+# python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+ buffers: dict()
+ param_shapes: dict()
+ shared_params: list
+ ds_version: int
+ frozen_param_shapes: dict()
+ frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+ return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+ '''
+ alist.sort(key=natural_keys) sorts in human order
+ http://nedbatchelder.com/blog/200712/human_sorting.html
+ (See Toothy's implementation in the comments)
+ '''
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+ if not os.path.isdir(checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+ # there should be only one file
+ if zero_stage <= 2:
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+ elif zero_stage == 3:
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+ if not os.path.exists(file):
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+ return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+ # XXX: need to test that this simple glob rule works for multi-node setup too
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+ if len(ckpt_files) == 0:
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+ return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+ zero_model_states = []
+ for file in files:
+ state_dict = torch.load(file, map_location=device, weights_only=False)
+
+ if BUFFER_NAMES not in state_dict:
+ raise ValueError(f"{file} is not a model state checkpoint")
+ buffer_names = state_dict[BUFFER_NAMES]
+ if debug:
+ print("Found buffers:", buffer_names)
+
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+ param_shapes = state_dict[PARAM_SHAPES]
+
+ # collect parameters that are included in param_shapes
+ param_names = []
+ for s in param_shapes:
+ for name in s.keys():
+ param_names.append(name)
+
+ # update with frozen parameters
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+ if frozen_param_shapes is not None:
+ if debug:
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+ param_names += list(frozen_param_shapes.keys())
+
+ # handle shared params
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+ ds_version = state_dict.get(DS_VERSION, None)
+
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+ z_model_state = zero_model_state(buffers=buffers,
+ param_shapes=param_shapes,
+ shared_params=shared_params,
+ ds_version=ds_version,
+ frozen_param_shapes=frozen_param_shapes,
+ frozen_param_fragments=frozen_param_fragments)
+ zero_model_states.append(z_model_state)
+
+ return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+ total_files = len(files)
+ state_dicts = []
+ for f in tqdm(files, desc='Loading checkpoint shards'):
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+ # and also handle the case where it was already removed by another helper script
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+ state_dicts.append(state_dict)
+
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
+ # use the max of the partition_count to get the dp world_size.
+
+ if type(world_size) is list:
+ world_size = max(world_size)
+
+ if world_size != total_files:
+ raise ValueError(
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+ )
+
+ # the groups are named differently in each stage
+ if zero_stage <= 2:
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+ elif zero_stage == 3:
+ fp32_groups_key = FP32_FLAT_GROUPS
+ else:
+ raise ValueError(f"unknown zero stage {zero_stage}")
+
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+ return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+ """
+ Returns fp32 state_dict reconstructed from ds checkpoint
+
+ Args:
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+ """
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+ optim_files = get_optim_files(ds_checkpoint_dir)
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+ model_files = get_model_state_files(ds_checkpoint_dir)
+
+ zero_model_states = parse_model_states(model_files)
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+ if zero_stage <= 2:
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+ elif zero_stage == 3:
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+ if debug:
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ state_dict[name] = frozen_param_fragments[name]
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+ attr = getattr(obj, fn, None)
+ return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+
+ # Reconstruction protocol:
+ #
+ # XXX: document this
+
+ if debug:
+ for i in range(world_size):
+ for j in range(len(fp32_flat_groups[0])):
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+ # XXX: memory usage doubles here (zero2)
+ num_param_groups = len(fp32_flat_groups[0])
+ merged_single_partition_of_fp32_groups = []
+ for i in range(num_param_groups):
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+ avail_numel = sum(
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+ if debug:
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+ # not asserting if there is a mismatch due to possible padding
+ print(f"Have {avail_numel} numels to process.")
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ total_numel = 0
+ total_params = 0
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+ offset = 0
+ avail_numel = full_single_fp32_vector.numel()
+ for name, shape in shapes.items():
+
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+ total_numel += unpartitioned_numel
+ total_params += 1
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+ offset += unpartitioned_numel
+
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+ # live optimizer object, so we are checking that the numbers are within the right range
+ align_to = 2 * world_size
+
+ def zero2_align(x):
+ return align_to * math.ceil(x / align_to)
+
+ if debug:
+ print(f"original offset={offset}, avail_numel={avail_numel}")
+
+ offset = zero2_align(offset)
+ avail_numel = zero2_align(avail_numel)
+
+ if debug:
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+ remainder = unpartitioned_numel % world_size
+ padding_numel = (world_size - remainder) if remainder else 0
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+ return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ if debug:
+ for i in range(world_size):
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+ """
+ A pseudo tensor that collects partitioned weights.
+ It is more memory efficient when there are multiple groups.
+ """
+
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+ self.flat_groups = flat_groups
+ self.flat_groups_offset = flat_groups_offset
+ self.offset = offset
+ self.partitioned_numel = partitioned_numel
+ self.shape = shape
+ self.dtype = self.flat_groups[0][0].dtype
+
+ def contiguous(self):
+ """
+ Merge partitioned weights from flat_groups into a single tensor.
+ """
+ end_idx = self.offset + self.partitioned_numel
+ world_size = len(self.flat_groups)
+ pad_flat_param_chunks = []
+
+ for rank_i in range(world_size):
+ # for each rank, we need to collect weights from related group/groups
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
+ start_group_id = None
+ end_group_id = None
+ for group_id in range(len(self.flat_groups_offset)):
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+ start_group_id = group_id
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+ end_group_id = group_id
+ break
+ # collect weights from related group/groups
+ for group_id in range(start_group_id, end_group_id + 1):
+ flat_tensor = flat_groups_at_rank_i[group_id]
+ start_offset = self.offset - self.flat_groups_offset[group_id]
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+ # collect weights from all ranks
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+ return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+ # param, re-consolidating each param, while dealing with padding if any
+
+ # merge list of dicts, preserving order
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+ if debug:
+ for i in range(world_size):
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+ wanted_params = len(param_shapes)
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+ # not asserting if there is a mismatch due to possible padding
+ avail_numel = fp32_flat_groups[0].numel() * world_size
+ print(f"Trainable params: Have {avail_numel} numels to process.")
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ offset = 0
+ total_numel = 0
+ total_params = 0
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+ total_params += 1
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ # memory efficient tensor
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+ state_dict[name] = tensor
+ offset += partitioned_numel
+
+ offset *= world_size
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+ """
+ Convert state_dict of GatheredTensor to torch tensor
+ """
+ torch_state_dict = {}
+ converted_tensors = {}
+ for name, tensor in state_dict.items():
+ tensor_id = id(tensor)
+ if tensor_id in converted_tensors: # shared tensors
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+ torch_state_dict[name] = shared_tensor
+ else:
+ converted_tensors[tensor_id] = name
+ if return_empty_tensor:
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+ else:
+ torch_state_dict[name] = tensor.contiguous()
+ return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag=None,
+ exclude_frozen_parameters=False,
+ lazy_mode=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+ via a model hub.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+ Returns:
+ - pytorch ``state_dict``
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ # do the training and checkpoint saving
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+ model = model.cpu() # move to cpu
+ model.load_state_dict(state_dict)
+ # submit to model hub or save the model to share with others
+
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
+ application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+ the checkpoint. Or you can load state_dict in lazy mode ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+ for name, lazy_tensor in state_dict.item():
+ tensor = lazy_tensor.contiguous() # to cpu
+ print(name, tensor)
+ # del tensor to release memory if it no longer in use
+ """
+ if tag is None:
+ latest_path = os.path.join(checkpoint_dir, 'latest')
+ if os.path.isfile(latest_path):
+ with open(latest_path, 'r') as fd:
+ tag = fd.read().strip()
+ else:
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+ if not os.path.isdir(ds_checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+ if lazy_mode:
+ return state_dict
+ else:
+ return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+ output_dir,
+ max_shard_size="5GB",
+ safe_serialization=False,
+ tag=None,
+ exclude_frozen_parameters=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ """
+
+ # Dependency pre-check
+ if safe_serialization:
+ try:
+ from safetensors.torch import save_file
+ except ImportError:
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
+ raise
+ if max_shard_size is not None:
+ try:
+ from huggingface_hub import split_torch_state_dict_into_shards
+ except ImportError:
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+ raise
+
+ # Convert zero checkpoint to state_dict
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag,
+ exclude_frozen_parameters,
+ lazy_mode=True)
+
+ # Shard the model if it is too big.
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+ if max_shard_size is not None:
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+ # an memory-efficient approach for sharding
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+ filename_pattern=filename_pattern,
+ max_shard_size=max_shard_size)
+ else:
+ from collections import namedtuple
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+ state_dict_split = StateDictSplit(is_sharded=False,
+ filename_to_tensors={weights_name: list(state_dict.keys())})
+
+ # Save the model by shard
+ os.makedirs(output_dir, exist_ok=True)
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+ shard_state_dict = to_torch_tensor(shard_state_dict)
+ output_path = os.path.join(output_dir, shard_file)
+ if safe_serialization:
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+ else:
+ torch.save(shard_state_dict, output_path)
+ # release the memory of current shard
+ for tensor_name in list(shard_state_dict.keys()):
+ del state_dict[tensor_name]
+ del shard_state_dict[tensor_name]
+ del shard_state_dict
+ gc.collect()
+
+ # Save index if sharded
+ if state_dict_split.is_sharded:
+ index = {
+ "metadata": state_dict_split.metadata,
+ "weight_map": state_dict_split.tensor_to_filename,
+ }
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+ save_index_file = os.path.join(output_dir, save_index_file)
+ with open(save_index_file, "w", encoding="utf-8") as f:
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+ f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+ """
+ 1. Put the provided model to cpu
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+ 3. Load it into the provided model
+
+ Args:
+ - ``model``: the model object to update
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+ Returns:
+ - ``model`: modified model
+
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+ conveniently placed for you in the checkpoint folder.
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+ # submit to model hub or save the model to share with others
+
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ """
+ logger.info(f"Extracting fp32 weights")
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+ logger.info(f"Overwriting model with fp32 weights")
+ model = model.cpu()
+ model.load_state_dict(state_dict, strict=False)
+
+ return model
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser()
+ parser.add_argument("checkpoint_dir",
+ type=str,
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+ parser.add_argument("output_dir",
+ type=str,
+ help="directory to the pytorch fp32 state_dict output files"
+ "(e.g. path/checkpoint-12-output/)")
+ parser.add_argument(
+ "--max_shard_size",
+ type=str,
+ default="5GB",
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+ "without CPU OOM issues.")
+ parser.add_argument(
+ "--safe_serialization",
+ default=False,
+ action='store_true',
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+ parser.add_argument("-t",
+ "--tag",
+ type=str,
+ default=None,
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+ args = parser.parse_args()
+
+ debug = args.debug
+
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+ args.output_dir,
+ max_shard_size=args.max_shard_size,
+ safe_serialization=args.safe_serialization,
+ tag=args.tag,
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
diff --git a/s2/README.md b/s2/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s2/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s2/adapter_config.json b/s2/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b3542974131b4b8f02885f2c26ca2523389a5828
--- /dev/null
+++ b/s2/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "gate_proj",
+ "o_proj",
+ "up_proj",
+ "down_proj",
+ "v_proj",
+ "k_proj",
+ "q_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s2/adapter_model.bin b/s2/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..3217f78c35cc7a2d1ca855f74c373ac7f1ec49c8
--- /dev/null
+++ b/s2/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:657313672dfd64ea5545a29eeda7f79040e21b724057b163e61b4f9bde5254c1
+size 639786637
diff --git a/s2/config.json b/s2/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s2/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s2/global_step6000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s2/global_step6000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..548911aeb1b6878b0726ef576ba82b535c215fdb
--- /dev/null
+++ b/s2/global_step6000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:235c3c2d5ebc5cc4e622b2f6c86c0936e6fbfe1d05cb8e7314f0d6f521c98d02
+size 4089599575
diff --git a/s2/global_step6000/mp_rank_00_model_states.pt b/s2/global_step6000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c4621b7e83eaff5d82c32740183f7d764e622a83
--- /dev/null
+++ b/s2/global_step6000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb4898e6378bd9b8e2d3817dfa868077e19aa7156a78daf0eab939a965962028
+size 28850200126
diff --git a/s2/latest b/s2/latest
new file mode 100644
index 0000000000000000000000000000000000000000..3ccb5110452b4a1eef79a1c432a18944dc25a985
--- /dev/null
+++ b/s2/latest
@@ -0,0 +1 @@
+global_step6000
\ No newline at end of file
diff --git a/s2/non_lora_trainables.bin b/s2/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..55a412581fe8bb1f2d7ffa236eb840a6b13a9716
--- /dev/null
+++ b/s2/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ca16573ba2935eb1df5756e9b23cc2487e048a9ce1e8a02e76ef2271b47feea
+size 41961191
diff --git a/s2/rng_state.pth b/s2/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..c81887d20aef8fe906571547bc7c25344cbf66f8
--- /dev/null
+++ b/s2/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58b571a0ec7a60a1d36dde4a941ac919260dbdf1599a2822ccc2feaeb81b4eee
+size 14575
diff --git a/s2/special_tokens_map.json b/s2/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s2/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s2/tokenizer.model b/s2/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s2/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s2/tokenizer_config.json b/s2/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s2/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s2/trainer_state.json b/s2/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..cbf4ee80f5fd087dbf9c372ad7ca75a5302f8fb1
--- /dev/null
+++ b/s2/trainer_state.json
@@ -0,0 +1,3616 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 13.392857142857142,
+ "global_step": 6000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 6.9406,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 6.9625,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.4781,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.2969,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 3.7406,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.3547,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 2.9609,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.9625,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.5938,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 2.2594,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 1.9625,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 1.8102,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.625,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 1.4773,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.3578,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.2027,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.0359,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 0.917,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 0.9805,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 0.7047,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 0.8602,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 0.9711,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 0.4922,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 0.6744,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 1.0547,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 0.6121,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 0.5311,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 0.4962,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 0.4607,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 0.6136,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 0.6199,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 0.5402,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 0.5375,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 0.4029,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 0.4354,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 0.7154,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 0.4764,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 0.4047,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 0.4208,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 0.3448,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 0.4429,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 0.7079,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 0.5989,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 0.4419,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 0.3883,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 0.3784,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 0.3872,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.4399,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 0.346,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 0.3869,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.3469,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.3942,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.497,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.3417,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 0.41,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.5033,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 0.4127,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.4914,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 0.2726,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.3555,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.4463,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 0.3625,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.559,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 0.2857,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.339,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.4425,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 0.386,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.2583,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.2287,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.4197,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.5672,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.4251,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 0.2164,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.4329,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.3667,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.5154,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.4191,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.4372,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.3543,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.2781,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.3889,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.3639,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 0.2839,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.3239,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.2576,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.3892,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.2446,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.3531,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.4144,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.3633,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.3056,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.2005,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.3426,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.2796,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.37,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.2958,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.2895,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.2656,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.2885,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.3175,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.2314,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.3122,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.1354,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.2356,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.2,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.2901,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.2357,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.3321,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.227,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.2477,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.2848,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.2755,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.2539,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.2153,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.3135,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.2413,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.2698,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.3131,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.2826,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.2425,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.2397,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.2258,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.2616,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.2745,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.2235,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.281,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.2768,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.1956,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.4028,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.2349,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.3584,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.2127,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.2892,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.2294,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.1796,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.1547,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.2344,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.1938,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.1642,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.2037,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.1806,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.1858,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.2068,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.1787,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.1947,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.1804,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.2003,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.141,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.1854,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.1827,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.2692,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.1865,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.1885,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.2657,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.1626,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.237,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.2337,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.2103,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.217,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.1827,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.2204,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.2232,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.1535,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.1905,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.1729,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.2792,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.2183,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.1739,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.1829,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.1702,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.2107,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.2757,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.2264,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.1797,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.1772,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.294,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.2024,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.2291,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.1425,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.1358,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.143,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.1534,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.1792,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.1508,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.1316,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.1316,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.1558,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.136,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.1113,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.1663,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.1571,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.1738,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.1491,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.206,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.1792,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.1409,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.1634,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.1211,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.2405,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.1525,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.1682,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.1604,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.1401,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.1205,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.1611,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.1259,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.1404,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.1372,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.1626,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.1819,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.1552,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.1783,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.147,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.1625,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.1435,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.1255,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.0939,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.1851,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.1553,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.1919,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.1703,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.1302,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.1564,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.152,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.1249,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.1187,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.091,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.127,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.11,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.1098,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.1008,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.1226,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.1098,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.1296,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.0975,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.1068,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.1191,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.102,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.123,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.1195,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.1326,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.1262,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.0987,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.1205,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.1237,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.1588,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.1303,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.1057,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.1137,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.1101,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.1159,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.129,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.1154,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.1108,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.1231,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.1221,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.1374,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.1277,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.1126,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.1074,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.1211,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.124,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.122,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.1246,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.1281,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.1117,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.1462,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.093,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.1543,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.0921,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.1139,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.1115,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.0748,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.1428,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.1,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.1159,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.1254,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.0951,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.0798,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.0978,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.1121,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.0811,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.0841,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.1058,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.1103,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.1101,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.0994,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.1068,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.0887,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.0963,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.1173,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.1129,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.1068,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.0932,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.0824,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.0976,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.1071,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.1087,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.1006,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.1139,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.1065,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.1183,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.1106,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.1263,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.0902,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.1197,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.1006,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.1146,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.1128,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.0985,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.091,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.0872,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.0998,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.0923,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.071,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.0893,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.0929,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.0892,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.0739,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.0783,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.0783,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.0847,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.0886,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.088,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.082,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.0779,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.0926,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.1093,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.1051,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.0914,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.0837,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.0982,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.0883,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.0949,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.0872,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.0851,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.1099,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.0928,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.0986,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.067,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.0957,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.0998,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.0705,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.1001,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.0987,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.0999,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.0924,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.0821,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.0928,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.0997,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.0995,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.1074,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.0789,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.1031,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.0861,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.1113,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.0801,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.1016,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.0901,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.0762,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.0881,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.0787,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.0877,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.0935,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.0776,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.0814,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.0736,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.0783,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.0766,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.0738,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.0875,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.0804,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.0701,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.0844,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.0812,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.0979,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.0721,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.0816,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.0714,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.0623,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.0618,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.0842,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.0792,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.0794,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.0866,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.0755,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.0825,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.0842,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.0776,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.097,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.094,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.0982,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.0863,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.0702,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.0982,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.0843,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.0759,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.0956,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.0852,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.0903,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.0797,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.0898,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.079,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.1005,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.0708,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.0667,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.06,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.074,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.0723,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.0722,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.0655,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.0819,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.0928,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.0822,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.0908,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.0813,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.0859,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.0794,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.0855,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.0984,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.0797,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.0772,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.0666,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.0728,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.0825,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.0698,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.074,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.0862,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.0768,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.0814,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.0829,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.0844,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.0803,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.0819,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.0692,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.0777,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.089,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.092,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.1097,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.0755,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.0787,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.0708,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.0778,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.0725,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.077,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.0809,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.0978,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.0759,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.0678,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.0706,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.0665,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.0575,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.0684,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.0869,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.0648,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.0673,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.0677,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.0687,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.0701,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.0645,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.084,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.0649,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.0762,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.0827,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.0691,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.0816,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.0832,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.0778,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.0765,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.0632,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.0822,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.0615,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.0701,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.0682,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.0991,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.0764,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.0706,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.0863,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.0813,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.0847,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.0805,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.0877,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.0713,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.0776,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.0685,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.0825,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.0747,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.086,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.0729,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.0688,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.0755,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.0877,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.0785,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.0587,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.0584,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.0641,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.0678,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.0597,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.0667,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.0651,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.0648,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.0671,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.0708,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.084,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.0592,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.0681,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.0765,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.0853,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.0756,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.0922,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.0704,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.0697,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.0698,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.0672,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.0586,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.073,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.0759,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.0817,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.0642,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.0669,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.0647,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.0711,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.0764,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.069,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.0733,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.0714,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.0848,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.0675,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.0739,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.085,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.0685,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.0852,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.0655,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.0669,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.0915,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.072,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.0792,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.0632,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.0591,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.0628,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.0641,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.0625,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.0632,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.0757,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.0576,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.0825,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.0584,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.0748,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.0755,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.068,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.0621,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.06,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.0608,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.074,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.0655,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0627,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.0769,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0869,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.0603,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.0733,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.062,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.0846,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.068,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.0653,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.0623,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.0627,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.0644,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.0594,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.0666,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.0626,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.0587,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.0645,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.0907,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.059,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.0696,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.083,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.0683,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.0665,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.0703,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.0663,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.0764,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.0664,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.065,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.0622,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.059,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.0547,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.0512,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.0548,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.0645,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.0667,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.0612,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.0558,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.0833,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.0526,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.0816,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.0648,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.0613,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.0567,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0538,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.0582,
+ "step": 6000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 5.287881617912627e+16,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s2/training_args.bin b/s2/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..f7c81f651c6227a959d59c69b506bf884384e3fc
--- /dev/null
+++ b/s2/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e58f3087a2c566412b2d6ccff502b7548a8643944ffa0bb008f5e780246067ae
+size 5819
diff --git a/s2/zero_to_fp32.py b/s2/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..c5246ff52274e1d6142001ccf085186d3545ce57
--- /dev/null
+++ b/s2/zero_to_fp32.py
@@ -0,0 +1,578 @@
+#!/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 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_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_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 _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()
+ 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("-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)
diff --git a/s2_en/README.md b/s2_en/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s2_en/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s2_en/adapter_config.json b/s2_en/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..f33773d9e4a506301f3a455aefd54ea4cdd6e8c0
--- /dev/null
+++ b/s2_en/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "k_proj",
+ "up_proj",
+ "down_proj",
+ "v_proj",
+ "o_proj",
+ "q_proj",
+ "gate_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s2_en/adapter_model.bin b/s2_en/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..76f85bf226fd0162e22cd343dffeccb5a1db95ea
--- /dev/null
+++ b/s2_en/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:422c0b3bfbe12adcec40ae8e0a603d2bea66f01127e811eb210d95159de1bf73
+size 639787082
diff --git a/s2_en/config.json b/s2_en/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s2_en/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s2_en/global_step11000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s2_en/global_step11000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..67bcf56f3162ee80157c79a9a537c6a207fc9855
--- /dev/null
+++ b/s2_en/global_step11000/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:ed9fd169999c88b798a023cbc5a6bff9c471a528683ee25ba2eb508c9009582e
+size 4089600080
diff --git a/s2_en/global_step11000/mp_rank_00_model_states.pt b/s2_en/global_step11000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c1c9438f8b1848cd5ecaeddffed05d87ed43398e
--- /dev/null
+++ b/s2_en/global_step11000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cea6959d4621df1a43c9037dc46cf2843a03cb83e0d35924bb817a270e4d7e3e
+size 28850200603
diff --git a/s2_en/latest b/s2_en/latest
new file mode 100644
index 0000000000000000000000000000000000000000..2b8686c7ed8bb8587fb5bfce4b266e4264df02e6
--- /dev/null
+++ b/s2_en/latest
@@ -0,0 +1 @@
+global_step11000
\ No newline at end of file
diff --git a/s2_en/non_lora_trainables.bin b/s2_en/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..0740879f1d1568a59084b2ad009d141bd66e680b
--- /dev/null
+++ b/s2_en/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:490f8fa67c3424a6208c09d6c57f97199a1924bf1b290b212ec7b6cdded6b252
+size 41961648
diff --git a/s2_en/rng_state.pth b/s2_en/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..3803a49c9028b35771d92e33ef4d5ec2f9f2080a
--- /dev/null
+++ b/s2_en/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f103970eca31341cb0472268569b3d2f277aa7c323fe83e03b8f2aec4ec5dde
+size 14244
diff --git a/s2_en/special_tokens_map.json b/s2_en/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s2_en/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s2_en/tokenizer.model b/s2_en/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s2_en/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s2_en/tokenizer_config.json b/s2_en/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s2_en/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s2_en/trainer_state.json b/s2_en/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..cc91ca937ec34bdf8d943f1ec2842281b0d23053
--- /dev/null
+++ b/s2_en/trainer_state.json
@@ -0,0 +1,6616 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 24.553571428571427,
+ "global_step": 11000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 5.4969,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 5.7625,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.3281,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.8969,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 4.0547,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.3922,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 3.0609,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.8453,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.3508,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 2.4211,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 2.2891,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 2.0117,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.7234,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 2.175,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.2992,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.2973,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.5531,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 1.25,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 1.1742,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 1.3242,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 1.234,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 1.1578,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 1.2152,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 1.2215,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 1.4211,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 1.0383,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 1.0629,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 1.3832,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 1.0078,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 1.8035,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 1.1969,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 0.7822,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 0.943,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 0.9402,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 1.0994,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 1.0188,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 1.2152,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 1.05,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 1.0902,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 0.9207,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 1.3006,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 0.7423,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 0.7264,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 0.7605,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 1.0223,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 0.6748,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 0.8947,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.6562,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 1.0439,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 1.1691,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.8621,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.818,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.6332,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.8918,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 0.8881,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.8555,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 0.7271,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.9385,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 0.8719,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.703,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.6535,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 0.8262,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.783,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 0.6609,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.7942,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.8227,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 1.009,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.9586,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.7836,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.8104,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.6854,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.7107,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 0.6771,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.8244,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.6863,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.793,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.8221,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.7005,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.7057,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.8062,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.9057,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.7797,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 0.9398,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.6892,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.6339,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.7377,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.8047,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.6106,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.7189,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.7957,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.5467,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.7066,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.5595,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.4314,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.3828,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.7207,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.5498,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.5375,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.6203,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.6041,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.5675,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.5832,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.5791,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.6242,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.4392,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.6139,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.401,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.5512,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.603,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.5771,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.4462,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.576,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.5742,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.4595,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.4734,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.5546,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.4874,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.4997,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.5466,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.5512,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.3668,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.4967,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.4352,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.5797,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.3668,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.3524,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.7638,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.5214,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.5143,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.5393,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.7084,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.4321,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.6334,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.7473,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.4093,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.3285,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.5238,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.4093,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.3202,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.3607,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.3255,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.3821,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.43,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.3271,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.5113,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.3202,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.3178,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.4218,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.3942,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.4392,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.4372,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.3706,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.4427,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.3347,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.2392,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.353,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.4637,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.3287,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.382,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.3413,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.3233,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.2846,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.3512,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.4272,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.3992,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.407,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.4494,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.2791,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.4276,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.3395,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.3548,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.3003,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.415,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.3812,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.361,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.4472,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.3828,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.2568,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.4671,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.4138,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.2093,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.1923,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.2213,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.2253,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.3045,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.2749,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.2329,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.2785,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.2412,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.2403,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.2864,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.2575,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.2621,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.2937,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.2597,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.2299,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.3193,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.2214,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.2271,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.2829,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.3148,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.2828,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.2846,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.2779,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.2877,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.3089,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.3351,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.2175,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.3142,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.2378,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.3042,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.2508,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.1733,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.2595,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.3652,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.3227,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.2818,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.3277,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.2576,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.3362,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.3035,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.2588,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.2766,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.2879,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.1825,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.2012,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.2014,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.1917,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.1853,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.1726,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.1738,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.2221,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.152,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.1959,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.221,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.2443,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.269,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.2092,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.2421,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.2037,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.3727,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.1812,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.1861,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.1987,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.2456,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.1986,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.1954,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.2107,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.2205,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.1677,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.1757,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.1849,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.2591,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.2387,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.1754,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.1956,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.199,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.195,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.27,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.2301,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.2423,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.2396,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.209,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.2518,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.2025,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.1896,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.2762,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.2312,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.2104,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.1653,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.2042,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.1439,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.1415,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.1458,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.1255,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.1407,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.1808,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.1939,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.1663,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.2042,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.2018,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.1778,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.1583,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.1483,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.2115,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.1656,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.1889,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.1656,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.1844,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.1918,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.1731,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.1824,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.1369,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.2788,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.1722,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.1454,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.1644,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.2158,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.1916,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.2604,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.2261,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.1846,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.2215,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.1609,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.1814,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.1731,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.1599,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.1508,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.1586,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.1836,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.1639,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.1322,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.1665,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.2296,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.166,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.1786,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.1708,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.1554,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.1067,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.1548,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.1609,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.1485,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.1499,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.1625,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.2173,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.1799,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.1554,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.1544,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.1347,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.1699,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.1382,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.1356,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.134,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.1597,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.1383,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.214,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.1322,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.1664,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.1222,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.1477,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.1433,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.114,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.16,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.143,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.1586,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.2475,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.1663,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.1411,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.156,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.2059,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.1535,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.1605,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.153,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.1622,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.1757,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.1776,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.1494,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.138,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.1342,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.1856,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.1542,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.132,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.1367,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.1024,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.1216,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.1546,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.1421,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.1191,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.1476,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.1406,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.1636,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.1358,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.1187,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.1813,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.2404,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.1502,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.1166,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.1509,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.1222,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.1202,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.1354,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.1861,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.1477,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.1162,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.1183,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.168,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.1765,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.1573,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.1615,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.1164,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.1042,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.1322,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.1346,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.1488,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.15,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.1608,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.129,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.1404,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.1542,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.1611,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.1089,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.1659,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.2411,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.1263,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.1005,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.1692,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.0977,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.0963,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.1449,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.1491,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.0934,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.1384,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.167,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.1378,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.1286,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.1483,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.1236,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.1181,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.1946,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.1184,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.1348,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.1342,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.1261,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.1277,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.1325,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.0981,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.137,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.103,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.1506,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.134,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.1274,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.1171,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.128,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.1936,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.1356,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.1126,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.1356,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.135,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.0998,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.1443,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.1485,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.1455,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.1356,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.1497,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.1463,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.1413,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.1611,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.1405,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.0849,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.1303,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.126,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.0996,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.0939,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.133,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.1169,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.1122,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.1198,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.0999,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.1395,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.099,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.165,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.0973,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.1153,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.1615,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.1608,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.1321,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.1102,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.1674,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.155,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.1256,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.1126,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.1321,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.1292,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.0945,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.1223,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.1241,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.1155,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.1248,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.1301,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.1341,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.1007,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.1041,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.1271,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.1184,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.1254,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.1574,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.1189,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.1134,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.1508,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.1292,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.0917,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.1032,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.0984,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.1162,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.088,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.0844,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.153,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.1279,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.1098,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.0935,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.1088,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.1226,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.1202,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.0983,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.1028,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.0997,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.1018,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.1135,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.1053,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.1099,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.1319,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.11,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.1343,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.1054,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.1285,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.1402,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.1584,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.1066,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.1021,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.0997,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.1108,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.1394,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.133,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.1151,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.0964,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.103,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.1951,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.1511,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.1219,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.1096,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.0845,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.1231,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.1064,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.1247,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.1105,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.1168,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.1313,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.1124,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.0774,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.1125,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.0943,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.0996,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.0839,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.1085,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.0751,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.0946,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.0975,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.1105,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.0876,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.1014,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.087,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.1555,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.0878,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.1181,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.1031,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0998,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.0926,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0829,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.1246,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.08,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.1063,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.175,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.1021,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.1085,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.1007,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.107,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.1331,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.1165,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.1051,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.1058,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.1249,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.1124,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.0974,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.1393,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.0852,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.1135,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.1014,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.0875,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.1782,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.1207,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.1528,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.1701,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.0959,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.0841,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.1153,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.101,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.0999,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.1071,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.101,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.107,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.0975,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.0876,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.1238,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.1028,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.0883,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.0945,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.1148,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.1207,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0874,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.0927,
+ "step": 6000
+ },
+ {
+ "epoch": 13.42,
+ "learning_rate": 1.5495257562515308e-05,
+ "loss": 0.0971,
+ "step": 6010
+ },
+ {
+ "epoch": 13.44,
+ "learning_rate": 1.5480148332320562e-05,
+ "loss": 0.0997,
+ "step": 6020
+ },
+ {
+ "epoch": 13.46,
+ "learning_rate": 1.5465021200501046e-05,
+ "loss": 0.0915,
+ "step": 6030
+ },
+ {
+ "epoch": 13.48,
+ "learning_rate": 1.5449876216471525e-05,
+ "loss": 0.0958,
+ "step": 6040
+ },
+ {
+ "epoch": 13.5,
+ "learning_rate": 1.5434713429705078e-05,
+ "loss": 0.1452,
+ "step": 6050
+ },
+ {
+ "epoch": 13.53,
+ "learning_rate": 1.5419532889732943e-05,
+ "loss": 0.0934,
+ "step": 6060
+ },
+ {
+ "epoch": 13.55,
+ "learning_rate": 1.540433464614435e-05,
+ "loss": 0.0974,
+ "step": 6070
+ },
+ {
+ "epoch": 13.57,
+ "learning_rate": 1.5389118748586357e-05,
+ "loss": 0.1125,
+ "step": 6080
+ },
+ {
+ "epoch": 13.59,
+ "learning_rate": 1.537388524676369e-05,
+ "loss": 0.1034,
+ "step": 6090
+ },
+ {
+ "epoch": 13.62,
+ "learning_rate": 1.5358634190438592e-05,
+ "loss": 0.1037,
+ "step": 6100
+ },
+ {
+ "epoch": 13.64,
+ "learning_rate": 1.5343365629430638e-05,
+ "loss": 0.0932,
+ "step": 6110
+ },
+ {
+ "epoch": 13.66,
+ "learning_rate": 1.5328079613616592e-05,
+ "loss": 0.0992,
+ "step": 6120
+ },
+ {
+ "epoch": 13.68,
+ "learning_rate": 1.531277619293023e-05,
+ "loss": 0.105,
+ "step": 6130
+ },
+ {
+ "epoch": 13.71,
+ "learning_rate": 1.5297455417362194e-05,
+ "loss": 0.0902,
+ "step": 6140
+ },
+ {
+ "epoch": 13.73,
+ "learning_rate": 1.52821173369598e-05,
+ "loss": 0.1047,
+ "step": 6150
+ },
+ {
+ "epoch": 13.75,
+ "learning_rate": 1.526676200182691e-05,
+ "loss": 0.0969,
+ "step": 6160
+ },
+ {
+ "epoch": 13.77,
+ "learning_rate": 1.5251389462123748e-05,
+ "loss": 0.0982,
+ "step": 6170
+ },
+ {
+ "epoch": 13.79,
+ "learning_rate": 1.5235999768066729e-05,
+ "loss": 0.0928,
+ "step": 6180
+ },
+ {
+ "epoch": 13.82,
+ "learning_rate": 1.5220592969928313e-05,
+ "loss": 0.1082,
+ "step": 6190
+ },
+ {
+ "epoch": 13.84,
+ "learning_rate": 1.5205169118036831e-05,
+ "loss": 0.082,
+ "step": 6200
+ },
+ {
+ "epoch": 13.86,
+ "learning_rate": 1.5189728262776325e-05,
+ "loss": 0.0939,
+ "step": 6210
+ },
+ {
+ "epoch": 13.88,
+ "learning_rate": 1.5174270454586375e-05,
+ "loss": 0.1805,
+ "step": 6220
+ },
+ {
+ "epoch": 13.91,
+ "learning_rate": 1.5158795743961942e-05,
+ "loss": 0.1409,
+ "step": 6230
+ },
+ {
+ "epoch": 13.93,
+ "learning_rate": 1.5143304181453204e-05,
+ "loss": 0.1409,
+ "step": 6240
+ },
+ {
+ "epoch": 13.95,
+ "learning_rate": 1.5127795817665389e-05,
+ "loss": 0.1206,
+ "step": 6250
+ },
+ {
+ "epoch": 13.97,
+ "learning_rate": 1.5112270703258602e-05,
+ "loss": 0.1379,
+ "step": 6260
+ },
+ {
+ "epoch": 14.0,
+ "learning_rate": 1.5096728888947669e-05,
+ "loss": 0.0794,
+ "step": 6270
+ },
+ {
+ "epoch": 14.02,
+ "learning_rate": 1.508117042550197e-05,
+ "loss": 0.1078,
+ "step": 6280
+ },
+ {
+ "epoch": 14.04,
+ "learning_rate": 1.5065595363745272e-05,
+ "loss": 0.0941,
+ "step": 6290
+ },
+ {
+ "epoch": 14.06,
+ "learning_rate": 1.505000375455556e-05,
+ "loss": 0.0893,
+ "step": 6300
+ },
+ {
+ "epoch": 14.08,
+ "learning_rate": 1.503439564886487e-05,
+ "loss": 0.1003,
+ "step": 6310
+ },
+ {
+ "epoch": 14.11,
+ "learning_rate": 1.501877109765914e-05,
+ "loss": 0.0851,
+ "step": 6320
+ },
+ {
+ "epoch": 14.13,
+ "learning_rate": 1.5003130151978012e-05,
+ "loss": 0.0819,
+ "step": 6330
+ },
+ {
+ "epoch": 14.15,
+ "learning_rate": 1.4987472862914697e-05,
+ "loss": 0.0819,
+ "step": 6340
+ },
+ {
+ "epoch": 14.17,
+ "learning_rate": 1.4971799281615782e-05,
+ "loss": 0.0852,
+ "step": 6350
+ },
+ {
+ "epoch": 14.2,
+ "learning_rate": 1.4956109459281083e-05,
+ "loss": 0.0748,
+ "step": 6360
+ },
+ {
+ "epoch": 14.22,
+ "learning_rate": 1.4940403447163467e-05,
+ "loss": 0.1047,
+ "step": 6370
+ },
+ {
+ "epoch": 14.24,
+ "learning_rate": 1.4924681296568689e-05,
+ "loss": 0.0865,
+ "step": 6380
+ },
+ {
+ "epoch": 14.26,
+ "learning_rate": 1.4908943058855213e-05,
+ "loss": 0.1038,
+ "step": 6390
+ },
+ {
+ "epoch": 14.29,
+ "learning_rate": 1.4893188785434067e-05,
+ "loss": 0.1166,
+ "step": 6400
+ },
+ {
+ "epoch": 14.31,
+ "learning_rate": 1.4877418527768654e-05,
+ "loss": 0.1093,
+ "step": 6410
+ },
+ {
+ "epoch": 14.33,
+ "learning_rate": 1.4861632337374596e-05,
+ "loss": 0.0827,
+ "step": 6420
+ },
+ {
+ "epoch": 14.35,
+ "learning_rate": 1.4845830265819552e-05,
+ "loss": 0.1074,
+ "step": 6430
+ },
+ {
+ "epoch": 14.38,
+ "learning_rate": 1.483001236472307e-05,
+ "loss": 0.0929,
+ "step": 6440
+ },
+ {
+ "epoch": 14.4,
+ "learning_rate": 1.4814178685756405e-05,
+ "loss": 0.083,
+ "step": 6450
+ },
+ {
+ "epoch": 14.42,
+ "learning_rate": 1.4798329280642345e-05,
+ "loss": 0.0842,
+ "step": 6460
+ },
+ {
+ "epoch": 14.44,
+ "learning_rate": 1.4782464201155057e-05,
+ "loss": 0.0988,
+ "step": 6470
+ },
+ {
+ "epoch": 14.46,
+ "learning_rate": 1.476658349911991e-05,
+ "loss": 0.0776,
+ "step": 6480
+ },
+ {
+ "epoch": 14.49,
+ "learning_rate": 1.4750687226413305e-05,
+ "loss": 0.0863,
+ "step": 6490
+ },
+ {
+ "epoch": 14.51,
+ "learning_rate": 1.4734775434962504e-05,
+ "loss": 0.0995,
+ "step": 6500
+ },
+ {
+ "epoch": 14.53,
+ "learning_rate": 1.471884817674546e-05,
+ "loss": 0.0917,
+ "step": 6510
+ },
+ {
+ "epoch": 14.55,
+ "learning_rate": 1.4702905503790668e-05,
+ "loss": 0.2156,
+ "step": 6520
+ },
+ {
+ "epoch": 14.58,
+ "learning_rate": 1.4686947468176955e-05,
+ "loss": 0.084,
+ "step": 6530
+ },
+ {
+ "epoch": 14.6,
+ "learning_rate": 1.467097412203334e-05,
+ "loss": 0.1124,
+ "step": 6540
+ },
+ {
+ "epoch": 14.62,
+ "learning_rate": 1.4654985517538864e-05,
+ "loss": 0.1854,
+ "step": 6550
+ },
+ {
+ "epoch": 14.64,
+ "learning_rate": 1.4638981706922401e-05,
+ "loss": 0.0905,
+ "step": 6560
+ },
+ {
+ "epoch": 14.67,
+ "learning_rate": 1.4622962742462503e-05,
+ "loss": 0.0947,
+ "step": 6570
+ },
+ {
+ "epoch": 14.69,
+ "learning_rate": 1.4606928676487223e-05,
+ "loss": 0.1043,
+ "step": 6580
+ },
+ {
+ "epoch": 14.71,
+ "learning_rate": 1.459087956137394e-05,
+ "loss": 0.1081,
+ "step": 6590
+ },
+ {
+ "epoch": 14.73,
+ "learning_rate": 1.4574815449549209e-05,
+ "loss": 0.0771,
+ "step": 6600
+ },
+ {
+ "epoch": 14.75,
+ "learning_rate": 1.4558736393488553e-05,
+ "loss": 0.1048,
+ "step": 6610
+ },
+ {
+ "epoch": 14.78,
+ "learning_rate": 1.4542642445716326e-05,
+ "loss": 0.1048,
+ "step": 6620
+ },
+ {
+ "epoch": 14.8,
+ "learning_rate": 1.4526533658805517e-05,
+ "loss": 0.0987,
+ "step": 6630
+ },
+ {
+ "epoch": 14.82,
+ "learning_rate": 1.4510410085377606e-05,
+ "loss": 0.1066,
+ "step": 6640
+ },
+ {
+ "epoch": 14.84,
+ "learning_rate": 1.4494271778102358e-05,
+ "loss": 0.0792,
+ "step": 6650
+ },
+ {
+ "epoch": 14.87,
+ "learning_rate": 1.4478118789697675e-05,
+ "loss": 0.0925,
+ "step": 6660
+ },
+ {
+ "epoch": 14.89,
+ "learning_rate": 1.4461951172929419e-05,
+ "loss": 0.1078,
+ "step": 6670
+ },
+ {
+ "epoch": 14.91,
+ "learning_rate": 1.4445768980611233e-05,
+ "loss": 0.0942,
+ "step": 6680
+ },
+ {
+ "epoch": 14.93,
+ "learning_rate": 1.4429572265604375e-05,
+ "loss": 0.0829,
+ "step": 6690
+ },
+ {
+ "epoch": 14.96,
+ "learning_rate": 1.4413361080817545e-05,
+ "loss": 0.1201,
+ "step": 6700
+ },
+ {
+ "epoch": 14.98,
+ "learning_rate": 1.4397135479206705e-05,
+ "loss": 0.1201,
+ "step": 6710
+ },
+ {
+ "epoch": 15.0,
+ "learning_rate": 1.4380895513774922e-05,
+ "loss": 0.0985,
+ "step": 6720
+ },
+ {
+ "epoch": 15.02,
+ "learning_rate": 1.436464123757217e-05,
+ "loss": 0.068,
+ "step": 6730
+ },
+ {
+ "epoch": 15.04,
+ "learning_rate": 1.4348372703695184e-05,
+ "loss": 0.0923,
+ "step": 6740
+ },
+ {
+ "epoch": 15.07,
+ "learning_rate": 1.4332089965287266e-05,
+ "loss": 0.0808,
+ "step": 6750
+ },
+ {
+ "epoch": 15.09,
+ "learning_rate": 1.431579307553812e-05,
+ "loss": 0.079,
+ "step": 6760
+ },
+ {
+ "epoch": 15.11,
+ "learning_rate": 1.429948208768368e-05,
+ "loss": 0.074,
+ "step": 6770
+ },
+ {
+ "epoch": 15.13,
+ "learning_rate": 1.4283157055005928e-05,
+ "loss": 0.0812,
+ "step": 6780
+ },
+ {
+ "epoch": 15.16,
+ "learning_rate": 1.4266818030832732e-05,
+ "loss": 0.0744,
+ "step": 6790
+ },
+ {
+ "epoch": 15.18,
+ "learning_rate": 1.4250465068537664e-05,
+ "loss": 0.0786,
+ "step": 6800
+ },
+ {
+ "epoch": 15.2,
+ "learning_rate": 1.4234098221539818e-05,
+ "loss": 0.08,
+ "step": 6810
+ },
+ {
+ "epoch": 15.22,
+ "learning_rate": 1.4217717543303657e-05,
+ "loss": 0.1014,
+ "step": 6820
+ },
+ {
+ "epoch": 15.25,
+ "learning_rate": 1.4201323087338816e-05,
+ "loss": 0.0751,
+ "step": 6830
+ },
+ {
+ "epoch": 15.27,
+ "learning_rate": 1.4184914907199942e-05,
+ "loss": 0.0937,
+ "step": 6840
+ },
+ {
+ "epoch": 15.29,
+ "learning_rate": 1.4168493056486512e-05,
+ "loss": 0.0852,
+ "step": 6850
+ },
+ {
+ "epoch": 15.31,
+ "learning_rate": 1.4152057588842657e-05,
+ "loss": 0.0896,
+ "step": 6860
+ },
+ {
+ "epoch": 15.33,
+ "learning_rate": 1.4135608557956992e-05,
+ "loss": 0.1104,
+ "step": 6870
+ },
+ {
+ "epoch": 15.36,
+ "learning_rate": 1.4119146017562441e-05,
+ "loss": 0.0914,
+ "step": 6880
+ },
+ {
+ "epoch": 15.38,
+ "learning_rate": 1.4102670021436059e-05,
+ "loss": 0.075,
+ "step": 6890
+ },
+ {
+ "epoch": 15.4,
+ "learning_rate": 1.4086180623398842e-05,
+ "loss": 0.0746,
+ "step": 6900
+ },
+ {
+ "epoch": 15.42,
+ "learning_rate": 1.4069677877315587e-05,
+ "loss": 0.1284,
+ "step": 6910
+ },
+ {
+ "epoch": 15.45,
+ "learning_rate": 1.4053161837094675e-05,
+ "loss": 0.0999,
+ "step": 6920
+ },
+ {
+ "epoch": 15.47,
+ "learning_rate": 1.4036632556687927e-05,
+ "loss": 0.094,
+ "step": 6930
+ },
+ {
+ "epoch": 15.49,
+ "learning_rate": 1.4020090090090408e-05,
+ "loss": 0.1239,
+ "step": 6940
+ },
+ {
+ "epoch": 15.51,
+ "learning_rate": 1.4003534491340259e-05,
+ "loss": 0.1013,
+ "step": 6950
+ },
+ {
+ "epoch": 15.54,
+ "learning_rate": 1.3986965814518521e-05,
+ "loss": 0.1134,
+ "step": 6960
+ },
+ {
+ "epoch": 15.56,
+ "learning_rate": 1.3970384113748951e-05,
+ "loss": 0.0859,
+ "step": 6970
+ },
+ {
+ "epoch": 15.58,
+ "learning_rate": 1.3953789443197857e-05,
+ "loss": 0.1392,
+ "step": 6980
+ },
+ {
+ "epoch": 15.6,
+ "learning_rate": 1.3937181857073912e-05,
+ "loss": 0.0904,
+ "step": 6990
+ },
+ {
+ "epoch": 15.62,
+ "learning_rate": 1.3920561409627974e-05,
+ "loss": 0.0792,
+ "step": 7000
+ },
+ {
+ "epoch": 15.65,
+ "learning_rate": 1.3903928155152926e-05,
+ "loss": 0.0778,
+ "step": 7010
+ },
+ {
+ "epoch": 15.67,
+ "learning_rate": 1.3887282147983472e-05,
+ "loss": 0.0995,
+ "step": 7020
+ },
+ {
+ "epoch": 15.69,
+ "learning_rate": 1.3870623442495987e-05,
+ "loss": 0.1112,
+ "step": 7030
+ },
+ {
+ "epoch": 15.71,
+ "learning_rate": 1.3853952093108323e-05,
+ "loss": 0.088,
+ "step": 7040
+ },
+ {
+ "epoch": 15.74,
+ "learning_rate": 1.3837268154279628e-05,
+ "loss": 0.0833,
+ "step": 7050
+ },
+ {
+ "epoch": 15.76,
+ "learning_rate": 1.3820571680510187e-05,
+ "loss": 0.1009,
+ "step": 7060
+ },
+ {
+ "epoch": 15.78,
+ "learning_rate": 1.3803862726341224e-05,
+ "loss": 0.1223,
+ "step": 7070
+ },
+ {
+ "epoch": 15.8,
+ "learning_rate": 1.3787141346354733e-05,
+ "loss": 0.1115,
+ "step": 7080
+ },
+ {
+ "epoch": 15.83,
+ "learning_rate": 1.3770407595173301e-05,
+ "loss": 0.0905,
+ "step": 7090
+ },
+ {
+ "epoch": 15.85,
+ "learning_rate": 1.375366152745992e-05,
+ "loss": 0.0843,
+ "step": 7100
+ },
+ {
+ "epoch": 15.87,
+ "learning_rate": 1.373690319791783e-05,
+ "loss": 0.0878,
+ "step": 7110
+ },
+ {
+ "epoch": 15.89,
+ "learning_rate": 1.3720132661290311e-05,
+ "loss": 0.1105,
+ "step": 7120
+ },
+ {
+ "epoch": 15.92,
+ "learning_rate": 1.3703349972360527e-05,
+ "loss": 0.0942,
+ "step": 7130
+ },
+ {
+ "epoch": 15.94,
+ "learning_rate": 1.3686555185951334e-05,
+ "loss": 0.1013,
+ "step": 7140
+ },
+ {
+ "epoch": 15.96,
+ "learning_rate": 1.3669748356925112e-05,
+ "loss": 0.1016,
+ "step": 7150
+ },
+ {
+ "epoch": 15.98,
+ "learning_rate": 1.3652929540183578e-05,
+ "loss": 0.0893,
+ "step": 7160
+ },
+ {
+ "epoch": 16.0,
+ "learning_rate": 1.3636098790667605e-05,
+ "loss": 0.0836,
+ "step": 7170
+ },
+ {
+ "epoch": 16.03,
+ "learning_rate": 1.3619256163357046e-05,
+ "loss": 0.0896,
+ "step": 7180
+ },
+ {
+ "epoch": 16.05,
+ "learning_rate": 1.3602401713270566e-05,
+ "loss": 0.0786,
+ "step": 7190
+ },
+ {
+ "epoch": 16.07,
+ "learning_rate": 1.3585535495465432e-05,
+ "loss": 0.0731,
+ "step": 7200
+ },
+ {
+ "epoch": 16.09,
+ "learning_rate": 1.3568657565037365e-05,
+ "loss": 0.1016,
+ "step": 7210
+ },
+ {
+ "epoch": 16.12,
+ "learning_rate": 1.3551767977120341e-05,
+ "loss": 0.0826,
+ "step": 7220
+ },
+ {
+ "epoch": 16.14,
+ "learning_rate": 1.353486678688642e-05,
+ "loss": 0.0733,
+ "step": 7230
+ },
+ {
+ "epoch": 16.16,
+ "learning_rate": 1.351795404954556e-05,
+ "loss": 0.0677,
+ "step": 7240
+ },
+ {
+ "epoch": 16.18,
+ "learning_rate": 1.3501029820345446e-05,
+ "loss": 0.0948,
+ "step": 7250
+ },
+ {
+ "epoch": 16.21,
+ "learning_rate": 1.3484094154571286e-05,
+ "loss": 0.1281,
+ "step": 7260
+ },
+ {
+ "epoch": 16.23,
+ "learning_rate": 1.3467147107545668e-05,
+ "loss": 0.0812,
+ "step": 7270
+ },
+ {
+ "epoch": 16.25,
+ "learning_rate": 1.3450188734628344e-05,
+ "loss": 0.0653,
+ "step": 7280
+ },
+ {
+ "epoch": 16.27,
+ "learning_rate": 1.3433219091216069e-05,
+ "loss": 0.0809,
+ "step": 7290
+ },
+ {
+ "epoch": 16.29,
+ "learning_rate": 1.3416238232742414e-05,
+ "loss": 0.0919,
+ "step": 7300
+ },
+ {
+ "epoch": 16.32,
+ "learning_rate": 1.3399246214677583e-05,
+ "loss": 0.0706,
+ "step": 7310
+ },
+ {
+ "epoch": 16.34,
+ "learning_rate": 1.338224309252824e-05,
+ "loss": 0.0947,
+ "step": 7320
+ },
+ {
+ "epoch": 16.36,
+ "learning_rate": 1.3365228921837314e-05,
+ "loss": 0.0592,
+ "step": 7330
+ },
+ {
+ "epoch": 16.38,
+ "learning_rate": 1.3348203758183831e-05,
+ "loss": 0.0839,
+ "step": 7340
+ },
+ {
+ "epoch": 16.41,
+ "learning_rate": 1.3331167657182726e-05,
+ "loss": 0.0852,
+ "step": 7350
+ },
+ {
+ "epoch": 16.43,
+ "learning_rate": 1.3314120674484663e-05,
+ "loss": 0.0893,
+ "step": 7360
+ },
+ {
+ "epoch": 16.45,
+ "learning_rate": 1.3297062865775851e-05,
+ "loss": 0.0943,
+ "step": 7370
+ },
+ {
+ "epoch": 16.47,
+ "learning_rate": 1.327999428677786e-05,
+ "loss": 0.0839,
+ "step": 7380
+ },
+ {
+ "epoch": 16.5,
+ "learning_rate": 1.3262914993247454e-05,
+ "loss": 0.0985,
+ "step": 7390
+ },
+ {
+ "epoch": 16.52,
+ "learning_rate": 1.324582504097638e-05,
+ "loss": 0.0892,
+ "step": 7400
+ },
+ {
+ "epoch": 16.54,
+ "learning_rate": 1.3228724485791225e-05,
+ "loss": 0.0724,
+ "step": 7410
+ },
+ {
+ "epoch": 16.56,
+ "learning_rate": 1.321161338355319e-05,
+ "loss": 0.0732,
+ "step": 7420
+ },
+ {
+ "epoch": 16.58,
+ "learning_rate": 1.3194491790157947e-05,
+ "loss": 0.072,
+ "step": 7430
+ },
+ {
+ "epoch": 16.61,
+ "learning_rate": 1.3177359761535427e-05,
+ "loss": 0.1024,
+ "step": 7440
+ },
+ {
+ "epoch": 16.63,
+ "learning_rate": 1.3160217353649652e-05,
+ "loss": 0.0986,
+ "step": 7450
+ },
+ {
+ "epoch": 16.65,
+ "learning_rate": 1.3143064622498551e-05,
+ "loss": 0.0729,
+ "step": 7460
+ },
+ {
+ "epoch": 16.67,
+ "learning_rate": 1.312590162411378e-05,
+ "loss": 0.0892,
+ "step": 7470
+ },
+ {
+ "epoch": 16.7,
+ "learning_rate": 1.310872841456052e-05,
+ "loss": 0.0931,
+ "step": 7480
+ },
+ {
+ "epoch": 16.72,
+ "learning_rate": 1.3091545049937322e-05,
+ "loss": 0.128,
+ "step": 7490
+ },
+ {
+ "epoch": 16.74,
+ "learning_rate": 1.3074351586375906e-05,
+ "loss": 0.1005,
+ "step": 7500
+ },
+ {
+ "epoch": 16.76,
+ "learning_rate": 1.305714808004098e-05,
+ "loss": 0.0697,
+ "step": 7510
+ },
+ {
+ "epoch": 16.79,
+ "learning_rate": 1.3039934587130056e-05,
+ "loss": 0.1304,
+ "step": 7520
+ },
+ {
+ "epoch": 16.81,
+ "learning_rate": 1.3022711163873272e-05,
+ "loss": 0.078,
+ "step": 7530
+ },
+ {
+ "epoch": 16.83,
+ "learning_rate": 1.3005477866533202e-05,
+ "loss": 0.0796,
+ "step": 7540
+ },
+ {
+ "epoch": 16.85,
+ "learning_rate": 1.2988234751404683e-05,
+ "loss": 0.08,
+ "step": 7550
+ },
+ {
+ "epoch": 16.88,
+ "learning_rate": 1.2970981874814613e-05,
+ "loss": 0.0796,
+ "step": 7560
+ },
+ {
+ "epoch": 16.9,
+ "learning_rate": 1.2953719293121775e-05,
+ "loss": 0.1008,
+ "step": 7570
+ },
+ {
+ "epoch": 16.92,
+ "learning_rate": 1.2936447062716668e-05,
+ "loss": 0.1491,
+ "step": 7580
+ },
+ {
+ "epoch": 16.94,
+ "learning_rate": 1.2919165240021303e-05,
+ "loss": 0.0754,
+ "step": 7590
+ },
+ {
+ "epoch": 16.96,
+ "learning_rate": 1.2901873881489021e-05,
+ "loss": 0.0843,
+ "step": 7600
+ },
+ {
+ "epoch": 16.99,
+ "learning_rate": 1.288457304360432e-05,
+ "loss": 0.0734,
+ "step": 7610
+ },
+ {
+ "epoch": 17.01,
+ "learning_rate": 1.2867262782882662e-05,
+ "loss": 0.0672,
+ "step": 7620
+ },
+ {
+ "epoch": 17.03,
+ "learning_rate": 1.2849943155870284e-05,
+ "loss": 0.0677,
+ "step": 7630
+ },
+ {
+ "epoch": 17.05,
+ "learning_rate": 1.2832614219144027e-05,
+ "loss": 0.1288,
+ "step": 7640
+ },
+ {
+ "epoch": 17.08,
+ "learning_rate": 1.2815276029311138e-05,
+ "loss": 0.0643,
+ "step": 7650
+ },
+ {
+ "epoch": 17.1,
+ "learning_rate": 1.2797928643009097e-05,
+ "loss": 0.0823,
+ "step": 7660
+ },
+ {
+ "epoch": 17.12,
+ "learning_rate": 1.2780572116905418e-05,
+ "loss": 0.071,
+ "step": 7670
+ },
+ {
+ "epoch": 17.14,
+ "learning_rate": 1.276320650769748e-05,
+ "loss": 0.0754,
+ "step": 7680
+ },
+ {
+ "epoch": 17.17,
+ "learning_rate": 1.2745831872112318e-05,
+ "loss": 0.0638,
+ "step": 7690
+ },
+ {
+ "epoch": 17.19,
+ "learning_rate": 1.2728448266906468e-05,
+ "loss": 0.0794,
+ "step": 7700
+ },
+ {
+ "epoch": 17.21,
+ "learning_rate": 1.2711055748865765e-05,
+ "loss": 0.0865,
+ "step": 7710
+ },
+ {
+ "epoch": 17.23,
+ "learning_rate": 1.2693654374805148e-05,
+ "loss": 0.0854,
+ "step": 7720
+ },
+ {
+ "epoch": 17.25,
+ "learning_rate": 1.2676244201568498e-05,
+ "loss": 0.0831,
+ "step": 7730
+ },
+ {
+ "epoch": 17.28,
+ "learning_rate": 1.2658825286028428e-05,
+ "loss": 0.0749,
+ "step": 7740
+ },
+ {
+ "epoch": 17.3,
+ "learning_rate": 1.2641397685086124e-05,
+ "loss": 0.0809,
+ "step": 7750
+ },
+ {
+ "epoch": 17.32,
+ "learning_rate": 1.2623961455671125e-05,
+ "loss": 0.071,
+ "step": 7760
+ },
+ {
+ "epoch": 17.34,
+ "learning_rate": 1.2606516654741172e-05,
+ "loss": 0.0777,
+ "step": 7770
+ },
+ {
+ "epoch": 17.37,
+ "learning_rate": 1.2589063339281995e-05,
+ "loss": 0.1106,
+ "step": 7780
+ },
+ {
+ "epoch": 17.39,
+ "learning_rate": 1.257160156630715e-05,
+ "loss": 0.0739,
+ "step": 7790
+ },
+ {
+ "epoch": 17.41,
+ "learning_rate": 1.2554131392857812e-05,
+ "loss": 0.0669,
+ "step": 7800
+ },
+ {
+ "epoch": 17.43,
+ "learning_rate": 1.253665287600259e-05,
+ "loss": 0.0927,
+ "step": 7810
+ },
+ {
+ "epoch": 17.46,
+ "learning_rate": 1.2519166072837368e-05,
+ "loss": 0.0701,
+ "step": 7820
+ },
+ {
+ "epoch": 17.48,
+ "learning_rate": 1.250167104048508e-05,
+ "loss": 0.076,
+ "step": 7830
+ },
+ {
+ "epoch": 17.5,
+ "learning_rate": 1.248416783609555e-05,
+ "loss": 0.087,
+ "step": 7840
+ },
+ {
+ "epoch": 17.52,
+ "learning_rate": 1.2466656516845293e-05,
+ "loss": 0.0743,
+ "step": 7850
+ },
+ {
+ "epoch": 17.54,
+ "learning_rate": 1.244913713993734e-05,
+ "loss": 0.0949,
+ "step": 7860
+ },
+ {
+ "epoch": 17.57,
+ "learning_rate": 1.2431609762601036e-05,
+ "loss": 0.0738,
+ "step": 7870
+ },
+ {
+ "epoch": 17.59,
+ "learning_rate": 1.241407444209186e-05,
+ "loss": 0.0623,
+ "step": 7880
+ },
+ {
+ "epoch": 17.61,
+ "learning_rate": 1.2396531235691245e-05,
+ "loss": 0.0812,
+ "step": 7890
+ },
+ {
+ "epoch": 17.63,
+ "learning_rate": 1.2378980200706376e-05,
+ "loss": 0.0899,
+ "step": 7900
+ },
+ {
+ "epoch": 17.66,
+ "learning_rate": 1.236142139447002e-05,
+ "loss": 0.089,
+ "step": 7910
+ },
+ {
+ "epoch": 17.68,
+ "learning_rate": 1.2343854874340324e-05,
+ "loss": 0.0817,
+ "step": 7920
+ },
+ {
+ "epoch": 17.7,
+ "learning_rate": 1.2326280697700632e-05,
+ "loss": 0.1199,
+ "step": 7930
+ },
+ {
+ "epoch": 17.72,
+ "learning_rate": 1.2308698921959306e-05,
+ "loss": 0.0904,
+ "step": 7940
+ },
+ {
+ "epoch": 17.75,
+ "learning_rate": 1.2291109604549525e-05,
+ "loss": 0.0667,
+ "step": 7950
+ },
+ {
+ "epoch": 17.77,
+ "learning_rate": 1.2273512802929107e-05,
+ "loss": 0.1035,
+ "step": 7960
+ },
+ {
+ "epoch": 17.79,
+ "learning_rate": 1.2255908574580311e-05,
+ "loss": 0.0842,
+ "step": 7970
+ },
+ {
+ "epoch": 17.81,
+ "learning_rate": 1.2238296977009672e-05,
+ "loss": 0.0892,
+ "step": 7980
+ },
+ {
+ "epoch": 17.83,
+ "learning_rate": 1.2220678067747785e-05,
+ "loss": 0.0846,
+ "step": 7990
+ },
+ {
+ "epoch": 17.86,
+ "learning_rate": 1.2203051904349128e-05,
+ "loss": 0.0829,
+ "step": 8000
+ },
+ {
+ "epoch": 17.88,
+ "learning_rate": 1.2185418544391885e-05,
+ "loss": 0.0822,
+ "step": 8010
+ },
+ {
+ "epoch": 17.9,
+ "learning_rate": 1.2167778045477743e-05,
+ "loss": 0.088,
+ "step": 8020
+ },
+ {
+ "epoch": 17.92,
+ "learning_rate": 1.215013046523171e-05,
+ "loss": 0.0973,
+ "step": 8030
+ },
+ {
+ "epoch": 17.95,
+ "learning_rate": 1.2132475861301928e-05,
+ "loss": 0.0811,
+ "step": 8040
+ },
+ {
+ "epoch": 17.97,
+ "learning_rate": 1.2114814291359476e-05,
+ "loss": 0.0749,
+ "step": 8050
+ },
+ {
+ "epoch": 17.99,
+ "learning_rate": 1.20971458130982e-05,
+ "loss": 0.0894,
+ "step": 8060
+ },
+ {
+ "epoch": 18.01,
+ "learning_rate": 1.20794704842345e-05,
+ "loss": 0.0807,
+ "step": 8070
+ },
+ {
+ "epoch": 18.04,
+ "learning_rate": 1.2061788362507168e-05,
+ "loss": 0.0713,
+ "step": 8080
+ },
+ {
+ "epoch": 18.06,
+ "learning_rate": 1.204409950567717e-05,
+ "loss": 0.0682,
+ "step": 8090
+ },
+ {
+ "epoch": 18.08,
+ "learning_rate": 1.2026403971527487e-05,
+ "loss": 0.0653,
+ "step": 8100
+ },
+ {
+ "epoch": 18.1,
+ "learning_rate": 1.2008701817862906e-05,
+ "loss": 0.0656,
+ "step": 8110
+ },
+ {
+ "epoch": 18.12,
+ "learning_rate": 1.1990993102509838e-05,
+ "loss": 0.0801,
+ "step": 8120
+ },
+ {
+ "epoch": 18.15,
+ "learning_rate": 1.1973277883316128e-05,
+ "loss": 0.0697,
+ "step": 8130
+ },
+ {
+ "epoch": 18.17,
+ "learning_rate": 1.1955556218150872e-05,
+ "loss": 0.0742,
+ "step": 8140
+ },
+ {
+ "epoch": 18.19,
+ "learning_rate": 1.1937828164904216e-05,
+ "loss": 0.076,
+ "step": 8150
+ },
+ {
+ "epoch": 18.21,
+ "learning_rate": 1.1920093781487175e-05,
+ "loss": 0.1006,
+ "step": 8160
+ },
+ {
+ "epoch": 18.24,
+ "learning_rate": 1.1902353125831441e-05,
+ "loss": 0.1085,
+ "step": 8170
+ },
+ {
+ "epoch": 18.26,
+ "learning_rate": 1.1884606255889203e-05,
+ "loss": 0.096,
+ "step": 8180
+ },
+ {
+ "epoch": 18.28,
+ "learning_rate": 1.1866853229632942e-05,
+ "loss": 0.0735,
+ "step": 8190
+ },
+ {
+ "epoch": 18.3,
+ "learning_rate": 1.1849094105055248e-05,
+ "loss": 0.0591,
+ "step": 8200
+ },
+ {
+ "epoch": 18.33,
+ "learning_rate": 1.1831328940168638e-05,
+ "loss": 0.1013,
+ "step": 8210
+ },
+ {
+ "epoch": 18.35,
+ "learning_rate": 1.181355779300536e-05,
+ "loss": 0.0744,
+ "step": 8220
+ },
+ {
+ "epoch": 18.37,
+ "learning_rate": 1.1795780721617199e-05,
+ "loss": 0.0704,
+ "step": 8230
+ },
+ {
+ "epoch": 18.39,
+ "learning_rate": 1.1777997784075294e-05,
+ "loss": 0.0693,
+ "step": 8240
+ },
+ {
+ "epoch": 18.42,
+ "learning_rate": 1.176020903846995e-05,
+ "loss": 0.067,
+ "step": 8250
+ },
+ {
+ "epoch": 18.44,
+ "learning_rate": 1.1742414542910444e-05,
+ "loss": 0.0709,
+ "step": 8260
+ },
+ {
+ "epoch": 18.46,
+ "learning_rate": 1.1724614355524832e-05,
+ "loss": 0.0848,
+ "step": 8270
+ },
+ {
+ "epoch": 18.48,
+ "learning_rate": 1.1706808534459768e-05,
+ "loss": 0.0898,
+ "step": 8280
+ },
+ {
+ "epoch": 18.5,
+ "learning_rate": 1.16889971378803e-05,
+ "loss": 0.0763,
+ "step": 8290
+ },
+ {
+ "epoch": 18.53,
+ "learning_rate": 1.1671180223969705e-05,
+ "loss": 0.0886,
+ "step": 8300
+ },
+ {
+ "epoch": 18.55,
+ "learning_rate": 1.1653357850929268e-05,
+ "loss": 0.0829,
+ "step": 8310
+ },
+ {
+ "epoch": 18.57,
+ "learning_rate": 1.1635530076978115e-05,
+ "loss": 0.0736,
+ "step": 8320
+ },
+ {
+ "epoch": 18.59,
+ "learning_rate": 1.161769696035301e-05,
+ "loss": 0.0867,
+ "step": 8330
+ },
+ {
+ "epoch": 18.62,
+ "learning_rate": 1.1599858559308175e-05,
+ "loss": 0.088,
+ "step": 8340
+ },
+ {
+ "epoch": 18.64,
+ "learning_rate": 1.158201493211509e-05,
+ "loss": 0.0732,
+ "step": 8350
+ },
+ {
+ "epoch": 18.66,
+ "learning_rate": 1.156416613706231e-05,
+ "loss": 0.0675,
+ "step": 8360
+ },
+ {
+ "epoch": 18.68,
+ "learning_rate": 1.1546312232455266e-05,
+ "loss": 0.0729,
+ "step": 8370
+ },
+ {
+ "epoch": 18.71,
+ "learning_rate": 1.152845327661609e-05,
+ "loss": 0.0635,
+ "step": 8380
+ },
+ {
+ "epoch": 18.73,
+ "learning_rate": 1.1510589327883406e-05,
+ "loss": 0.0776,
+ "step": 8390
+ },
+ {
+ "epoch": 18.75,
+ "learning_rate": 1.1492720444612148e-05,
+ "loss": 0.0625,
+ "step": 8400
+ },
+ {
+ "epoch": 18.77,
+ "learning_rate": 1.1474846685173374e-05,
+ "loss": 0.0653,
+ "step": 8410
+ },
+ {
+ "epoch": 18.79,
+ "learning_rate": 1.1456968107954066e-05,
+ "loss": 0.0592,
+ "step": 8420
+ },
+ {
+ "epoch": 18.82,
+ "learning_rate": 1.143908477135695e-05,
+ "loss": 0.0811,
+ "step": 8430
+ },
+ {
+ "epoch": 18.84,
+ "learning_rate": 1.1421196733800291e-05,
+ "loss": 0.0989,
+ "step": 8440
+ },
+ {
+ "epoch": 18.86,
+ "learning_rate": 1.1403304053717719e-05,
+ "loss": 0.0673,
+ "step": 8450
+ },
+ {
+ "epoch": 18.88,
+ "learning_rate": 1.138540678955802e-05,
+ "loss": 0.0691,
+ "step": 8460
+ },
+ {
+ "epoch": 18.91,
+ "learning_rate": 1.1367504999784963e-05,
+ "loss": 0.0656,
+ "step": 8470
+ },
+ {
+ "epoch": 18.93,
+ "learning_rate": 1.1349598742877097e-05,
+ "loss": 0.0964,
+ "step": 8480
+ },
+ {
+ "epoch": 18.95,
+ "learning_rate": 1.1331688077327563e-05,
+ "loss": 0.1015,
+ "step": 8490
+ },
+ {
+ "epoch": 18.97,
+ "learning_rate": 1.1313773061643905e-05,
+ "loss": 0.1236,
+ "step": 8500
+ },
+ {
+ "epoch": 19.0,
+ "learning_rate": 1.1295853754347876e-05,
+ "loss": 0.0762,
+ "step": 8510
+ },
+ {
+ "epoch": 19.02,
+ "learning_rate": 1.1277930213975249e-05,
+ "loss": 0.0847,
+ "step": 8520
+ },
+ {
+ "epoch": 19.04,
+ "learning_rate": 1.1260002499075617e-05,
+ "loss": 0.0643,
+ "step": 8530
+ },
+ {
+ "epoch": 19.06,
+ "learning_rate": 1.1242070668212227e-05,
+ "loss": 0.0858,
+ "step": 8540
+ },
+ {
+ "epoch": 19.08,
+ "learning_rate": 1.1224134779961758e-05,
+ "loss": 0.0586,
+ "step": 8550
+ },
+ {
+ "epoch": 19.11,
+ "learning_rate": 1.1206194892914142e-05,
+ "loss": 0.0868,
+ "step": 8560
+ },
+ {
+ "epoch": 19.13,
+ "learning_rate": 1.1188251065672382e-05,
+ "loss": 0.0629,
+ "step": 8570
+ },
+ {
+ "epoch": 19.15,
+ "learning_rate": 1.117030335685235e-05,
+ "loss": 0.0594,
+ "step": 8580
+ },
+ {
+ "epoch": 19.17,
+ "learning_rate": 1.1152351825082588e-05,
+ "loss": 0.0558,
+ "step": 8590
+ },
+ {
+ "epoch": 19.2,
+ "learning_rate": 1.1134396529004143e-05,
+ "loss": 0.0621,
+ "step": 8600
+ },
+ {
+ "epoch": 19.22,
+ "learning_rate": 1.1116437527270343e-05,
+ "loss": 0.0687,
+ "step": 8610
+ },
+ {
+ "epoch": 19.24,
+ "learning_rate": 1.109847487854663e-05,
+ "loss": 0.0739,
+ "step": 8620
+ },
+ {
+ "epoch": 19.26,
+ "learning_rate": 1.1080508641510357e-05,
+ "loss": 0.069,
+ "step": 8630
+ },
+ {
+ "epoch": 19.29,
+ "learning_rate": 1.1062538874850597e-05,
+ "loss": 0.078,
+ "step": 8640
+ },
+ {
+ "epoch": 19.31,
+ "learning_rate": 1.1044565637267957e-05,
+ "loss": 0.0656,
+ "step": 8650
+ },
+ {
+ "epoch": 19.33,
+ "learning_rate": 1.1026588987474379e-05,
+ "loss": 0.1075,
+ "step": 8660
+ },
+ {
+ "epoch": 19.35,
+ "learning_rate": 1.100860898419295e-05,
+ "loss": 0.0751,
+ "step": 8670
+ },
+ {
+ "epoch": 19.38,
+ "learning_rate": 1.0990625686157714e-05,
+ "loss": 0.0578,
+ "step": 8680
+ },
+ {
+ "epoch": 19.4,
+ "learning_rate": 1.097263915211348e-05,
+ "loss": 0.0664,
+ "step": 8690
+ },
+ {
+ "epoch": 19.42,
+ "learning_rate": 1.0954649440815625e-05,
+ "loss": 0.0714,
+ "step": 8700
+ },
+ {
+ "epoch": 19.44,
+ "learning_rate": 1.0936656611029901e-05,
+ "loss": 0.0731,
+ "step": 8710
+ },
+ {
+ "epoch": 19.46,
+ "learning_rate": 1.091866072153226e-05,
+ "loss": 0.0647,
+ "step": 8720
+ },
+ {
+ "epoch": 19.49,
+ "learning_rate": 1.090066183110863e-05,
+ "loss": 0.0902,
+ "step": 8730
+ },
+ {
+ "epoch": 19.51,
+ "learning_rate": 1.0882659998554759e-05,
+ "loss": 0.0809,
+ "step": 8740
+ },
+ {
+ "epoch": 19.53,
+ "learning_rate": 1.0864655282675997e-05,
+ "loss": 0.0769,
+ "step": 8750
+ },
+ {
+ "epoch": 19.55,
+ "learning_rate": 1.0846647742287116e-05,
+ "loss": 0.0708,
+ "step": 8760
+ },
+ {
+ "epoch": 19.58,
+ "learning_rate": 1.0828637436212111e-05,
+ "loss": 0.0723,
+ "step": 8770
+ },
+ {
+ "epoch": 19.6,
+ "learning_rate": 1.0810624423284012e-05,
+ "loss": 0.0747,
+ "step": 8780
+ },
+ {
+ "epoch": 19.62,
+ "learning_rate": 1.07926087623447e-05,
+ "loss": 0.0634,
+ "step": 8790
+ },
+ {
+ "epoch": 19.64,
+ "learning_rate": 1.0774590512244694e-05,
+ "loss": 0.0708,
+ "step": 8800
+ },
+ {
+ "epoch": 19.67,
+ "learning_rate": 1.0756569731842978e-05,
+ "loss": 0.0698,
+ "step": 8810
+ },
+ {
+ "epoch": 19.69,
+ "learning_rate": 1.07385464800068e-05,
+ "loss": 0.0605,
+ "step": 8820
+ },
+ {
+ "epoch": 19.71,
+ "learning_rate": 1.0720520815611476e-05,
+ "loss": 0.0744,
+ "step": 8830
+ },
+ {
+ "epoch": 19.73,
+ "learning_rate": 1.0702492797540214e-05,
+ "loss": 0.0819,
+ "step": 8840
+ },
+ {
+ "epoch": 19.75,
+ "learning_rate": 1.06844624846839e-05,
+ "loss": 0.0727,
+ "step": 8850
+ },
+ {
+ "epoch": 19.78,
+ "learning_rate": 1.0666429935940925e-05,
+ "loss": 0.0738,
+ "step": 8860
+ },
+ {
+ "epoch": 19.8,
+ "learning_rate": 1.0648395210216975e-05,
+ "loss": 0.0805,
+ "step": 8870
+ },
+ {
+ "epoch": 19.82,
+ "learning_rate": 1.0630358366424856e-05,
+ "loss": 0.0687,
+ "step": 8880
+ },
+ {
+ "epoch": 19.84,
+ "learning_rate": 1.0612319463484286e-05,
+ "loss": 0.0622,
+ "step": 8890
+ },
+ {
+ "epoch": 19.87,
+ "learning_rate": 1.0594278560321713e-05,
+ "loss": 0.0894,
+ "step": 8900
+ },
+ {
+ "epoch": 19.89,
+ "learning_rate": 1.0576235715870119e-05,
+ "loss": 0.0682,
+ "step": 8910
+ },
+ {
+ "epoch": 19.91,
+ "learning_rate": 1.0558190989068822e-05,
+ "loss": 0.0812,
+ "step": 8920
+ },
+ {
+ "epoch": 19.93,
+ "learning_rate": 1.0540144438863302e-05,
+ "loss": 0.132,
+ "step": 8930
+ },
+ {
+ "epoch": 19.96,
+ "learning_rate": 1.052209612420498e-05,
+ "loss": 0.0985,
+ "step": 8940
+ },
+ {
+ "epoch": 19.98,
+ "learning_rate": 1.050404610405105e-05,
+ "loss": 0.0716,
+ "step": 8950
+ },
+ {
+ "epoch": 20.0,
+ "learning_rate": 1.0485994437364278e-05,
+ "loss": 0.0676,
+ "step": 8960
+ },
+ {
+ "epoch": 20.02,
+ "learning_rate": 1.0467941183112801e-05,
+ "loss": 0.0585,
+ "step": 8970
+ },
+ {
+ "epoch": 20.04,
+ "learning_rate": 1.0449886400269952e-05,
+ "loss": 0.0711,
+ "step": 8980
+ },
+ {
+ "epoch": 20.07,
+ "learning_rate": 1.0431830147814049e-05,
+ "loss": 0.0558,
+ "step": 8990
+ },
+ {
+ "epoch": 20.09,
+ "learning_rate": 1.0413772484728211e-05,
+ "loss": 0.0473,
+ "step": 9000
+ },
+ {
+ "epoch": 20.11,
+ "learning_rate": 1.0395713470000173e-05,
+ "loss": 0.058,
+ "step": 9010
+ },
+ {
+ "epoch": 20.13,
+ "learning_rate": 1.0377653162622076e-05,
+ "loss": 0.0631,
+ "step": 9020
+ },
+ {
+ "epoch": 20.16,
+ "learning_rate": 1.0359591621590292e-05,
+ "loss": 0.065,
+ "step": 9030
+ },
+ {
+ "epoch": 20.18,
+ "learning_rate": 1.034152890590521e-05,
+ "loss": 0.0676,
+ "step": 9040
+ },
+ {
+ "epoch": 20.2,
+ "learning_rate": 1.0323465074571078e-05,
+ "loss": 0.0845,
+ "step": 9050
+ },
+ {
+ "epoch": 20.22,
+ "learning_rate": 1.0305400186595764e-05,
+ "loss": 0.0723,
+ "step": 9060
+ },
+ {
+ "epoch": 20.25,
+ "learning_rate": 1.0287334300990602e-05,
+ "loss": 0.105,
+ "step": 9070
+ },
+ {
+ "epoch": 20.27,
+ "learning_rate": 1.026926747677018e-05,
+ "loss": 0.0714,
+ "step": 9080
+ },
+ {
+ "epoch": 20.29,
+ "learning_rate": 1.025119977295216e-05,
+ "loss": 0.0684,
+ "step": 9090
+ },
+ {
+ "epoch": 20.31,
+ "learning_rate": 1.0233131248557067e-05,
+ "loss": 0.0696,
+ "step": 9100
+ },
+ {
+ "epoch": 20.33,
+ "learning_rate": 1.0215061962608111e-05,
+ "loss": 0.0572,
+ "step": 9110
+ },
+ {
+ "epoch": 20.36,
+ "learning_rate": 1.0196991974130986e-05,
+ "loss": 0.0968,
+ "step": 9120
+ },
+ {
+ "epoch": 20.38,
+ "learning_rate": 1.017892134215369e-05,
+ "loss": 0.0655,
+ "step": 9130
+ },
+ {
+ "epoch": 20.4,
+ "learning_rate": 1.0160850125706314e-05,
+ "loss": 0.0611,
+ "step": 9140
+ },
+ {
+ "epoch": 20.42,
+ "learning_rate": 1.0142778383820861e-05,
+ "loss": 0.0839,
+ "step": 9150
+ },
+ {
+ "epoch": 20.45,
+ "learning_rate": 1.0124706175531054e-05,
+ "loss": 0.0655,
+ "step": 9160
+ },
+ {
+ "epoch": 20.47,
+ "learning_rate": 1.0106633559872135e-05,
+ "loss": 0.0528,
+ "step": 9170
+ },
+ {
+ "epoch": 20.49,
+ "learning_rate": 1.0088560595880676e-05,
+ "loss": 0.0706,
+ "step": 9180
+ },
+ {
+ "epoch": 20.51,
+ "learning_rate": 1.0070487342594392e-05,
+ "loss": 0.0676,
+ "step": 9190
+ },
+ {
+ "epoch": 20.54,
+ "learning_rate": 1.005241385905194e-05,
+ "loss": 0.0712,
+ "step": 9200
+ },
+ {
+ "epoch": 20.56,
+ "learning_rate": 1.0034340204292728e-05,
+ "loss": 0.0657,
+ "step": 9210
+ },
+ {
+ "epoch": 20.58,
+ "learning_rate": 1.0016266437356727e-05,
+ "loss": 0.0713,
+ "step": 9220
+ },
+ {
+ "epoch": 20.6,
+ "learning_rate": 9.998192617284271e-06,
+ "loss": 0.065,
+ "step": 9230
+ },
+ {
+ "epoch": 20.62,
+ "learning_rate": 9.980118803115867e-06,
+ "loss": 0.0734,
+ "step": 9240
+ },
+ {
+ "epoch": 20.65,
+ "learning_rate": 9.962045053892004e-06,
+ "loss": 0.061,
+ "step": 9250
+ },
+ {
+ "epoch": 20.67,
+ "learning_rate": 9.94397142865296e-06,
+ "loss": 0.0727,
+ "step": 9260
+ },
+ {
+ "epoch": 20.69,
+ "learning_rate": 9.925897986438613e-06,
+ "loss": 0.0702,
+ "step": 9270
+ },
+ {
+ "epoch": 20.71,
+ "learning_rate": 9.907824786288226e-06,
+ "loss": 0.0979,
+ "step": 9280
+ },
+ {
+ "epoch": 20.74,
+ "learning_rate": 9.889751887240296e-06,
+ "loss": 0.0534,
+ "step": 9290
+ },
+ {
+ "epoch": 20.76,
+ "learning_rate": 9.87167934833231e-06,
+ "loss": 0.0634,
+ "step": 9300
+ },
+ {
+ "epoch": 20.78,
+ "learning_rate": 9.853607228600602e-06,
+ "loss": 0.0566,
+ "step": 9310
+ },
+ {
+ "epoch": 20.8,
+ "learning_rate": 9.835535587080118e-06,
+ "loss": 0.0774,
+ "step": 9320
+ },
+ {
+ "epoch": 20.83,
+ "learning_rate": 9.817464482804257e-06,
+ "loss": 0.0737,
+ "step": 9330
+ },
+ {
+ "epoch": 20.85,
+ "learning_rate": 9.799393974804651e-06,
+ "loss": 0.0621,
+ "step": 9340
+ },
+ {
+ "epoch": 20.87,
+ "learning_rate": 9.781324122110993e-06,
+ "loss": 0.0648,
+ "step": 9350
+ },
+ {
+ "epoch": 20.89,
+ "learning_rate": 9.763254983750829e-06,
+ "loss": 0.0753,
+ "step": 9360
+ },
+ {
+ "epoch": 20.92,
+ "learning_rate": 9.745186618749373e-06,
+ "loss": 0.0769,
+ "step": 9370
+ },
+ {
+ "epoch": 20.94,
+ "learning_rate": 9.727119086129321e-06,
+ "loss": 0.0764,
+ "step": 9380
+ },
+ {
+ "epoch": 20.96,
+ "learning_rate": 9.709052444910636e-06,
+ "loss": 0.0763,
+ "step": 9390
+ },
+ {
+ "epoch": 20.98,
+ "learning_rate": 9.690986754110378e-06,
+ "loss": 0.0742,
+ "step": 9400
+ },
+ {
+ "epoch": 21.0,
+ "learning_rate": 9.6729220727425e-06,
+ "loss": 0.0998,
+ "step": 9410
+ },
+ {
+ "epoch": 21.03,
+ "learning_rate": 9.654858459817663e-06,
+ "loss": 0.0583,
+ "step": 9420
+ },
+ {
+ "epoch": 21.05,
+ "learning_rate": 9.636795974343023e-06,
+ "loss": 0.0642,
+ "step": 9430
+ },
+ {
+ "epoch": 21.07,
+ "learning_rate": 9.61873467532207e-06,
+ "loss": 0.0636,
+ "step": 9440
+ },
+ {
+ "epoch": 21.09,
+ "learning_rate": 9.600674621754406e-06,
+ "loss": 0.046,
+ "step": 9450
+ },
+ {
+ "epoch": 21.12,
+ "learning_rate": 9.582615872635578e-06,
+ "loss": 0.0551,
+ "step": 9460
+ },
+ {
+ "epoch": 21.14,
+ "learning_rate": 9.564558486956853e-06,
+ "loss": 0.0534,
+ "step": 9470
+ },
+ {
+ "epoch": 21.16,
+ "learning_rate": 9.546502523705057e-06,
+ "loss": 0.0825,
+ "step": 9480
+ },
+ {
+ "epoch": 21.18,
+ "learning_rate": 9.528448041862375e-06,
+ "loss": 0.0592,
+ "step": 9490
+ },
+ {
+ "epoch": 21.21,
+ "learning_rate": 9.510395100406136e-06,
+ "loss": 0.0598,
+ "step": 9500
+ },
+ {
+ "epoch": 21.23,
+ "learning_rate": 9.492343758308651e-06,
+ "loss": 0.0615,
+ "step": 9510
+ },
+ {
+ "epoch": 21.25,
+ "learning_rate": 9.474294074536996e-06,
+ "loss": 0.067,
+ "step": 9520
+ },
+ {
+ "epoch": 21.27,
+ "learning_rate": 9.456246108052844e-06,
+ "loss": 0.0526,
+ "step": 9530
+ },
+ {
+ "epoch": 21.29,
+ "learning_rate": 9.438199917812241e-06,
+ "loss": 0.0577,
+ "step": 9540
+ },
+ {
+ "epoch": 21.32,
+ "learning_rate": 9.420155562765443e-06,
+ "loss": 0.0627,
+ "step": 9550
+ },
+ {
+ "epoch": 21.34,
+ "learning_rate": 9.402113101856705e-06,
+ "loss": 0.0575,
+ "step": 9560
+ },
+ {
+ "epoch": 21.36,
+ "learning_rate": 9.384072594024103e-06,
+ "loss": 0.07,
+ "step": 9570
+ },
+ {
+ "epoch": 21.38,
+ "learning_rate": 9.366034098199317e-06,
+ "loss": 0.0578,
+ "step": 9580
+ },
+ {
+ "epoch": 21.41,
+ "learning_rate": 9.347997673307473e-06,
+ "loss": 0.0543,
+ "step": 9590
+ },
+ {
+ "epoch": 21.43,
+ "learning_rate": 9.329963378266919e-06,
+ "loss": 0.0553,
+ "step": 9600
+ },
+ {
+ "epoch": 21.45,
+ "learning_rate": 9.31193127198905e-06,
+ "loss": 0.0581,
+ "step": 9610
+ },
+ {
+ "epoch": 21.47,
+ "learning_rate": 9.293901413378116e-06,
+ "loss": 0.0494,
+ "step": 9620
+ },
+ {
+ "epoch": 21.5,
+ "learning_rate": 9.275873861331012e-06,
+ "loss": 0.0688,
+ "step": 9630
+ },
+ {
+ "epoch": 21.52,
+ "learning_rate": 9.257848674737112e-06,
+ "loss": 0.0548,
+ "step": 9640
+ },
+ {
+ "epoch": 21.54,
+ "learning_rate": 9.239825912478054e-06,
+ "loss": 0.0744,
+ "step": 9650
+ },
+ {
+ "epoch": 21.56,
+ "learning_rate": 9.221805633427564e-06,
+ "loss": 0.0701,
+ "step": 9660
+ },
+ {
+ "epoch": 21.58,
+ "learning_rate": 9.203787896451246e-06,
+ "loss": 0.0495,
+ "step": 9670
+ },
+ {
+ "epoch": 21.61,
+ "learning_rate": 9.185772760406408e-06,
+ "loss": 0.0528,
+ "step": 9680
+ },
+ {
+ "epoch": 21.63,
+ "learning_rate": 9.167760284141859e-06,
+ "loss": 0.0656,
+ "step": 9690
+ },
+ {
+ "epoch": 21.65,
+ "learning_rate": 9.149750526497725e-06,
+ "loss": 0.0549,
+ "step": 9700
+ },
+ {
+ "epoch": 21.67,
+ "learning_rate": 9.131743546305235e-06,
+ "loss": 0.0596,
+ "step": 9710
+ },
+ {
+ "epoch": 21.7,
+ "learning_rate": 9.113739402386566e-06,
+ "loss": 0.0602,
+ "step": 9720
+ },
+ {
+ "epoch": 21.72,
+ "learning_rate": 9.095738153554624e-06,
+ "loss": 0.0816,
+ "step": 9730
+ },
+ {
+ "epoch": 21.74,
+ "learning_rate": 9.077739858612843e-06,
+ "loss": 0.0752,
+ "step": 9740
+ },
+ {
+ "epoch": 21.76,
+ "learning_rate": 9.059744576355027e-06,
+ "loss": 0.0666,
+ "step": 9750
+ },
+ {
+ "epoch": 21.79,
+ "learning_rate": 9.041752365565125e-06,
+ "loss": 0.0584,
+ "step": 9760
+ },
+ {
+ "epoch": 21.81,
+ "learning_rate": 9.023763285017065e-06,
+ "loss": 0.063,
+ "step": 9770
+ },
+ {
+ "epoch": 21.83,
+ "learning_rate": 9.005777393474534e-06,
+ "loss": 0.0545,
+ "step": 9780
+ },
+ {
+ "epoch": 21.85,
+ "learning_rate": 8.987794749690819e-06,
+ "loss": 0.0552,
+ "step": 9790
+ },
+ {
+ "epoch": 21.88,
+ "learning_rate": 8.969815412408583e-06,
+ "loss": 0.0589,
+ "step": 9800
+ },
+ {
+ "epoch": 21.9,
+ "learning_rate": 8.951839440359701e-06,
+ "loss": 0.0698,
+ "step": 9810
+ },
+ {
+ "epoch": 21.92,
+ "learning_rate": 8.93386689226504e-06,
+ "loss": 0.0911,
+ "step": 9820
+ },
+ {
+ "epoch": 21.94,
+ "learning_rate": 8.915897826834295e-06,
+ "loss": 0.0779,
+ "step": 9830
+ },
+ {
+ "epoch": 21.96,
+ "learning_rate": 8.89793230276578e-06,
+ "loss": 0.0606,
+ "step": 9840
+ },
+ {
+ "epoch": 21.99,
+ "learning_rate": 8.879970378746238e-06,
+ "loss": 0.0787,
+ "step": 9850
+ },
+ {
+ "epoch": 22.01,
+ "learning_rate": 8.862012113450662e-06,
+ "loss": 0.0574,
+ "step": 9860
+ },
+ {
+ "epoch": 22.03,
+ "learning_rate": 8.844057565542074e-06,
+ "loss": 0.0581,
+ "step": 9870
+ },
+ {
+ "epoch": 22.05,
+ "learning_rate": 8.826106793671376e-06,
+ "loss": 0.0437,
+ "step": 9880
+ },
+ {
+ "epoch": 22.08,
+ "learning_rate": 8.808159856477115e-06,
+ "loss": 0.069,
+ "step": 9890
+ },
+ {
+ "epoch": 22.1,
+ "learning_rate": 8.790216812585327e-06,
+ "loss": 0.046,
+ "step": 9900
+ },
+ {
+ "epoch": 22.12,
+ "learning_rate": 8.772277720609312e-06,
+ "loss": 0.0825,
+ "step": 9910
+ },
+ {
+ "epoch": 22.14,
+ "learning_rate": 8.754342639149486e-06,
+ "loss": 0.0659,
+ "step": 9920
+ },
+ {
+ "epoch": 22.17,
+ "learning_rate": 8.736411626793139e-06,
+ "loss": 0.0649,
+ "step": 9930
+ },
+ {
+ "epoch": 22.19,
+ "learning_rate": 8.718484742114285e-06,
+ "loss": 0.0563,
+ "step": 9940
+ },
+ {
+ "epoch": 22.21,
+ "learning_rate": 8.700562043673448e-06,
+ "loss": 0.043,
+ "step": 9950
+ },
+ {
+ "epoch": 22.23,
+ "learning_rate": 8.682643590017474e-06,
+ "loss": 0.0482,
+ "step": 9960
+ },
+ {
+ "epoch": 22.25,
+ "learning_rate": 8.664729439679354e-06,
+ "loss": 0.0436,
+ "step": 9970
+ },
+ {
+ "epoch": 22.28,
+ "learning_rate": 8.646819651178008e-06,
+ "loss": 0.0536,
+ "step": 9980
+ },
+ {
+ "epoch": 22.3,
+ "learning_rate": 8.628914283018119e-06,
+ "loss": 0.0615,
+ "step": 9990
+ },
+ {
+ "epoch": 22.32,
+ "learning_rate": 8.61101339368992e-06,
+ "loss": 0.0461,
+ "step": 10000
+ },
+ {
+ "epoch": 22.34,
+ "learning_rate": 8.593117041669024e-06,
+ "loss": 0.0605,
+ "step": 10010
+ },
+ {
+ "epoch": 22.37,
+ "learning_rate": 8.57522528541621e-06,
+ "loss": 0.07,
+ "step": 10020
+ },
+ {
+ "epoch": 22.39,
+ "learning_rate": 8.55733818337726e-06,
+ "loss": 0.059,
+ "step": 10030
+ },
+ {
+ "epoch": 22.41,
+ "learning_rate": 8.539455793982737e-06,
+ "loss": 0.047,
+ "step": 10040
+ },
+ {
+ "epoch": 22.43,
+ "learning_rate": 8.521578175647823e-06,
+ "loss": 0.0859,
+ "step": 10050
+ },
+ {
+ "epoch": 22.46,
+ "learning_rate": 8.503705386772098e-06,
+ "loss": 0.0596,
+ "step": 10060
+ },
+ {
+ "epoch": 22.48,
+ "learning_rate": 8.485837485739384e-06,
+ "loss": 0.0516,
+ "step": 10070
+ },
+ {
+ "epoch": 22.5,
+ "learning_rate": 8.467974530917524e-06,
+ "loss": 0.0495,
+ "step": 10080
+ },
+ {
+ "epoch": 22.52,
+ "learning_rate": 8.450116580658208e-06,
+ "loss": 0.0562,
+ "step": 10090
+ },
+ {
+ "epoch": 22.54,
+ "learning_rate": 8.432263693296783e-06,
+ "loss": 0.0549,
+ "step": 10100
+ },
+ {
+ "epoch": 22.57,
+ "learning_rate": 8.414415927152042e-06,
+ "loss": 0.0671,
+ "step": 10110
+ },
+ {
+ "epoch": 22.59,
+ "learning_rate": 8.396573340526069e-06,
+ "loss": 0.0678,
+ "step": 10120
+ },
+ {
+ "epoch": 22.61,
+ "learning_rate": 8.37873599170401e-06,
+ "loss": 0.0668,
+ "step": 10130
+ },
+ {
+ "epoch": 22.63,
+ "learning_rate": 8.360903938953914e-06,
+ "loss": 0.0544,
+ "step": 10140
+ },
+ {
+ "epoch": 22.66,
+ "learning_rate": 8.343077240526522e-06,
+ "loss": 0.0684,
+ "step": 10150
+ },
+ {
+ "epoch": 22.68,
+ "learning_rate": 8.325255954655093e-06,
+ "loss": 0.0971,
+ "step": 10160
+ },
+ {
+ "epoch": 22.7,
+ "learning_rate": 8.307440139555192e-06,
+ "loss": 0.0519,
+ "step": 10170
+ },
+ {
+ "epoch": 22.72,
+ "learning_rate": 8.289629853424526e-06,
+ "loss": 0.057,
+ "step": 10180
+ },
+ {
+ "epoch": 22.75,
+ "learning_rate": 8.271825154442732e-06,
+ "loss": 0.0626,
+ "step": 10190
+ },
+ {
+ "epoch": 22.77,
+ "learning_rate": 8.2540261007712e-06,
+ "loss": 0.0477,
+ "step": 10200
+ },
+ {
+ "epoch": 22.79,
+ "learning_rate": 8.236232750552881e-06,
+ "loss": 0.0616,
+ "step": 10210
+ },
+ {
+ "epoch": 22.81,
+ "learning_rate": 8.218445161912088e-06,
+ "loss": 0.0492,
+ "step": 10220
+ },
+ {
+ "epoch": 22.83,
+ "learning_rate": 8.20066339295432e-06,
+ "loss": 0.0723,
+ "step": 10230
+ },
+ {
+ "epoch": 22.86,
+ "learning_rate": 8.182887501766059e-06,
+ "loss": 0.0562,
+ "step": 10240
+ },
+ {
+ "epoch": 22.88,
+ "learning_rate": 8.165117546414595e-06,
+ "loss": 0.0536,
+ "step": 10250
+ },
+ {
+ "epoch": 22.9,
+ "learning_rate": 8.147353584947818e-06,
+ "loss": 0.0571,
+ "step": 10260
+ },
+ {
+ "epoch": 22.92,
+ "learning_rate": 8.129595675394045e-06,
+ "loss": 0.0767,
+ "step": 10270
+ },
+ {
+ "epoch": 22.95,
+ "learning_rate": 8.11184387576182e-06,
+ "loss": 0.0885,
+ "step": 10280
+ },
+ {
+ "epoch": 22.97,
+ "learning_rate": 8.094098244039734e-06,
+ "loss": 0.0684,
+ "step": 10290
+ },
+ {
+ "epoch": 22.99,
+ "learning_rate": 8.076358838196216e-06,
+ "loss": 0.0601,
+ "step": 10300
+ },
+ {
+ "epoch": 23.01,
+ "learning_rate": 8.058625716179375e-06,
+ "loss": 0.0388,
+ "step": 10310
+ },
+ {
+ "epoch": 23.04,
+ "learning_rate": 8.04089893591678e-06,
+ "loss": 0.0422,
+ "step": 10320
+ },
+ {
+ "epoch": 23.06,
+ "learning_rate": 8.023178555315291e-06,
+ "loss": 0.0459,
+ "step": 10330
+ },
+ {
+ "epoch": 23.08,
+ "learning_rate": 8.005464632260862e-06,
+ "loss": 0.0329,
+ "step": 10340
+ },
+ {
+ "epoch": 23.1,
+ "learning_rate": 7.987757224618346e-06,
+ "loss": 0.051,
+ "step": 10350
+ },
+ {
+ "epoch": 23.12,
+ "learning_rate": 7.970056390231323e-06,
+ "loss": 0.0545,
+ "step": 10360
+ },
+ {
+ "epoch": 23.15,
+ "learning_rate": 7.952362186921889e-06,
+ "loss": 0.0463,
+ "step": 10370
+ },
+ {
+ "epoch": 23.17,
+ "learning_rate": 7.934674672490488e-06,
+ "loss": 0.0614,
+ "step": 10380
+ },
+ {
+ "epoch": 23.19,
+ "learning_rate": 7.916993904715708e-06,
+ "loss": 0.0491,
+ "step": 10390
+ },
+ {
+ "epoch": 23.21,
+ "learning_rate": 7.899319941354107e-06,
+ "loss": 0.0441,
+ "step": 10400
+ },
+ {
+ "epoch": 23.24,
+ "learning_rate": 7.881652840140001e-06,
+ "loss": 0.0505,
+ "step": 10410
+ },
+ {
+ "epoch": 23.26,
+ "learning_rate": 7.863992658785302e-06,
+ "loss": 0.0517,
+ "step": 10420
+ },
+ {
+ "epoch": 23.28,
+ "learning_rate": 7.846339454979312e-06,
+ "loss": 0.0604,
+ "step": 10430
+ },
+ {
+ "epoch": 23.3,
+ "learning_rate": 7.828693286388542e-06,
+ "loss": 0.0441,
+ "step": 10440
+ },
+ {
+ "epoch": 23.33,
+ "learning_rate": 7.811054210656526e-06,
+ "loss": 0.0535,
+ "step": 10450
+ },
+ {
+ "epoch": 23.35,
+ "learning_rate": 7.793422285403614e-06,
+ "loss": 0.0534,
+ "step": 10460
+ },
+ {
+ "epoch": 23.37,
+ "learning_rate": 7.775797568226816e-06,
+ "loss": 0.0566,
+ "step": 10470
+ },
+ {
+ "epoch": 23.39,
+ "learning_rate": 7.758180116699578e-06,
+ "loss": 0.085,
+ "step": 10480
+ },
+ {
+ "epoch": 23.42,
+ "learning_rate": 7.74056998837163e-06,
+ "loss": 0.0563,
+ "step": 10490
+ },
+ {
+ "epoch": 23.44,
+ "learning_rate": 7.722967240768761e-06,
+ "loss": 0.0613,
+ "step": 10500
+ },
+ {
+ "epoch": 23.46,
+ "learning_rate": 7.705371931392668e-06,
+ "loss": 0.0413,
+ "step": 10510
+ },
+ {
+ "epoch": 23.48,
+ "learning_rate": 7.687784117720736e-06,
+ "loss": 0.0538,
+ "step": 10520
+ },
+ {
+ "epoch": 23.5,
+ "learning_rate": 7.670203857205877e-06,
+ "loss": 0.0576,
+ "step": 10530
+ },
+ {
+ "epoch": 23.53,
+ "learning_rate": 7.652631207276311e-06,
+ "loss": 0.0467,
+ "step": 10540
+ },
+ {
+ "epoch": 23.55,
+ "learning_rate": 7.635066225335417e-06,
+ "loss": 0.0404,
+ "step": 10550
+ },
+ {
+ "epoch": 23.57,
+ "learning_rate": 7.617508968761519e-06,
+ "loss": 0.0625,
+ "step": 10560
+ },
+ {
+ "epoch": 23.59,
+ "learning_rate": 7.599959494907695e-06,
+ "loss": 0.0598,
+ "step": 10570
+ },
+ {
+ "epoch": 23.62,
+ "learning_rate": 7.582417861101614e-06,
+ "loss": 0.0449,
+ "step": 10580
+ },
+ {
+ "epoch": 23.64,
+ "learning_rate": 7.564884124645325e-06,
+ "loss": 0.072,
+ "step": 10590
+ },
+ {
+ "epoch": 23.66,
+ "learning_rate": 7.547358342815089e-06,
+ "loss": 0.0607,
+ "step": 10600
+ },
+ {
+ "epoch": 23.68,
+ "learning_rate": 7.5298405728611645e-06,
+ "loss": 0.0687,
+ "step": 10610
+ },
+ {
+ "epoch": 23.71,
+ "learning_rate": 7.512330872007659e-06,
+ "loss": 0.0524,
+ "step": 10620
+ },
+ {
+ "epoch": 23.73,
+ "learning_rate": 7.494829297452306e-06,
+ "loss": 0.048,
+ "step": 10630
+ },
+ {
+ "epoch": 23.75,
+ "learning_rate": 7.4773359063663045e-06,
+ "loss": 0.0569,
+ "step": 10640
+ },
+ {
+ "epoch": 23.77,
+ "learning_rate": 7.459850755894108e-06,
+ "loss": 0.0585,
+ "step": 10650
+ },
+ {
+ "epoch": 23.79,
+ "learning_rate": 7.442373903153266e-06,
+ "loss": 0.0859,
+ "step": 10660
+ },
+ {
+ "epoch": 23.82,
+ "learning_rate": 7.424905405234209e-06,
+ "loss": 0.0498,
+ "step": 10670
+ },
+ {
+ "epoch": 23.84,
+ "learning_rate": 7.407445319200083e-06,
+ "loss": 0.0523,
+ "step": 10680
+ },
+ {
+ "epoch": 23.86,
+ "learning_rate": 7.38999370208656e-06,
+ "loss": 0.0711,
+ "step": 10690
+ },
+ {
+ "epoch": 23.88,
+ "learning_rate": 7.37255061090163e-06,
+ "loss": 0.0398,
+ "step": 10700
+ },
+ {
+ "epoch": 23.91,
+ "learning_rate": 7.355116102625451e-06,
+ "loss": 0.0765,
+ "step": 10710
+ },
+ {
+ "epoch": 23.93,
+ "learning_rate": 7.337690234210132e-06,
+ "loss": 0.0547,
+ "step": 10720
+ },
+ {
+ "epoch": 23.95,
+ "learning_rate": 7.320273062579568e-06,
+ "loss": 0.055,
+ "step": 10730
+ },
+ {
+ "epoch": 23.97,
+ "learning_rate": 7.3028646446292295e-06,
+ "loss": 0.0524,
+ "step": 10740
+ },
+ {
+ "epoch": 24.0,
+ "learning_rate": 7.28546503722601e-06,
+ "loss": 0.0479,
+ "step": 10750
+ },
+ {
+ "epoch": 24.02,
+ "learning_rate": 7.268074297208008e-06,
+ "loss": 0.0418,
+ "step": 10760
+ },
+ {
+ "epoch": 24.04,
+ "learning_rate": 7.250692481384366e-06,
+ "loss": 0.0531,
+ "step": 10770
+ },
+ {
+ "epoch": 24.06,
+ "learning_rate": 7.233319646535067e-06,
+ "loss": 0.0401,
+ "step": 10780
+ },
+ {
+ "epoch": 24.08,
+ "learning_rate": 7.21595584941076e-06,
+ "loss": 0.0646,
+ "step": 10790
+ },
+ {
+ "epoch": 24.11,
+ "learning_rate": 7.198601146732573e-06,
+ "loss": 0.0477,
+ "step": 10800
+ },
+ {
+ "epoch": 24.13,
+ "learning_rate": 7.181255595191919e-06,
+ "loss": 0.0331,
+ "step": 10810
+ },
+ {
+ "epoch": 24.15,
+ "learning_rate": 7.1639192514503265e-06,
+ "loss": 0.0463,
+ "step": 10820
+ },
+ {
+ "epoch": 24.17,
+ "learning_rate": 7.146592172139234e-06,
+ "loss": 0.0472,
+ "step": 10830
+ },
+ {
+ "epoch": 24.2,
+ "learning_rate": 7.129274413859832e-06,
+ "loss": 0.0462,
+ "step": 10840
+ },
+ {
+ "epoch": 24.22,
+ "learning_rate": 7.111966033182845e-06,
+ "loss": 0.0404,
+ "step": 10850
+ },
+ {
+ "epoch": 24.24,
+ "learning_rate": 7.094667086648381e-06,
+ "loss": 0.0602,
+ "step": 10860
+ },
+ {
+ "epoch": 24.26,
+ "learning_rate": 7.077377630765716e-06,
+ "loss": 0.044,
+ "step": 10870
+ },
+ {
+ "epoch": 24.29,
+ "learning_rate": 7.060097722013137e-06,
+ "loss": 0.0413,
+ "step": 10880
+ },
+ {
+ "epoch": 24.31,
+ "learning_rate": 7.042827416837728e-06,
+ "loss": 0.0369,
+ "step": 10890
+ },
+ {
+ "epoch": 24.33,
+ "learning_rate": 7.025566771655219e-06,
+ "loss": 0.055,
+ "step": 10900
+ },
+ {
+ "epoch": 24.35,
+ "learning_rate": 7.00831584284977e-06,
+ "loss": 0.0454,
+ "step": 10910
+ },
+ {
+ "epoch": 24.38,
+ "learning_rate": 6.991074686773809e-06,
+ "loss": 0.0475,
+ "step": 10920
+ },
+ {
+ "epoch": 24.4,
+ "learning_rate": 6.973843359747845e-06,
+ "loss": 0.0545,
+ "step": 10930
+ },
+ {
+ "epoch": 24.42,
+ "learning_rate": 6.95662191806026e-06,
+ "loss": 0.0459,
+ "step": 10940
+ },
+ {
+ "epoch": 24.44,
+ "learning_rate": 6.939410417967168e-06,
+ "loss": 0.053,
+ "step": 10950
+ },
+ {
+ "epoch": 24.46,
+ "learning_rate": 6.922208915692186e-06,
+ "loss": 0.0417,
+ "step": 10960
+ },
+ {
+ "epoch": 24.49,
+ "learning_rate": 6.905017467426291e-06,
+ "loss": 0.0568,
+ "step": 10970
+ },
+ {
+ "epoch": 24.51,
+ "learning_rate": 6.887836129327602e-06,
+ "loss": 0.076,
+ "step": 10980
+ },
+ {
+ "epoch": 24.53,
+ "learning_rate": 6.870664957521225e-06,
+ "loss": 0.0478,
+ "step": 10990
+ },
+ {
+ "epoch": 24.55,
+ "learning_rate": 6.85350400809904e-06,
+ "loss": 0.0418,
+ "step": 11000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 6.444053730584166e+16,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s2_en/training_args.bin b/s2_en/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..12611490df2adf725c294654cec732ca6eb72544
--- /dev/null
+++ b/s2_en/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb231ed72d187834e038ca24c913fab03998bea4ceebcd929ea504841108c3c1
+size 6264
diff --git a/s2_en/zero_to_fp32.py b/s2_en/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..0e759146cadd92ddfefab3680146c2bd6a2b5c04
--- /dev/null
+++ b/s2_en/zero_to_fp32.py
@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+# python zero_to_fp32.py . output_dir/
+# or
+# python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+ buffers: dict()
+ param_shapes: dict()
+ shared_params: list
+ ds_version: int
+ frozen_param_shapes: dict()
+ frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+ return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+ '''
+ alist.sort(key=natural_keys) sorts in human order
+ http://nedbatchelder.com/blog/200712/human_sorting.html
+ (See Toothy's implementation in the comments)
+ '''
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+ if not os.path.isdir(checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+ # there should be only one file
+ if zero_stage <= 2:
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+ elif zero_stage == 3:
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+ if not os.path.exists(file):
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+ return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+ # XXX: need to test that this simple glob rule works for multi-node setup too
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+ if len(ckpt_files) == 0:
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+ return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+ zero_model_states = []
+ for file in files:
+ state_dict = torch.load(file, map_location=device, weights_only=False)
+
+ if BUFFER_NAMES not in state_dict:
+ raise ValueError(f"{file} is not a model state checkpoint")
+ buffer_names = state_dict[BUFFER_NAMES]
+ if debug:
+ print("Found buffers:", buffer_names)
+
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+ param_shapes = state_dict[PARAM_SHAPES]
+
+ # collect parameters that are included in param_shapes
+ param_names = []
+ for s in param_shapes:
+ for name in s.keys():
+ param_names.append(name)
+
+ # update with frozen parameters
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+ if frozen_param_shapes is not None:
+ if debug:
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+ param_names += list(frozen_param_shapes.keys())
+
+ # handle shared params
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+ ds_version = state_dict.get(DS_VERSION, None)
+
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+ z_model_state = zero_model_state(buffers=buffers,
+ param_shapes=param_shapes,
+ shared_params=shared_params,
+ ds_version=ds_version,
+ frozen_param_shapes=frozen_param_shapes,
+ frozen_param_fragments=frozen_param_fragments)
+ zero_model_states.append(z_model_state)
+
+ return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+ total_files = len(files)
+ state_dicts = []
+ for f in tqdm(files, desc='Loading checkpoint shards'):
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+ # and also handle the case where it was already removed by another helper script
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+ state_dicts.append(state_dict)
+
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
+ # use the max of the partition_count to get the dp world_size.
+
+ if type(world_size) is list:
+ world_size = max(world_size)
+
+ if world_size != total_files:
+ raise ValueError(
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+ )
+
+ # the groups are named differently in each stage
+ if zero_stage <= 2:
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+ elif zero_stage == 3:
+ fp32_groups_key = FP32_FLAT_GROUPS
+ else:
+ raise ValueError(f"unknown zero stage {zero_stage}")
+
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+ return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+ """
+ Returns fp32 state_dict reconstructed from ds checkpoint
+
+ Args:
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+ """
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+ optim_files = get_optim_files(ds_checkpoint_dir)
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+ model_files = get_model_state_files(ds_checkpoint_dir)
+
+ zero_model_states = parse_model_states(model_files)
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+ if zero_stage <= 2:
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+ elif zero_stage == 3:
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+ if debug:
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ state_dict[name] = frozen_param_fragments[name]
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+ attr = getattr(obj, fn, None)
+ return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+
+ # Reconstruction protocol:
+ #
+ # XXX: document this
+
+ if debug:
+ for i in range(world_size):
+ for j in range(len(fp32_flat_groups[0])):
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+ # XXX: memory usage doubles here (zero2)
+ num_param_groups = len(fp32_flat_groups[0])
+ merged_single_partition_of_fp32_groups = []
+ for i in range(num_param_groups):
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+ avail_numel = sum(
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+ if debug:
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+ # not asserting if there is a mismatch due to possible padding
+ print(f"Have {avail_numel} numels to process.")
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ total_numel = 0
+ total_params = 0
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+ offset = 0
+ avail_numel = full_single_fp32_vector.numel()
+ for name, shape in shapes.items():
+
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+ total_numel += unpartitioned_numel
+ total_params += 1
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+ offset += unpartitioned_numel
+
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+ # live optimizer object, so we are checking that the numbers are within the right range
+ align_to = 2 * world_size
+
+ def zero2_align(x):
+ return align_to * math.ceil(x / align_to)
+
+ if debug:
+ print(f"original offset={offset}, avail_numel={avail_numel}")
+
+ offset = zero2_align(offset)
+ avail_numel = zero2_align(avail_numel)
+
+ if debug:
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+ remainder = unpartitioned_numel % world_size
+ padding_numel = (world_size - remainder) if remainder else 0
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+ return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ if debug:
+ for i in range(world_size):
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+ """
+ A pseudo tensor that collects partitioned weights.
+ It is more memory efficient when there are multiple groups.
+ """
+
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+ self.flat_groups = flat_groups
+ self.flat_groups_offset = flat_groups_offset
+ self.offset = offset
+ self.partitioned_numel = partitioned_numel
+ self.shape = shape
+ self.dtype = self.flat_groups[0][0].dtype
+
+ def contiguous(self):
+ """
+ Merge partitioned weights from flat_groups into a single tensor.
+ """
+ end_idx = self.offset + self.partitioned_numel
+ world_size = len(self.flat_groups)
+ pad_flat_param_chunks = []
+
+ for rank_i in range(world_size):
+ # for each rank, we need to collect weights from related group/groups
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
+ start_group_id = None
+ end_group_id = None
+ for group_id in range(len(self.flat_groups_offset)):
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+ start_group_id = group_id
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+ end_group_id = group_id
+ break
+ # collect weights from related group/groups
+ for group_id in range(start_group_id, end_group_id + 1):
+ flat_tensor = flat_groups_at_rank_i[group_id]
+ start_offset = self.offset - self.flat_groups_offset[group_id]
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+ # collect weights from all ranks
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+ return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+ # param, re-consolidating each param, while dealing with padding if any
+
+ # merge list of dicts, preserving order
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+ if debug:
+ for i in range(world_size):
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+ wanted_params = len(param_shapes)
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+ # not asserting if there is a mismatch due to possible padding
+ avail_numel = fp32_flat_groups[0].numel() * world_size
+ print(f"Trainable params: Have {avail_numel} numels to process.")
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ offset = 0
+ total_numel = 0
+ total_params = 0
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+ total_params += 1
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ # memory efficient tensor
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+ state_dict[name] = tensor
+ offset += partitioned_numel
+
+ offset *= world_size
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+ """
+ Convert state_dict of GatheredTensor to torch tensor
+ """
+ torch_state_dict = {}
+ converted_tensors = {}
+ for name, tensor in state_dict.items():
+ tensor_id = id(tensor)
+ if tensor_id in converted_tensors: # shared tensors
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+ torch_state_dict[name] = shared_tensor
+ else:
+ converted_tensors[tensor_id] = name
+ if return_empty_tensor:
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+ else:
+ torch_state_dict[name] = tensor.contiguous()
+ return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag=None,
+ exclude_frozen_parameters=False,
+ lazy_mode=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+ via a model hub.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+ Returns:
+ - pytorch ``state_dict``
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ # do the training and checkpoint saving
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+ model = model.cpu() # move to cpu
+ model.load_state_dict(state_dict)
+ # submit to model hub or save the model to share with others
+
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
+ application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+ the checkpoint. Or you can load state_dict in lazy mode ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+ for name, lazy_tensor in state_dict.item():
+ tensor = lazy_tensor.contiguous() # to cpu
+ print(name, tensor)
+ # del tensor to release memory if it no longer in use
+ """
+ if tag is None:
+ latest_path = os.path.join(checkpoint_dir, 'latest')
+ if os.path.isfile(latest_path):
+ with open(latest_path, 'r') as fd:
+ tag = fd.read().strip()
+ else:
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+ if not os.path.isdir(ds_checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+ if lazy_mode:
+ return state_dict
+ else:
+ return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+ output_dir,
+ max_shard_size="5GB",
+ safe_serialization=False,
+ tag=None,
+ exclude_frozen_parameters=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ """
+
+ # Dependency pre-check
+ if safe_serialization:
+ try:
+ from safetensors.torch import save_file
+ except ImportError:
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
+ raise
+ if max_shard_size is not None:
+ try:
+ from huggingface_hub import split_torch_state_dict_into_shards
+ except ImportError:
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+ raise
+
+ # Convert zero checkpoint to state_dict
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag,
+ exclude_frozen_parameters,
+ lazy_mode=True)
+
+ # Shard the model if it is too big.
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+ if max_shard_size is not None:
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+ # an memory-efficient approach for sharding
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+ filename_pattern=filename_pattern,
+ max_shard_size=max_shard_size)
+ else:
+ from collections import namedtuple
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+ state_dict_split = StateDictSplit(is_sharded=False,
+ filename_to_tensors={weights_name: list(state_dict.keys())})
+
+ # Save the model by shard
+ os.makedirs(output_dir, exist_ok=True)
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+ shard_state_dict = to_torch_tensor(shard_state_dict)
+ output_path = os.path.join(output_dir, shard_file)
+ if safe_serialization:
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+ else:
+ torch.save(shard_state_dict, output_path)
+ # release the memory of current shard
+ for tensor_name in list(shard_state_dict.keys()):
+ del state_dict[tensor_name]
+ del shard_state_dict[tensor_name]
+ del shard_state_dict
+ gc.collect()
+
+ # Save index if sharded
+ if state_dict_split.is_sharded:
+ index = {
+ "metadata": state_dict_split.metadata,
+ "weight_map": state_dict_split.tensor_to_filename,
+ }
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+ save_index_file = os.path.join(output_dir, save_index_file)
+ with open(save_index_file, "w", encoding="utf-8") as f:
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+ f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+ """
+ 1. Put the provided model to cpu
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+ 3. Load it into the provided model
+
+ Args:
+ - ``model``: the model object to update
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+ Returns:
+ - ``model`: modified model
+
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+ conveniently placed for you in the checkpoint folder.
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+ # submit to model hub or save the model to share with others
+
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ """
+ logger.info(f"Extracting fp32 weights")
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+ logger.info(f"Overwriting model with fp32 weights")
+ model = model.cpu()
+ model.load_state_dict(state_dict, strict=False)
+
+ return model
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser()
+ parser.add_argument("checkpoint_dir",
+ type=str,
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+ parser.add_argument("output_dir",
+ type=str,
+ help="directory to the pytorch fp32 state_dict output files"
+ "(e.g. path/checkpoint-12-output/)")
+ parser.add_argument(
+ "--max_shard_size",
+ type=str,
+ default="5GB",
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+ "without CPU OOM issues.")
+ parser.add_argument(
+ "--safe_serialization",
+ default=False,
+ action='store_true',
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+ parser.add_argument("-t",
+ "--tag",
+ type=str,
+ default=None,
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+ args = parser.parse_args()
+
+ debug = args.debug
+
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+ args.output_dir,
+ max_shard_size=args.max_shard_size,
+ safe_serialization=args.safe_serialization,
+ tag=args.tag,
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
diff --git a/s3/README.md b/s3/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s3/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s3/adapter_config.json b/s3/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..01ee4b8aca52410dbed3c8f18b9b57163c93bf06
--- /dev/null
+++ b/s3/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "k_proj",
+ "down_proj",
+ "gate_proj",
+ "v_proj",
+ "q_proj",
+ "up_proj",
+ "o_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s3/adapter_model.bin b/s3/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..c7e43204df06ab2de9fd7bba5e8672828c253609
--- /dev/null
+++ b/s3/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e22332b2f3d384bf190b30314fbdcf66329025e225ce4db9a42337f31b4bc134
+size 639786637
diff --git a/s3/config.json b/s3/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s3/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s3/global_step8000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s3/global_step8000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e91ba84932489afccee550bfd67b580efa819b52
--- /dev/null
+++ b/s3/global_step8000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bf0aa7542a4e6400bc6e919eb2b11915841ed51640c1536afde5e9adbee9da6
+size 4089599575
diff --git a/s3/global_step8000/mp_rank_00_model_states.pt b/s3/global_step8000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..54f07b6fc74112320011d167ec5445b7ee42af59
--- /dev/null
+++ b/s3/global_step8000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f1b1fa71ac60f2752968b9460034538a3da85fa568b02ce0b100869896ce246
+size 28850200126
diff --git a/s3/latest b/s3/latest
new file mode 100644
index 0000000000000000000000000000000000000000..6c558c76ac9e7515267f79d24d9fde4a7f8688f1
--- /dev/null
+++ b/s3/latest
@@ -0,0 +1 @@
+global_step8000
\ No newline at end of file
diff --git a/s3/non_lora_trainables.bin b/s3/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..07fb0160fd4e5e8a4b9397665c38f7d733b90056
--- /dev/null
+++ b/s3/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a094e2067cc06a07978a15e5f7debc8e30146ddbfed52ac5ca1fddb75717fe52
+size 41961191
diff --git a/s3/rng_state.pth b/s3/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..56b07c17d76cb7d553a46c6ba95eb0c3c23eb448
--- /dev/null
+++ b/s3/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7306a6fab6b9f2f5ab021541dab748c46c613a31be79d659f1f30528994bc7ce
+size 14575
diff --git a/s3/special_tokens_map.json b/s3/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s3/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s3/tokenizer.model b/s3/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s3/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s3/tokenizer_config.json b/s3/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s3/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s3/trainer_state.json b/s3/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6f16bb02afdbe2d6ea88da7d47a9f5598ece1c57
--- /dev/null
+++ b/s3/trainer_state.json
@@ -0,0 +1,4816 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 17.857142857142858,
+ "global_step": 8000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 7.2406,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 6.5969,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.4641,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.0734,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 3.9453,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.4391,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 3.1609,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.9453,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.8047,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 2.2758,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 2.0383,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 1.7414,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.5773,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 1.3703,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.2703,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.1812,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.0328,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 0.943,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 1.1418,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 0.784,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 0.6648,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 0.7697,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 0.6514,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 0.6775,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 0.5932,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 0.6855,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 0.5135,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 0.5949,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 0.5139,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 0.4232,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 0.4325,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 0.4629,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 0.3893,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 0.5352,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 0.5012,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 0.4872,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 0.6109,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 0.4701,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 0.5756,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 0.4864,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 0.5752,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 0.5738,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 0.4296,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 0.4486,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 0.5186,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 0.3414,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 0.4656,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.361,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 0.4122,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 0.331,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.4964,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.3147,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.3456,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.324,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 0.3336,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.3744,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 0.392,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.4078,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 0.4142,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.3998,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.3499,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 0.3383,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.3296,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 0.3723,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.4653,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.3805,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 0.7628,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.3567,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.4469,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.2795,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.5107,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.4593,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 0.4607,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.4134,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.4903,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.3674,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.3483,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.3344,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.2979,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.3349,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.3062,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.3093,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 0.3229,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.3677,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.2956,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.4208,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.2229,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.3343,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.4137,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.2521,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.3819,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.2854,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.2011,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.2441,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.277,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.2837,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.1958,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.2656,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.227,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.2097,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.2186,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.2874,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.1967,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.263,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.2814,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.3031,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.273,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.3278,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.2603,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.2474,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.3549,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.2238,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.2438,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.2068,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.2335,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.2469,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.3177,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.2442,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.2069,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.2246,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.2917,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.2328,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.2774,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.3456,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.2944,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.1965,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.2508,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.2811,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.3268,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.2356,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.2328,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.2515,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.4871,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.2741,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.3244,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.2452,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.1495,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.133,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.2249,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.1724,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.1692,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.1605,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.2,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.2069,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.194,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.2604,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.1781,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.167,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.2232,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.1978,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.2211,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.2264,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.2231,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.203,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.1621,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.2412,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.2393,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.2143,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.197,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.2317,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.1935,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.2081,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.1856,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.1731,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.201,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.251,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.2249,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.1694,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.217,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.1834,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.1892,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.1855,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.1846,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.1942,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.1904,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.2,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.2034,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.2563,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.2639,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.17,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.1453,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.1692,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.1397,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.1819,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.1902,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.1498,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.1418,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.1641,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.1604,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.1473,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.1341,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.1327,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.1508,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.175,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.1844,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.1875,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.1373,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.126,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.1428,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.2099,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.1505,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.1498,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.158,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.13,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.1162,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.1587,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.163,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.1551,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.1859,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.1807,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.148,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.1809,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.1431,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.1473,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.1512,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.1578,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.1495,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.167,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.1947,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.1611,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.1301,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.1348,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.1641,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.1603,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.1247,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.1154,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.1265,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.1025,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.11,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.1142,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.0908,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.1159,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.1335,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.1093,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.1255,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.1081,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.144,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.1115,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.1108,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.1386,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.1375,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.1292,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.1444,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.1225,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.1526,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.131,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.114,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.1298,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.0971,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.1352,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.1494,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.134,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.1332,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.1328,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.1235,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.1609,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.1408,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.1463,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.1548,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.1282,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.1387,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.1259,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.144,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.1063,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.1188,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.1581,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.1357,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.1446,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.1305,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.1042,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.0902,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.0978,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.0895,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.0901,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.0958,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.0952,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.0866,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.1057,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.0941,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.1057,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.121,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.1101,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.1027,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.0803,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.1232,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.0986,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.1098,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.1063,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.0962,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.0951,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.1091,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.096,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.1044,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.0912,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.0998,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.1122,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.098,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.0913,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.1545,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.0895,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.1202,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.1023,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.0949,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.1018,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.1039,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.1055,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.1352,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.1144,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.1253,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.1012,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.125,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.1364,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.1229,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.1061,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.0676,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.0866,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.073,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.0978,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.0781,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.0948,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.0655,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.0934,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.1012,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.0872,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.0904,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.093,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.0921,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.0907,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.0783,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.0941,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.1035,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.0814,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.0937,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.0916,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.0981,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.0897,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.0911,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.1,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.1036,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.1104,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.107,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.1009,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.0983,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.0831,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.0984,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.0921,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.1029,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.1106,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.13,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.1031,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.0851,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.0879,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.0817,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.1026,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.0979,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.0904,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.0887,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.0993,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.0935,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.081,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.0646,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.075,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.0654,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.0885,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.0861,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.0781,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.0802,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.0633,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.0585,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.0751,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.0834,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.0924,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.0761,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.0866,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.0847,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.0706,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.0917,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.0748,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.0767,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.0702,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.0851,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.0713,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.0869,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.0823,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.0872,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.0775,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.0862,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.1073,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.0979,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.0851,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.0912,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.0803,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.0788,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.0764,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.1058,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.1,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.09,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.0848,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.0852,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.1047,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.096,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.0942,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.0985,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.0743,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.0912,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.0667,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.0882,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.0683,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.0732,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.0658,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.0709,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.0759,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.0793,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.0936,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.0862,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.0682,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.0783,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.0736,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.063,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.0746,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.0806,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.081,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.0712,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.0719,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.0617,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.1026,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.0749,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.0857,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.0699,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.0845,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.0678,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.0859,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.0802,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.0708,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.0808,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.0799,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.0734,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.0818,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.0706,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.0775,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.0839,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.0785,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.0739,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.0892,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.0835,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.0785,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.0927,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.1003,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.0677,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.076,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.0721,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.0616,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.0682,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.0621,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.07,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.0971,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.0721,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.0736,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.0674,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.0675,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.0677,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.0677,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.0595,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.0675,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.0688,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.0588,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.0733,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.0745,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.0731,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.0991,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.0693,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.0782,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.0831,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.0764,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.0632,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.06,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.0835,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.0731,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.0789,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.0813,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.0744,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.0639,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.076,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.0798,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.0736,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.0759,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.0781,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.0787,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.0717,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.0961,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.0717,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.0863,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.0834,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.0523,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.0642,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.0741,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.0673,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.0599,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.067,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.0747,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.0662,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.0572,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.0635,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.0675,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.0637,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.0649,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.0656,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.0773,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.0664,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.0864,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.0873,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.0736,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.0716,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.0653,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.0621,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.067,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.0693,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.0696,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.079,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.0593,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.0737,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.0909,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.0885,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.0679,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.073,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.0751,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.0575,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.064,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.0773,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.059,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.0674,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.0746,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.0781,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.0619,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.0694,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.074,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.0714,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.0652,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.0722,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.0635,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.0502,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.0481,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.0622,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.0765,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.0598,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.0578,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.0714,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.0503,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.0532,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.0534,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.0687,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.0721,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.0604,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.0712,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.0704,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0701,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.0629,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0712,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.0794,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.0689,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.067,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.0595,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.0716,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.074,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.0628,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.0742,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.0745,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.0544,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.0643,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.061,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.0602,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.069,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.0646,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.0649,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.0705,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.0566,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.0815,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.0844,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.0728,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.0701,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.0592,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.0654,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.0604,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.0587,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.0545,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.0531,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.0651,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.0521,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.0496,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.0479,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.0563,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.061,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.0489,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.0558,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.0708,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.0541,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.074,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.075,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0741,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.0765,
+ "step": 6000
+ },
+ {
+ "epoch": 13.42,
+ "learning_rate": 1.5495257562515308e-05,
+ "loss": 0.0761,
+ "step": 6010
+ },
+ {
+ "epoch": 13.44,
+ "learning_rate": 1.5480148332320562e-05,
+ "loss": 0.0858,
+ "step": 6020
+ },
+ {
+ "epoch": 13.46,
+ "learning_rate": 1.5465021200501046e-05,
+ "loss": 0.0736,
+ "step": 6030
+ },
+ {
+ "epoch": 13.48,
+ "learning_rate": 1.5449876216471525e-05,
+ "loss": 0.0637,
+ "step": 6040
+ },
+ {
+ "epoch": 13.5,
+ "learning_rate": 1.5434713429705078e-05,
+ "loss": 0.082,
+ "step": 6050
+ },
+ {
+ "epoch": 13.53,
+ "learning_rate": 1.5419532889732943e-05,
+ "loss": 0.0609,
+ "step": 6060
+ },
+ {
+ "epoch": 13.55,
+ "learning_rate": 1.540433464614435e-05,
+ "loss": 0.0865,
+ "step": 6070
+ },
+ {
+ "epoch": 13.57,
+ "learning_rate": 1.5389118748586357e-05,
+ "loss": 0.0696,
+ "step": 6080
+ },
+ {
+ "epoch": 13.59,
+ "learning_rate": 1.537388524676369e-05,
+ "loss": 0.0639,
+ "step": 6090
+ },
+ {
+ "epoch": 13.62,
+ "learning_rate": 1.5358634190438592e-05,
+ "loss": 0.0698,
+ "step": 6100
+ },
+ {
+ "epoch": 13.64,
+ "learning_rate": 1.5343365629430638e-05,
+ "loss": 0.0603,
+ "step": 6110
+ },
+ {
+ "epoch": 13.66,
+ "learning_rate": 1.5328079613616592e-05,
+ "loss": 0.0865,
+ "step": 6120
+ },
+ {
+ "epoch": 13.68,
+ "learning_rate": 1.531277619293023e-05,
+ "loss": 0.0667,
+ "step": 6130
+ },
+ {
+ "epoch": 13.71,
+ "learning_rate": 1.5297455417362194e-05,
+ "loss": 0.0736,
+ "step": 6140
+ },
+ {
+ "epoch": 13.73,
+ "learning_rate": 1.52821173369598e-05,
+ "loss": 0.0717,
+ "step": 6150
+ },
+ {
+ "epoch": 13.75,
+ "learning_rate": 1.526676200182691e-05,
+ "loss": 0.0644,
+ "step": 6160
+ },
+ {
+ "epoch": 13.77,
+ "learning_rate": 1.5251389462123748e-05,
+ "loss": 0.06,
+ "step": 6170
+ },
+ {
+ "epoch": 13.79,
+ "learning_rate": 1.5235999768066729e-05,
+ "loss": 0.0661,
+ "step": 6180
+ },
+ {
+ "epoch": 13.82,
+ "learning_rate": 1.5220592969928313e-05,
+ "loss": 0.0659,
+ "step": 6190
+ },
+ {
+ "epoch": 13.84,
+ "learning_rate": 1.5205169118036831e-05,
+ "loss": 0.0663,
+ "step": 6200
+ },
+ {
+ "epoch": 13.86,
+ "learning_rate": 1.5189728262776325e-05,
+ "loss": 0.074,
+ "step": 6210
+ },
+ {
+ "epoch": 13.88,
+ "learning_rate": 1.5174270454586375e-05,
+ "loss": 0.0685,
+ "step": 6220
+ },
+ {
+ "epoch": 13.91,
+ "learning_rate": 1.5158795743961942e-05,
+ "loss": 0.0735,
+ "step": 6230
+ },
+ {
+ "epoch": 13.93,
+ "learning_rate": 1.5143304181453204e-05,
+ "loss": 0.0651,
+ "step": 6240
+ },
+ {
+ "epoch": 13.95,
+ "learning_rate": 1.5127795817665389e-05,
+ "loss": 0.0663,
+ "step": 6250
+ },
+ {
+ "epoch": 13.97,
+ "learning_rate": 1.5112270703258602e-05,
+ "loss": 0.0624,
+ "step": 6260
+ },
+ {
+ "epoch": 14.0,
+ "learning_rate": 1.5096728888947669e-05,
+ "loss": 0.0694,
+ "step": 6270
+ },
+ {
+ "epoch": 14.02,
+ "learning_rate": 1.508117042550197e-05,
+ "loss": 0.0576,
+ "step": 6280
+ },
+ {
+ "epoch": 14.04,
+ "learning_rate": 1.5065595363745272e-05,
+ "loss": 0.0617,
+ "step": 6290
+ },
+ {
+ "epoch": 14.06,
+ "learning_rate": 1.505000375455556e-05,
+ "loss": 0.07,
+ "step": 6300
+ },
+ {
+ "epoch": 14.08,
+ "learning_rate": 1.503439564886487e-05,
+ "loss": 0.062,
+ "step": 6310
+ },
+ {
+ "epoch": 14.11,
+ "learning_rate": 1.501877109765914e-05,
+ "loss": 0.0588,
+ "step": 6320
+ },
+ {
+ "epoch": 14.13,
+ "learning_rate": 1.5003130151978012e-05,
+ "loss": 0.0494,
+ "step": 6330
+ },
+ {
+ "epoch": 14.15,
+ "learning_rate": 1.4987472862914697e-05,
+ "loss": 0.0531,
+ "step": 6340
+ },
+ {
+ "epoch": 14.17,
+ "learning_rate": 1.4971799281615782e-05,
+ "loss": 0.0543,
+ "step": 6350
+ },
+ {
+ "epoch": 14.2,
+ "learning_rate": 1.4956109459281083e-05,
+ "loss": 0.0561,
+ "step": 6360
+ },
+ {
+ "epoch": 14.22,
+ "learning_rate": 1.4940403447163467e-05,
+ "loss": 0.0581,
+ "step": 6370
+ },
+ {
+ "epoch": 14.24,
+ "learning_rate": 1.4924681296568689e-05,
+ "loss": 0.0728,
+ "step": 6380
+ },
+ {
+ "epoch": 14.26,
+ "learning_rate": 1.4908943058855213e-05,
+ "loss": 0.0556,
+ "step": 6390
+ },
+ {
+ "epoch": 14.29,
+ "learning_rate": 1.4893188785434067e-05,
+ "loss": 0.0574,
+ "step": 6400
+ },
+ {
+ "epoch": 14.31,
+ "learning_rate": 1.4877418527768654e-05,
+ "loss": 0.0706,
+ "step": 6410
+ },
+ {
+ "epoch": 14.33,
+ "learning_rate": 1.4861632337374596e-05,
+ "loss": 0.0667,
+ "step": 6420
+ },
+ {
+ "epoch": 14.35,
+ "learning_rate": 1.4845830265819552e-05,
+ "loss": 0.0584,
+ "step": 6430
+ },
+ {
+ "epoch": 14.38,
+ "learning_rate": 1.483001236472307e-05,
+ "loss": 0.054,
+ "step": 6440
+ },
+ {
+ "epoch": 14.4,
+ "learning_rate": 1.4814178685756405e-05,
+ "loss": 0.0609,
+ "step": 6450
+ },
+ {
+ "epoch": 14.42,
+ "learning_rate": 1.4798329280642345e-05,
+ "loss": 0.0667,
+ "step": 6460
+ },
+ {
+ "epoch": 14.44,
+ "learning_rate": 1.4782464201155057e-05,
+ "loss": 0.0561,
+ "step": 6470
+ },
+ {
+ "epoch": 14.46,
+ "learning_rate": 1.476658349911991e-05,
+ "loss": 0.0611,
+ "step": 6480
+ },
+ {
+ "epoch": 14.49,
+ "learning_rate": 1.4750687226413305e-05,
+ "loss": 0.0637,
+ "step": 6490
+ },
+ {
+ "epoch": 14.51,
+ "learning_rate": 1.4734775434962504e-05,
+ "loss": 0.0629,
+ "step": 6500
+ },
+ {
+ "epoch": 14.53,
+ "learning_rate": 1.471884817674546e-05,
+ "loss": 0.0655,
+ "step": 6510
+ },
+ {
+ "epoch": 14.55,
+ "learning_rate": 1.4702905503790668e-05,
+ "loss": 0.0625,
+ "step": 6520
+ },
+ {
+ "epoch": 14.58,
+ "learning_rate": 1.4686947468176955e-05,
+ "loss": 0.0559,
+ "step": 6530
+ },
+ {
+ "epoch": 14.6,
+ "learning_rate": 1.467097412203334e-05,
+ "loss": 0.069,
+ "step": 6540
+ },
+ {
+ "epoch": 14.62,
+ "learning_rate": 1.4654985517538864e-05,
+ "loss": 0.0687,
+ "step": 6550
+ },
+ {
+ "epoch": 14.64,
+ "learning_rate": 1.4638981706922401e-05,
+ "loss": 0.0493,
+ "step": 6560
+ },
+ {
+ "epoch": 14.67,
+ "learning_rate": 1.4622962742462503e-05,
+ "loss": 0.0691,
+ "step": 6570
+ },
+ {
+ "epoch": 14.69,
+ "learning_rate": 1.4606928676487223e-05,
+ "loss": 0.0516,
+ "step": 6580
+ },
+ {
+ "epoch": 14.71,
+ "learning_rate": 1.459087956137394e-05,
+ "loss": 0.0676,
+ "step": 6590
+ },
+ {
+ "epoch": 14.73,
+ "learning_rate": 1.4574815449549209e-05,
+ "loss": 0.0629,
+ "step": 6600
+ },
+ {
+ "epoch": 14.75,
+ "learning_rate": 1.4558736393488553e-05,
+ "loss": 0.0646,
+ "step": 6610
+ },
+ {
+ "epoch": 14.78,
+ "learning_rate": 1.4542642445716326e-05,
+ "loss": 0.0594,
+ "step": 6620
+ },
+ {
+ "epoch": 14.8,
+ "learning_rate": 1.4526533658805517e-05,
+ "loss": 0.0718,
+ "step": 6630
+ },
+ {
+ "epoch": 14.82,
+ "learning_rate": 1.4510410085377606e-05,
+ "loss": 0.055,
+ "step": 6640
+ },
+ {
+ "epoch": 14.84,
+ "learning_rate": 1.4494271778102358e-05,
+ "loss": 0.0549,
+ "step": 6650
+ },
+ {
+ "epoch": 14.87,
+ "learning_rate": 1.4478118789697675e-05,
+ "loss": 0.0587,
+ "step": 6660
+ },
+ {
+ "epoch": 14.89,
+ "learning_rate": 1.4461951172929419e-05,
+ "loss": 0.0508,
+ "step": 6670
+ },
+ {
+ "epoch": 14.91,
+ "learning_rate": 1.4445768980611233e-05,
+ "loss": 0.0717,
+ "step": 6680
+ },
+ {
+ "epoch": 14.93,
+ "learning_rate": 1.4429572265604375e-05,
+ "loss": 0.0635,
+ "step": 6690
+ },
+ {
+ "epoch": 14.96,
+ "learning_rate": 1.4413361080817545e-05,
+ "loss": 0.0743,
+ "step": 6700
+ },
+ {
+ "epoch": 14.98,
+ "learning_rate": 1.4397135479206705e-05,
+ "loss": 0.0793,
+ "step": 6710
+ },
+ {
+ "epoch": 15.0,
+ "learning_rate": 1.4380895513774922e-05,
+ "loss": 0.0636,
+ "step": 6720
+ },
+ {
+ "epoch": 15.02,
+ "learning_rate": 1.436464123757217e-05,
+ "loss": 0.0548,
+ "step": 6730
+ },
+ {
+ "epoch": 15.04,
+ "learning_rate": 1.4348372703695184e-05,
+ "loss": 0.0564,
+ "step": 6740
+ },
+ {
+ "epoch": 15.07,
+ "learning_rate": 1.4332089965287266e-05,
+ "loss": 0.052,
+ "step": 6750
+ },
+ {
+ "epoch": 15.09,
+ "learning_rate": 1.431579307553812e-05,
+ "loss": 0.0621,
+ "step": 6760
+ },
+ {
+ "epoch": 15.11,
+ "learning_rate": 1.429948208768368e-05,
+ "loss": 0.0538,
+ "step": 6770
+ },
+ {
+ "epoch": 15.13,
+ "learning_rate": 1.4283157055005928e-05,
+ "loss": 0.0598,
+ "step": 6780
+ },
+ {
+ "epoch": 15.16,
+ "learning_rate": 1.4266818030832732e-05,
+ "loss": 0.0495,
+ "step": 6790
+ },
+ {
+ "epoch": 15.18,
+ "learning_rate": 1.4250465068537664e-05,
+ "loss": 0.0549,
+ "step": 6800
+ },
+ {
+ "epoch": 15.2,
+ "learning_rate": 1.4234098221539818e-05,
+ "loss": 0.0596,
+ "step": 6810
+ },
+ {
+ "epoch": 15.22,
+ "learning_rate": 1.4217717543303657e-05,
+ "loss": 0.0586,
+ "step": 6820
+ },
+ {
+ "epoch": 15.25,
+ "learning_rate": 1.4201323087338816e-05,
+ "loss": 0.0634,
+ "step": 6830
+ },
+ {
+ "epoch": 15.27,
+ "learning_rate": 1.4184914907199942e-05,
+ "loss": 0.0582,
+ "step": 6840
+ },
+ {
+ "epoch": 15.29,
+ "learning_rate": 1.4168493056486512e-05,
+ "loss": 0.0531,
+ "step": 6850
+ },
+ {
+ "epoch": 15.31,
+ "learning_rate": 1.4152057588842657e-05,
+ "loss": 0.0523,
+ "step": 6860
+ },
+ {
+ "epoch": 15.33,
+ "learning_rate": 1.4135608557956992e-05,
+ "loss": 0.0532,
+ "step": 6870
+ },
+ {
+ "epoch": 15.36,
+ "learning_rate": 1.4119146017562441e-05,
+ "loss": 0.0528,
+ "step": 6880
+ },
+ {
+ "epoch": 15.38,
+ "learning_rate": 1.4102670021436059e-05,
+ "loss": 0.0503,
+ "step": 6890
+ },
+ {
+ "epoch": 15.4,
+ "learning_rate": 1.4086180623398842e-05,
+ "loss": 0.0463,
+ "step": 6900
+ },
+ {
+ "epoch": 15.42,
+ "learning_rate": 1.4069677877315587e-05,
+ "loss": 0.0672,
+ "step": 6910
+ },
+ {
+ "epoch": 15.45,
+ "learning_rate": 1.4053161837094675e-05,
+ "loss": 0.0583,
+ "step": 6920
+ },
+ {
+ "epoch": 15.47,
+ "learning_rate": 1.4036632556687927e-05,
+ "loss": 0.0539,
+ "step": 6930
+ },
+ {
+ "epoch": 15.49,
+ "learning_rate": 1.4020090090090408e-05,
+ "loss": 0.0426,
+ "step": 6940
+ },
+ {
+ "epoch": 15.51,
+ "learning_rate": 1.4003534491340259e-05,
+ "loss": 0.0569,
+ "step": 6950
+ },
+ {
+ "epoch": 15.54,
+ "learning_rate": 1.3986965814518521e-05,
+ "loss": 0.0527,
+ "step": 6960
+ },
+ {
+ "epoch": 15.56,
+ "learning_rate": 1.3970384113748951e-05,
+ "loss": 0.0593,
+ "step": 6970
+ },
+ {
+ "epoch": 15.58,
+ "learning_rate": 1.3953789443197857e-05,
+ "loss": 0.0618,
+ "step": 6980
+ },
+ {
+ "epoch": 15.6,
+ "learning_rate": 1.3937181857073912e-05,
+ "loss": 0.0595,
+ "step": 6990
+ },
+ {
+ "epoch": 15.62,
+ "learning_rate": 1.3920561409627974e-05,
+ "loss": 0.0626,
+ "step": 7000
+ },
+ {
+ "epoch": 15.65,
+ "learning_rate": 1.3903928155152926e-05,
+ "loss": 0.0622,
+ "step": 7010
+ },
+ {
+ "epoch": 15.67,
+ "learning_rate": 1.3887282147983472e-05,
+ "loss": 0.0583,
+ "step": 7020
+ },
+ {
+ "epoch": 15.69,
+ "learning_rate": 1.3870623442495987e-05,
+ "loss": 0.0582,
+ "step": 7030
+ },
+ {
+ "epoch": 15.71,
+ "learning_rate": 1.3853952093108323e-05,
+ "loss": 0.0948,
+ "step": 7040
+ },
+ {
+ "epoch": 15.74,
+ "learning_rate": 1.3837268154279628e-05,
+ "loss": 0.101,
+ "step": 7050
+ },
+ {
+ "epoch": 15.76,
+ "learning_rate": 1.3820571680510187e-05,
+ "loss": 0.0675,
+ "step": 7060
+ },
+ {
+ "epoch": 15.78,
+ "learning_rate": 1.3803862726341224e-05,
+ "loss": 0.0683,
+ "step": 7070
+ },
+ {
+ "epoch": 15.8,
+ "learning_rate": 1.3787141346354733e-05,
+ "loss": 0.0725,
+ "step": 7080
+ },
+ {
+ "epoch": 15.83,
+ "learning_rate": 1.3770407595173301e-05,
+ "loss": 0.0792,
+ "step": 7090
+ },
+ {
+ "epoch": 15.85,
+ "learning_rate": 1.375366152745992e-05,
+ "loss": 0.0737,
+ "step": 7100
+ },
+ {
+ "epoch": 15.87,
+ "learning_rate": 1.373690319791783e-05,
+ "loss": 0.067,
+ "step": 7110
+ },
+ {
+ "epoch": 15.89,
+ "learning_rate": 1.3720132661290311e-05,
+ "loss": 0.0694,
+ "step": 7120
+ },
+ {
+ "epoch": 15.92,
+ "learning_rate": 1.3703349972360527e-05,
+ "loss": 0.0655,
+ "step": 7130
+ },
+ {
+ "epoch": 15.94,
+ "learning_rate": 1.3686555185951334e-05,
+ "loss": 0.0683,
+ "step": 7140
+ },
+ {
+ "epoch": 15.96,
+ "learning_rate": 1.3669748356925112e-05,
+ "loss": 0.0771,
+ "step": 7150
+ },
+ {
+ "epoch": 15.98,
+ "learning_rate": 1.3652929540183578e-05,
+ "loss": 0.0552,
+ "step": 7160
+ },
+ {
+ "epoch": 16.0,
+ "learning_rate": 1.3636098790667605e-05,
+ "loss": 0.0605,
+ "step": 7170
+ },
+ {
+ "epoch": 16.03,
+ "learning_rate": 1.3619256163357046e-05,
+ "loss": 0.0511,
+ "step": 7180
+ },
+ {
+ "epoch": 16.05,
+ "learning_rate": 1.3602401713270566e-05,
+ "loss": 0.0545,
+ "step": 7190
+ },
+ {
+ "epoch": 16.07,
+ "learning_rate": 1.3585535495465432e-05,
+ "loss": 0.0435,
+ "step": 7200
+ },
+ {
+ "epoch": 16.09,
+ "learning_rate": 1.3568657565037365e-05,
+ "loss": 0.0541,
+ "step": 7210
+ },
+ {
+ "epoch": 16.12,
+ "learning_rate": 1.3551767977120341e-05,
+ "loss": 0.0466,
+ "step": 7220
+ },
+ {
+ "epoch": 16.14,
+ "learning_rate": 1.353486678688642e-05,
+ "loss": 0.0594,
+ "step": 7230
+ },
+ {
+ "epoch": 16.16,
+ "learning_rate": 1.351795404954556e-05,
+ "loss": 0.0581,
+ "step": 7240
+ },
+ {
+ "epoch": 16.18,
+ "learning_rate": 1.3501029820345446e-05,
+ "loss": 0.0542,
+ "step": 7250
+ },
+ {
+ "epoch": 16.21,
+ "learning_rate": 1.3484094154571286e-05,
+ "loss": 0.0551,
+ "step": 7260
+ },
+ {
+ "epoch": 16.23,
+ "learning_rate": 1.3467147107545668e-05,
+ "loss": 0.0636,
+ "step": 7270
+ },
+ {
+ "epoch": 16.25,
+ "learning_rate": 1.3450188734628344e-05,
+ "loss": 0.0518,
+ "step": 7280
+ },
+ {
+ "epoch": 16.27,
+ "learning_rate": 1.3433219091216069e-05,
+ "loss": 0.059,
+ "step": 7290
+ },
+ {
+ "epoch": 16.29,
+ "learning_rate": 1.3416238232742414e-05,
+ "loss": 0.0601,
+ "step": 7300
+ },
+ {
+ "epoch": 16.32,
+ "learning_rate": 1.3399246214677583e-05,
+ "loss": 0.0518,
+ "step": 7310
+ },
+ {
+ "epoch": 16.34,
+ "learning_rate": 1.338224309252824e-05,
+ "loss": 0.0554,
+ "step": 7320
+ },
+ {
+ "epoch": 16.36,
+ "learning_rate": 1.3365228921837314e-05,
+ "loss": 0.0554,
+ "step": 7330
+ },
+ {
+ "epoch": 16.38,
+ "learning_rate": 1.3348203758183831e-05,
+ "loss": 0.0533,
+ "step": 7340
+ },
+ {
+ "epoch": 16.41,
+ "learning_rate": 1.3331167657182726e-05,
+ "loss": 0.048,
+ "step": 7350
+ },
+ {
+ "epoch": 16.43,
+ "learning_rate": 1.3314120674484663e-05,
+ "loss": 0.0533,
+ "step": 7360
+ },
+ {
+ "epoch": 16.45,
+ "learning_rate": 1.3297062865775851e-05,
+ "loss": 0.0553,
+ "step": 7370
+ },
+ {
+ "epoch": 16.47,
+ "learning_rate": 1.327999428677786e-05,
+ "loss": 0.0521,
+ "step": 7380
+ },
+ {
+ "epoch": 16.5,
+ "learning_rate": 1.3262914993247454e-05,
+ "loss": 0.0573,
+ "step": 7390
+ },
+ {
+ "epoch": 16.52,
+ "learning_rate": 1.324582504097638e-05,
+ "loss": 0.0546,
+ "step": 7400
+ },
+ {
+ "epoch": 16.54,
+ "learning_rate": 1.3228724485791225e-05,
+ "loss": 0.0527,
+ "step": 7410
+ },
+ {
+ "epoch": 16.56,
+ "learning_rate": 1.321161338355319e-05,
+ "loss": 0.0555,
+ "step": 7420
+ },
+ {
+ "epoch": 16.58,
+ "learning_rate": 1.3194491790157947e-05,
+ "loss": 0.0528,
+ "step": 7430
+ },
+ {
+ "epoch": 16.61,
+ "learning_rate": 1.3177359761535427e-05,
+ "loss": 0.0493,
+ "step": 7440
+ },
+ {
+ "epoch": 16.63,
+ "learning_rate": 1.3160217353649652e-05,
+ "loss": 0.0492,
+ "step": 7450
+ },
+ {
+ "epoch": 16.65,
+ "learning_rate": 1.3143064622498551e-05,
+ "loss": 0.0389,
+ "step": 7460
+ },
+ {
+ "epoch": 16.67,
+ "learning_rate": 1.312590162411378e-05,
+ "loss": 0.064,
+ "step": 7470
+ },
+ {
+ "epoch": 16.7,
+ "learning_rate": 1.310872841456052e-05,
+ "loss": 0.0567,
+ "step": 7480
+ },
+ {
+ "epoch": 16.72,
+ "learning_rate": 1.3091545049937322e-05,
+ "loss": 0.0606,
+ "step": 7490
+ },
+ {
+ "epoch": 16.74,
+ "learning_rate": 1.3074351586375906e-05,
+ "loss": 0.0553,
+ "step": 7500
+ },
+ {
+ "epoch": 16.76,
+ "learning_rate": 1.305714808004098e-05,
+ "loss": 0.0594,
+ "step": 7510
+ },
+ {
+ "epoch": 16.79,
+ "learning_rate": 1.3039934587130056e-05,
+ "loss": 0.0504,
+ "step": 7520
+ },
+ {
+ "epoch": 16.81,
+ "learning_rate": 1.3022711163873272e-05,
+ "loss": 0.0567,
+ "step": 7530
+ },
+ {
+ "epoch": 16.83,
+ "learning_rate": 1.3005477866533202e-05,
+ "loss": 0.0516,
+ "step": 7540
+ },
+ {
+ "epoch": 16.85,
+ "learning_rate": 1.2988234751404683e-05,
+ "loss": 0.0469,
+ "step": 7550
+ },
+ {
+ "epoch": 16.88,
+ "learning_rate": 1.2970981874814613e-05,
+ "loss": 0.0529,
+ "step": 7560
+ },
+ {
+ "epoch": 16.9,
+ "learning_rate": 1.2953719293121775e-05,
+ "loss": 0.0524,
+ "step": 7570
+ },
+ {
+ "epoch": 16.92,
+ "learning_rate": 1.2936447062716668e-05,
+ "loss": 0.0564,
+ "step": 7580
+ },
+ {
+ "epoch": 16.94,
+ "learning_rate": 1.2919165240021303e-05,
+ "loss": 0.0623,
+ "step": 7590
+ },
+ {
+ "epoch": 16.96,
+ "learning_rate": 1.2901873881489021e-05,
+ "loss": 0.0711,
+ "step": 7600
+ },
+ {
+ "epoch": 16.99,
+ "learning_rate": 1.288457304360432e-05,
+ "loss": 0.0605,
+ "step": 7610
+ },
+ {
+ "epoch": 17.01,
+ "learning_rate": 1.2867262782882662e-05,
+ "loss": 0.0533,
+ "step": 7620
+ },
+ {
+ "epoch": 17.03,
+ "learning_rate": 1.2849943155870284e-05,
+ "loss": 0.0508,
+ "step": 7630
+ },
+ {
+ "epoch": 17.05,
+ "learning_rate": 1.2832614219144027e-05,
+ "loss": 0.0406,
+ "step": 7640
+ },
+ {
+ "epoch": 17.08,
+ "learning_rate": 1.2815276029311138e-05,
+ "loss": 0.0545,
+ "step": 7650
+ },
+ {
+ "epoch": 17.1,
+ "learning_rate": 1.2797928643009097e-05,
+ "loss": 0.0526,
+ "step": 7660
+ },
+ {
+ "epoch": 17.12,
+ "learning_rate": 1.2780572116905418e-05,
+ "loss": 0.0453,
+ "step": 7670
+ },
+ {
+ "epoch": 17.14,
+ "learning_rate": 1.276320650769748e-05,
+ "loss": 0.0373,
+ "step": 7680
+ },
+ {
+ "epoch": 17.17,
+ "learning_rate": 1.2745831872112318e-05,
+ "loss": 0.0534,
+ "step": 7690
+ },
+ {
+ "epoch": 17.19,
+ "learning_rate": 1.2728448266906468e-05,
+ "loss": 0.0424,
+ "step": 7700
+ },
+ {
+ "epoch": 17.21,
+ "learning_rate": 1.2711055748865765e-05,
+ "loss": 0.0443,
+ "step": 7710
+ },
+ {
+ "epoch": 17.23,
+ "learning_rate": 1.2693654374805148e-05,
+ "loss": 0.0491,
+ "step": 7720
+ },
+ {
+ "epoch": 17.25,
+ "learning_rate": 1.2676244201568498e-05,
+ "loss": 0.0369,
+ "step": 7730
+ },
+ {
+ "epoch": 17.28,
+ "learning_rate": 1.2658825286028428e-05,
+ "loss": 0.0428,
+ "step": 7740
+ },
+ {
+ "epoch": 17.3,
+ "learning_rate": 1.2641397685086124e-05,
+ "loss": 0.0455,
+ "step": 7750
+ },
+ {
+ "epoch": 17.32,
+ "learning_rate": 1.2623961455671125e-05,
+ "loss": 0.0429,
+ "step": 7760
+ },
+ {
+ "epoch": 17.34,
+ "learning_rate": 1.2606516654741172e-05,
+ "loss": 0.0456,
+ "step": 7770
+ },
+ {
+ "epoch": 17.37,
+ "learning_rate": 1.2589063339281995e-05,
+ "loss": 0.0584,
+ "step": 7780
+ },
+ {
+ "epoch": 17.39,
+ "learning_rate": 1.257160156630715e-05,
+ "loss": 0.0545,
+ "step": 7790
+ },
+ {
+ "epoch": 17.41,
+ "learning_rate": 1.2554131392857812e-05,
+ "loss": 0.073,
+ "step": 7800
+ },
+ {
+ "epoch": 17.43,
+ "learning_rate": 1.253665287600259e-05,
+ "loss": 0.0552,
+ "step": 7810
+ },
+ {
+ "epoch": 17.46,
+ "learning_rate": 1.2519166072837368e-05,
+ "loss": 0.0486,
+ "step": 7820
+ },
+ {
+ "epoch": 17.48,
+ "learning_rate": 1.250167104048508e-05,
+ "loss": 0.0489,
+ "step": 7830
+ },
+ {
+ "epoch": 17.5,
+ "learning_rate": 1.248416783609555e-05,
+ "loss": 0.0629,
+ "step": 7840
+ },
+ {
+ "epoch": 17.52,
+ "learning_rate": 1.2466656516845293e-05,
+ "loss": 0.0472,
+ "step": 7850
+ },
+ {
+ "epoch": 17.54,
+ "learning_rate": 1.244913713993734e-05,
+ "loss": 0.0588,
+ "step": 7860
+ },
+ {
+ "epoch": 17.57,
+ "learning_rate": 1.2431609762601036e-05,
+ "loss": 0.0581,
+ "step": 7870
+ },
+ {
+ "epoch": 17.59,
+ "learning_rate": 1.241407444209186e-05,
+ "loss": 0.0473,
+ "step": 7880
+ },
+ {
+ "epoch": 17.61,
+ "learning_rate": 1.2396531235691245e-05,
+ "loss": 0.0497,
+ "step": 7890
+ },
+ {
+ "epoch": 17.63,
+ "learning_rate": 1.2378980200706376e-05,
+ "loss": 0.0538,
+ "step": 7900
+ },
+ {
+ "epoch": 17.66,
+ "learning_rate": 1.236142139447002e-05,
+ "loss": 0.0606,
+ "step": 7910
+ },
+ {
+ "epoch": 17.68,
+ "learning_rate": 1.2343854874340324e-05,
+ "loss": 0.0633,
+ "step": 7920
+ },
+ {
+ "epoch": 17.7,
+ "learning_rate": 1.2326280697700632e-05,
+ "loss": 0.0562,
+ "step": 7930
+ },
+ {
+ "epoch": 17.72,
+ "learning_rate": 1.2308698921959306e-05,
+ "loss": 0.0545,
+ "step": 7940
+ },
+ {
+ "epoch": 17.75,
+ "learning_rate": 1.2291109604549525e-05,
+ "loss": 0.0502,
+ "step": 7950
+ },
+ {
+ "epoch": 17.77,
+ "learning_rate": 1.2273512802929107e-05,
+ "loss": 0.0442,
+ "step": 7960
+ },
+ {
+ "epoch": 17.79,
+ "learning_rate": 1.2255908574580311e-05,
+ "loss": 0.0564,
+ "step": 7970
+ },
+ {
+ "epoch": 17.81,
+ "learning_rate": 1.2238296977009672e-05,
+ "loss": 0.0526,
+ "step": 7980
+ },
+ {
+ "epoch": 17.83,
+ "learning_rate": 1.2220678067747785e-05,
+ "loss": 0.0612,
+ "step": 7990
+ },
+ {
+ "epoch": 17.86,
+ "learning_rate": 1.2203051904349128e-05,
+ "loss": 0.0516,
+ "step": 8000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 6.984706930455347e+16,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s3/training_args.bin b/s3/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..5c992638db49c0dcf625b34909edd822ec12411c
--- /dev/null
+++ b/s3/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06397119ca7fecb031c64ddfa438d042f564ebe99db5299306b46979a7b05e05
+size 5819
diff --git a/s3/zero_to_fp32.py b/s3/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..c5246ff52274e1d6142001ccf085186d3545ce57
--- /dev/null
+++ b/s3/zero_to_fp32.py
@@ -0,0 +1,578 @@
+#!/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 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_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_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 _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()
+ 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("-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)
diff --git a/s3_en/README.md b/s3_en/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d4576fe074287232d3836bf69c21d3f2593290d9
--- /dev/null
+++ b/s3_en/README.md
@@ -0,0 +1,9 @@
+---
+library_name: peft
+---
+## Training procedure
+
+### Framework versions
+
+
+- PEFT 0.4.0
diff --git a/s3_en/adapter_config.json b/s3_en/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7fd18317d116aed06a782f108f5c2e712c434118
--- /dev/null
+++ b/s3_en/adapter_config.json
@@ -0,0 +1,26 @@
+{
+ "auto_mapping": null,
+ "base_model_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "bias": "none",
+ "fan_in_fan_out": false,
+ "inference_mode": true,
+ "init_lora_weights": true,
+ "layers_pattern": null,
+ "layers_to_transform": null,
+ "lora_alpha": 256,
+ "lora_dropout": 0.05,
+ "modules_to_save": null,
+ "peft_type": "LORA",
+ "r": 128,
+ "revision": null,
+ "target_modules": [
+ "gate_proj",
+ "k_proj",
+ "up_proj",
+ "o_proj",
+ "q_proj",
+ "down_proj",
+ "v_proj"
+ ],
+ "task_type": "CAUSAL_LM"
+}
\ No newline at end of file
diff --git a/s3_en/adapter_model.bin b/s3_en/adapter_model.bin
new file mode 100644
index 0000000000000000000000000000000000000000..93361d395c16d61e15f45658c4527a8844e26dc0
--- /dev/null
+++ b/s3_en/adapter_model.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2493db32d83fb99da283bbab2ac37803d027eaa0e36d3aae890806a5069cb6fc
+size 639787082
diff --git a/s3_en/config.json b/s3_en/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..930a04c04daba4ec27407f60d4588b7440c5983c
--- /dev/null
+++ b/s3_en/config.json
@@ -0,0 +1,44 @@
+{
+ "_name_or_path": "liuhaotian/llava-v1.5-7b",
+ "architectures": [
+ "LlavaLlamaForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "freeze_mm_mlp_adapter": false,
+ "freeze_mm_vision_resampler": false,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "image_aspect_ratio": "pad",
+ "image_grid_pinpoints": null,
+ "initializer_range": 0.02,
+ "intermediate_size": 11008,
+ "max_length": 4096,
+ "max_position_embeddings": 4096,
+ "mm_hidden_size": 1024,
+ "mm_projector_lr": 2e-05,
+ "mm_projector_type": "mlp2x_gelu",
+ "mm_resampler_type": null,
+ "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": "openai/clip-vit-large-patch14-336",
+ "model_type": "llava",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 32,
+ "pad_token_id": 0,
+ "pretraining_tp": 1,
+ "rms_norm_eps": 1e-05,
+ "rope_scaling": null,
+ "tie_word_embeddings": false,
+ "torch_dtype": "float16",
+ "transformers_version": "4.31.0",
+ "tune_mm_mlp_adapter": false,
+ "tune_mm_vision_resampler": false,
+ "unfreeze_mm_vision_tower": false,
+ "use_cache": true,
+ "use_mm_proj": true,
+ "vocab_size": 32000
+}
diff --git a/s3_en/global_step11000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt b/s3_en/global_step11000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..a71ef3ee10bfe4199d252a618f44253d26e37059
--- /dev/null
+++ b/s3_en/global_step11000/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:dfcd06107008f4a07c0e09d6616bd68b06c5601b502b08568b5da1573732a0e9
+size 4089600080
diff --git a/s3_en/global_step11000/mp_rank_00_model_states.pt b/s3_en/global_step11000/mp_rank_00_model_states.pt
new file mode 100644
index 0000000000000000000000000000000000000000..699aec5cb7447e19e2348295db1d2e0df9283b2d
--- /dev/null
+++ b/s3_en/global_step11000/mp_rank_00_model_states.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e0202c974df5a685162c51854103799890526560592e6c092d17f85956f4652
+size 28850200603
diff --git a/s3_en/latest b/s3_en/latest
new file mode 100644
index 0000000000000000000000000000000000000000..2b8686c7ed8bb8587fb5bfce4b266e4264df02e6
--- /dev/null
+++ b/s3_en/latest
@@ -0,0 +1 @@
+global_step11000
\ No newline at end of file
diff --git a/s3_en/non_lora_trainables.bin b/s3_en/non_lora_trainables.bin
new file mode 100644
index 0000000000000000000000000000000000000000..25b2c7773258740632a97791a96b7738f75d9d2c
--- /dev/null
+++ b/s3_en/non_lora_trainables.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94713b553f33c6356082fe8f05550048b4585627113f8501ac8fbfb4085dbe01
+size 41961648
diff --git a/s3_en/rng_state.pth b/s3_en/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..9efd3f73d881b89600c2c7c18864f9adcabe7753
--- /dev/null
+++ b/s3_en/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09c567287e72a15f6f754053da4a17b797d562b9640019264671996875cbcc69
+size 14244
diff --git a/s3_en/special_tokens_map.json b/s3_en/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..14761dcf1466dc232bd41de9c21d4c617b15755e
--- /dev/null
+++ b/s3_en/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+ "bos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "eos_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "pad_token": "",
+ "unk_token": {
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s3_en/tokenizer.model b/s3_en/tokenizer.model
new file mode 100644
index 0000000000000000000000000000000000000000..6c00c742ce03c627d6cd5b795984876fa49fa899
--- /dev/null
+++ b/s3_en/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/s3_en/tokenizer_config.json b/s3_en/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..740756b4bef305e27d0bb4d2e1a40dd8847797f7
--- /dev/null
+++ b/s3_en/tokenizer_config.json
@@ -0,0 +1,35 @@
+{
+ "add_bos_token": true,
+ "add_eos_token": false,
+ "bos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "clean_up_tokenization_spaces": false,
+ "eos_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ },
+ "legacy": false,
+ "model_max_length": 2048,
+ "pad_token": null,
+ "padding_side": "right",
+ "sp_model_kwargs": {},
+ "tokenizer_class": "LlamaTokenizer",
+ "unk_token": {
+ "__type": "AddedToken",
+ "content": "",
+ "lstrip": false,
+ "normalized": false,
+ "rstrip": false,
+ "single_word": false
+ }
+}
diff --git a/s3_en/trainer_state.json b/s3_en/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d11de716f028ae52c76fe6216dd2ccd937cb5b5b
--- /dev/null
+++ b/s3_en/trainer_state.json
@@ -0,0 +1,6616 @@
+{
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 24.553571428571427,
+ "global_step": 11000,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "epoch": 0.02,
+ "learning_rate": 3.7174721189591085e-07,
+ "loss": 5.5375,
+ "step": 10
+ },
+ {
+ "epoch": 0.04,
+ "learning_rate": 7.434944237918217e-07,
+ "loss": 5.7375,
+ "step": 20
+ },
+ {
+ "epoch": 0.07,
+ "learning_rate": 1.1152416356877324e-06,
+ "loss": 5.4938,
+ "step": 30
+ },
+ {
+ "epoch": 0.09,
+ "learning_rate": 1.4869888475836434e-06,
+ "loss": 4.7687,
+ "step": 40
+ },
+ {
+ "epoch": 0.11,
+ "learning_rate": 1.858736059479554e-06,
+ "loss": 4.0375,
+ "step": 50
+ },
+ {
+ "epoch": 0.13,
+ "learning_rate": 2.2304832713754648e-06,
+ "loss": 3.375,
+ "step": 60
+ },
+ {
+ "epoch": 0.16,
+ "learning_rate": 2.6022304832713758e-06,
+ "loss": 3.1344,
+ "step": 70
+ },
+ {
+ "epoch": 0.18,
+ "learning_rate": 2.973977695167287e-06,
+ "loss": 2.6906,
+ "step": 80
+ },
+ {
+ "epoch": 0.2,
+ "learning_rate": 3.3457249070631974e-06,
+ "loss": 2.657,
+ "step": 90
+ },
+ {
+ "epoch": 0.22,
+ "learning_rate": 3.717472118959108e-06,
+ "loss": 3.0039,
+ "step": 100
+ },
+ {
+ "epoch": 0.25,
+ "learning_rate": 4.089219330855019e-06,
+ "loss": 2.5945,
+ "step": 110
+ },
+ {
+ "epoch": 0.27,
+ "learning_rate": 4.4609665427509296e-06,
+ "loss": 1.9672,
+ "step": 120
+ },
+ {
+ "epoch": 0.29,
+ "learning_rate": 4.83271375464684e-06,
+ "loss": 1.5867,
+ "step": 130
+ },
+ {
+ "epoch": 0.31,
+ "learning_rate": 5.2044609665427516e-06,
+ "loss": 1.7336,
+ "step": 140
+ },
+ {
+ "epoch": 0.33,
+ "learning_rate": 5.576208178438662e-06,
+ "loss": 1.6789,
+ "step": 150
+ },
+ {
+ "epoch": 0.36,
+ "learning_rate": 5.947955390334574e-06,
+ "loss": 1.3305,
+ "step": 160
+ },
+ {
+ "epoch": 0.38,
+ "learning_rate": 6.319702602230484e-06,
+ "loss": 1.4691,
+ "step": 170
+ },
+ {
+ "epoch": 0.4,
+ "learning_rate": 6.691449814126395e-06,
+ "loss": 1.3047,
+ "step": 180
+ },
+ {
+ "epoch": 0.42,
+ "learning_rate": 7.063197026022306e-06,
+ "loss": 1.3469,
+ "step": 190
+ },
+ {
+ "epoch": 0.45,
+ "learning_rate": 7.434944237918216e-06,
+ "loss": 1.3875,
+ "step": 200
+ },
+ {
+ "epoch": 0.47,
+ "learning_rate": 7.806691449814127e-06,
+ "loss": 1.2895,
+ "step": 210
+ },
+ {
+ "epoch": 0.49,
+ "learning_rate": 8.178438661710038e-06,
+ "loss": 1.348,
+ "step": 220
+ },
+ {
+ "epoch": 0.51,
+ "learning_rate": 8.550185873605949e-06,
+ "loss": 1.2574,
+ "step": 230
+ },
+ {
+ "epoch": 0.54,
+ "learning_rate": 8.921933085501859e-06,
+ "loss": 1.0988,
+ "step": 240
+ },
+ {
+ "epoch": 0.56,
+ "learning_rate": 9.29368029739777e-06,
+ "loss": 1.2254,
+ "step": 250
+ },
+ {
+ "epoch": 0.58,
+ "learning_rate": 9.66542750929368e-06,
+ "loss": 1.1318,
+ "step": 260
+ },
+ {
+ "epoch": 0.6,
+ "learning_rate": 1.0037174721189591e-05,
+ "loss": 1.232,
+ "step": 270
+ },
+ {
+ "epoch": 0.62,
+ "learning_rate": 1.0408921933085503e-05,
+ "loss": 0.9969,
+ "step": 280
+ },
+ {
+ "epoch": 0.65,
+ "learning_rate": 1.0780669144981412e-05,
+ "loss": 1.634,
+ "step": 290
+ },
+ {
+ "epoch": 0.67,
+ "learning_rate": 1.1152416356877324e-05,
+ "loss": 0.873,
+ "step": 300
+ },
+ {
+ "epoch": 0.69,
+ "learning_rate": 1.1524163568773235e-05,
+ "loss": 0.8949,
+ "step": 310
+ },
+ {
+ "epoch": 0.71,
+ "learning_rate": 1.1895910780669147e-05,
+ "loss": 1.1779,
+ "step": 320
+ },
+ {
+ "epoch": 0.74,
+ "learning_rate": 1.2267657992565056e-05,
+ "loss": 0.9756,
+ "step": 330
+ },
+ {
+ "epoch": 0.76,
+ "learning_rate": 1.2639405204460968e-05,
+ "loss": 0.9342,
+ "step": 340
+ },
+ {
+ "epoch": 0.78,
+ "learning_rate": 1.3011152416356879e-05,
+ "loss": 1.1662,
+ "step": 350
+ },
+ {
+ "epoch": 0.8,
+ "learning_rate": 1.338289962825279e-05,
+ "loss": 0.8674,
+ "step": 360
+ },
+ {
+ "epoch": 0.83,
+ "learning_rate": 1.37546468401487e-05,
+ "loss": 0.8068,
+ "step": 370
+ },
+ {
+ "epoch": 0.85,
+ "learning_rate": 1.4126394052044612e-05,
+ "loss": 0.9621,
+ "step": 380
+ },
+ {
+ "epoch": 0.87,
+ "learning_rate": 1.4498141263940521e-05,
+ "loss": 0.8658,
+ "step": 390
+ },
+ {
+ "epoch": 0.89,
+ "learning_rate": 1.4869888475836432e-05,
+ "loss": 0.9328,
+ "step": 400
+ },
+ {
+ "epoch": 0.92,
+ "learning_rate": 1.5241635687732344e-05,
+ "loss": 0.8383,
+ "step": 410
+ },
+ {
+ "epoch": 0.94,
+ "learning_rate": 1.5613382899628255e-05,
+ "loss": 1.4459,
+ "step": 420
+ },
+ {
+ "epoch": 0.96,
+ "learning_rate": 1.5985130111524165e-05,
+ "loss": 1.0838,
+ "step": 430
+ },
+ {
+ "epoch": 0.98,
+ "learning_rate": 1.6356877323420076e-05,
+ "loss": 1.1383,
+ "step": 440
+ },
+ {
+ "epoch": 1.0,
+ "learning_rate": 1.6728624535315986e-05,
+ "loss": 1.1641,
+ "step": 450
+ },
+ {
+ "epoch": 1.03,
+ "learning_rate": 1.7100371747211897e-05,
+ "loss": 1.1611,
+ "step": 460
+ },
+ {
+ "epoch": 1.05,
+ "learning_rate": 1.7472118959107808e-05,
+ "loss": 0.8883,
+ "step": 470
+ },
+ {
+ "epoch": 1.07,
+ "learning_rate": 1.7843866171003718e-05,
+ "loss": 0.9504,
+ "step": 480
+ },
+ {
+ "epoch": 1.09,
+ "learning_rate": 1.8215613382899632e-05,
+ "loss": 1.0572,
+ "step": 490
+ },
+ {
+ "epoch": 1.12,
+ "learning_rate": 1.858736059479554e-05,
+ "loss": 0.7316,
+ "step": 500
+ },
+ {
+ "epoch": 1.14,
+ "learning_rate": 1.8959107806691453e-05,
+ "loss": 0.8014,
+ "step": 510
+ },
+ {
+ "epoch": 1.16,
+ "learning_rate": 1.933085501858736e-05,
+ "loss": 0.785,
+ "step": 520
+ },
+ {
+ "epoch": 1.18,
+ "learning_rate": 1.970260223048327e-05,
+ "loss": 0.6114,
+ "step": 530
+ },
+ {
+ "epoch": 1.21,
+ "learning_rate": 1.9999999346673545e-05,
+ "loss": 0.7264,
+ "step": 540
+ },
+ {
+ "epoch": 1.23,
+ "learning_rate": 1.9999976480256544e-05,
+ "loss": 1.0746,
+ "step": 550
+ },
+ {
+ "epoch": 1.25,
+ "learning_rate": 1.9999920947602093e-05,
+ "loss": 0.569,
+ "step": 560
+ },
+ {
+ "epoch": 1.27,
+ "learning_rate": 1.9999832748891602e-05,
+ "loss": 1.0086,
+ "step": 570
+ },
+ {
+ "epoch": 1.29,
+ "learning_rate": 1.999971188441319e-05,
+ "loss": 0.9928,
+ "step": 580
+ },
+ {
+ "epoch": 1.32,
+ "learning_rate": 1.999955835456167e-05,
+ "loss": 1.0482,
+ "step": 590
+ },
+ {
+ "epoch": 1.34,
+ "learning_rate": 1.9999372159838563e-05,
+ "loss": 0.995,
+ "step": 600
+ },
+ {
+ "epoch": 1.36,
+ "learning_rate": 1.9999153300852108e-05,
+ "loss": 0.6276,
+ "step": 610
+ },
+ {
+ "epoch": 1.38,
+ "learning_rate": 1.999890177831723e-05,
+ "loss": 0.9965,
+ "step": 620
+ },
+ {
+ "epoch": 1.41,
+ "learning_rate": 1.9998617593055565e-05,
+ "loss": 0.9007,
+ "step": 630
+ },
+ {
+ "epoch": 1.43,
+ "learning_rate": 1.9998300745995437e-05,
+ "loss": 1.099,
+ "step": 640
+ },
+ {
+ "epoch": 1.45,
+ "learning_rate": 1.9997951238171875e-05,
+ "loss": 0.8074,
+ "step": 650
+ },
+ {
+ "epoch": 1.47,
+ "learning_rate": 1.9997569070726585e-05,
+ "loss": 0.7982,
+ "step": 660
+ },
+ {
+ "epoch": 1.5,
+ "learning_rate": 1.9997154244907972e-05,
+ "loss": 0.676,
+ "step": 670
+ },
+ {
+ "epoch": 1.52,
+ "learning_rate": 1.999670676207112e-05,
+ "loss": 0.8164,
+ "step": 680
+ },
+ {
+ "epoch": 1.54,
+ "learning_rate": 1.999622662367778e-05,
+ "loss": 0.6457,
+ "step": 690
+ },
+ {
+ "epoch": 1.56,
+ "learning_rate": 1.99957138312964e-05,
+ "loss": 0.843,
+ "step": 700
+ },
+ {
+ "epoch": 1.58,
+ "learning_rate": 1.999516838660208e-05,
+ "loss": 0.7602,
+ "step": 710
+ },
+ {
+ "epoch": 1.61,
+ "learning_rate": 1.9994590291376585e-05,
+ "loss": 0.7498,
+ "step": 720
+ },
+ {
+ "epoch": 1.63,
+ "learning_rate": 1.999397954750834e-05,
+ "loss": 0.6646,
+ "step": 730
+ },
+ {
+ "epoch": 1.65,
+ "learning_rate": 1.999333615699242e-05,
+ "loss": 0.5847,
+ "step": 740
+ },
+ {
+ "epoch": 1.67,
+ "learning_rate": 1.999266012193054e-05,
+ "loss": 0.8805,
+ "step": 750
+ },
+ {
+ "epoch": 1.7,
+ "learning_rate": 1.9991951444531067e-05,
+ "loss": 0.8666,
+ "step": 760
+ },
+ {
+ "epoch": 1.72,
+ "learning_rate": 1.999121012710898e-05,
+ "loss": 0.8045,
+ "step": 770
+ },
+ {
+ "epoch": 1.74,
+ "learning_rate": 1.9990436172085896e-05,
+ "loss": 0.6297,
+ "step": 780
+ },
+ {
+ "epoch": 1.76,
+ "learning_rate": 1.9989629581990038e-05,
+ "loss": 0.77,
+ "step": 790
+ },
+ {
+ "epoch": 1.79,
+ "learning_rate": 1.9988790359456236e-05,
+ "loss": 0.7264,
+ "step": 800
+ },
+ {
+ "epoch": 1.81,
+ "learning_rate": 1.998791850722593e-05,
+ "loss": 0.6923,
+ "step": 810
+ },
+ {
+ "epoch": 1.83,
+ "learning_rate": 1.9987014028147124e-05,
+ "loss": 0.714,
+ "step": 820
+ },
+ {
+ "epoch": 1.85,
+ "learning_rate": 1.9986076925174433e-05,
+ "loss": 1.0271,
+ "step": 830
+ },
+ {
+ "epoch": 1.88,
+ "learning_rate": 1.9985107201369024e-05,
+ "loss": 0.709,
+ "step": 840
+ },
+ {
+ "epoch": 1.9,
+ "learning_rate": 1.998410485989862e-05,
+ "loss": 0.8906,
+ "step": 850
+ },
+ {
+ "epoch": 1.92,
+ "learning_rate": 1.9983069904037506e-05,
+ "loss": 0.7855,
+ "step": 860
+ },
+ {
+ "epoch": 1.94,
+ "learning_rate": 1.99820023371665e-05,
+ "loss": 0.7781,
+ "step": 870
+ },
+ {
+ "epoch": 1.96,
+ "learning_rate": 1.9980902162772952e-05,
+ "loss": 0.6815,
+ "step": 880
+ },
+ {
+ "epoch": 1.99,
+ "learning_rate": 1.9979769384450728e-05,
+ "loss": 0.7196,
+ "step": 890
+ },
+ {
+ "epoch": 2.01,
+ "learning_rate": 1.9978604005900187e-05,
+ "loss": 0.5428,
+ "step": 900
+ },
+ {
+ "epoch": 2.03,
+ "learning_rate": 1.9977406030928205e-05,
+ "loss": 0.5729,
+ "step": 910
+ },
+ {
+ "epoch": 2.05,
+ "learning_rate": 1.9976175463448113e-05,
+ "loss": 0.4022,
+ "step": 920
+ },
+ {
+ "epoch": 2.08,
+ "learning_rate": 1.997491230747973e-05,
+ "loss": 0.812,
+ "step": 930
+ },
+ {
+ "epoch": 2.1,
+ "learning_rate": 1.9973616567149316e-05,
+ "loss": 0.6283,
+ "step": 940
+ },
+ {
+ "epoch": 2.12,
+ "learning_rate": 1.9972288246689576e-05,
+ "loss": 0.5121,
+ "step": 950
+ },
+ {
+ "epoch": 2.14,
+ "learning_rate": 1.9970927350439647e-05,
+ "loss": 0.7583,
+ "step": 960
+ },
+ {
+ "epoch": 2.17,
+ "learning_rate": 1.9969533882845076e-05,
+ "loss": 0.5348,
+ "step": 970
+ },
+ {
+ "epoch": 2.19,
+ "learning_rate": 1.9968107848457804e-05,
+ "loss": 0.502,
+ "step": 980
+ },
+ {
+ "epoch": 2.21,
+ "learning_rate": 1.9966649251936164e-05,
+ "loss": 0.5887,
+ "step": 990
+ },
+ {
+ "epoch": 2.23,
+ "learning_rate": 1.996515809804485e-05,
+ "loss": 0.6484,
+ "step": 1000
+ },
+ {
+ "epoch": 2.25,
+ "learning_rate": 1.9963634391654912e-05,
+ "loss": 0.5695,
+ "step": 1010
+ },
+ {
+ "epoch": 2.28,
+ "learning_rate": 1.996207813774374e-05,
+ "loss": 0.5599,
+ "step": 1020
+ },
+ {
+ "epoch": 2.3,
+ "learning_rate": 1.9960489341395043e-05,
+ "loss": 0.8953,
+ "step": 1030
+ },
+ {
+ "epoch": 2.32,
+ "learning_rate": 1.9958868007798828e-05,
+ "loss": 0.6059,
+ "step": 1040
+ },
+ {
+ "epoch": 2.34,
+ "learning_rate": 1.9957214142251392e-05,
+ "loss": 0.4453,
+ "step": 1050
+ },
+ {
+ "epoch": 2.37,
+ "learning_rate": 1.9955527750155315e-05,
+ "loss": 0.4588,
+ "step": 1060
+ },
+ {
+ "epoch": 2.39,
+ "learning_rate": 1.995380883701941e-05,
+ "loss": 0.5647,
+ "step": 1070
+ },
+ {
+ "epoch": 2.41,
+ "learning_rate": 1.995205740845874e-05,
+ "loss": 0.4607,
+ "step": 1080
+ },
+ {
+ "epoch": 2.43,
+ "learning_rate": 1.9950273470194566e-05,
+ "loss": 0.404,
+ "step": 1090
+ },
+ {
+ "epoch": 2.46,
+ "learning_rate": 1.9948457028054364e-05,
+ "loss": 0.4993,
+ "step": 1100
+ },
+ {
+ "epoch": 2.48,
+ "learning_rate": 1.994660808797178e-05,
+ "loss": 0.5283,
+ "step": 1110
+ },
+ {
+ "epoch": 2.5,
+ "learning_rate": 1.9944726655986618e-05,
+ "loss": 0.4921,
+ "step": 1120
+ },
+ {
+ "epoch": 2.52,
+ "learning_rate": 1.9942812738244827e-05,
+ "loss": 0.5538,
+ "step": 1130
+ },
+ {
+ "epoch": 2.54,
+ "learning_rate": 1.9940866340998464e-05,
+ "loss": 0.5472,
+ "step": 1140
+ },
+ {
+ "epoch": 2.57,
+ "learning_rate": 1.99388874706057e-05,
+ "loss": 0.6049,
+ "step": 1150
+ },
+ {
+ "epoch": 2.59,
+ "learning_rate": 1.9936876133530766e-05,
+ "loss": 0.5311,
+ "step": 1160
+ },
+ {
+ "epoch": 2.61,
+ "learning_rate": 1.9934832336343968e-05,
+ "loss": 0.5754,
+ "step": 1170
+ },
+ {
+ "epoch": 2.63,
+ "learning_rate": 1.993275608572163e-05,
+ "loss": 0.5119,
+ "step": 1180
+ },
+ {
+ "epoch": 2.66,
+ "learning_rate": 1.99306473884461e-05,
+ "loss": 0.5473,
+ "step": 1190
+ },
+ {
+ "epoch": 2.68,
+ "learning_rate": 1.992850625140572e-05,
+ "loss": 0.5799,
+ "step": 1200
+ },
+ {
+ "epoch": 2.7,
+ "learning_rate": 1.992633268159479e-05,
+ "loss": 0.47,
+ "step": 1210
+ },
+ {
+ "epoch": 2.72,
+ "learning_rate": 1.992412668611356e-05,
+ "loss": 0.4977,
+ "step": 1220
+ },
+ {
+ "epoch": 2.75,
+ "learning_rate": 1.992188827216821e-05,
+ "loss": 0.516,
+ "step": 1230
+ },
+ {
+ "epoch": 2.77,
+ "learning_rate": 1.9919617447070808e-05,
+ "loss": 0.6475,
+ "step": 1240
+ },
+ {
+ "epoch": 2.79,
+ "learning_rate": 1.991731421823931e-05,
+ "loss": 0.4234,
+ "step": 1250
+ },
+ {
+ "epoch": 2.81,
+ "learning_rate": 1.9914978593197507e-05,
+ "loss": 0.7242,
+ "step": 1260
+ },
+ {
+ "epoch": 2.83,
+ "learning_rate": 1.991261057957503e-05,
+ "loss": 0.5497,
+ "step": 1270
+ },
+ {
+ "epoch": 2.86,
+ "learning_rate": 1.9910210185107306e-05,
+ "loss": 0.5153,
+ "step": 1280
+ },
+ {
+ "epoch": 2.88,
+ "learning_rate": 1.9907777417635537e-05,
+ "loss": 0.5354,
+ "step": 1290
+ },
+ {
+ "epoch": 2.9,
+ "learning_rate": 1.9905312285106682e-05,
+ "loss": 0.5673,
+ "step": 1300
+ },
+ {
+ "epoch": 2.92,
+ "learning_rate": 1.9902814795573416e-05,
+ "loss": 0.6213,
+ "step": 1310
+ },
+ {
+ "epoch": 2.95,
+ "learning_rate": 1.9900284957194115e-05,
+ "loss": 0.5334,
+ "step": 1320
+ },
+ {
+ "epoch": 2.97,
+ "learning_rate": 1.9897722778232836e-05,
+ "loss": 0.6574,
+ "step": 1330
+ },
+ {
+ "epoch": 2.99,
+ "learning_rate": 1.9895128267059268e-05,
+ "loss": 0.5353,
+ "step": 1340
+ },
+ {
+ "epoch": 3.01,
+ "learning_rate": 1.9892501432148728e-05,
+ "loss": 0.4071,
+ "step": 1350
+ },
+ {
+ "epoch": 3.04,
+ "learning_rate": 1.988984228208211e-05,
+ "loss": 0.5495,
+ "step": 1360
+ },
+ {
+ "epoch": 3.06,
+ "learning_rate": 1.988715082554588e-05,
+ "loss": 0.4265,
+ "step": 1370
+ },
+ {
+ "epoch": 3.08,
+ "learning_rate": 1.988442707133204e-05,
+ "loss": 0.3707,
+ "step": 1380
+ },
+ {
+ "epoch": 3.1,
+ "learning_rate": 1.988167102833808e-05,
+ "loss": 0.4191,
+ "step": 1390
+ },
+ {
+ "epoch": 3.12,
+ "learning_rate": 1.9878882705566988e-05,
+ "loss": 0.3696,
+ "step": 1400
+ },
+ {
+ "epoch": 3.15,
+ "learning_rate": 1.9876062112127176e-05,
+ "loss": 0.4368,
+ "step": 1410
+ },
+ {
+ "epoch": 3.17,
+ "learning_rate": 1.9873209257232494e-05,
+ "loss": 0.2783,
+ "step": 1420
+ },
+ {
+ "epoch": 3.19,
+ "learning_rate": 1.987032415020216e-05,
+ "loss": 0.3593,
+ "step": 1430
+ },
+ {
+ "epoch": 3.21,
+ "learning_rate": 1.986740680046076e-05,
+ "loss": 0.2914,
+ "step": 1440
+ },
+ {
+ "epoch": 3.24,
+ "learning_rate": 1.9864457217538198e-05,
+ "loss": 0.4211,
+ "step": 1450
+ },
+ {
+ "epoch": 3.26,
+ "learning_rate": 1.986147541106967e-05,
+ "loss": 0.2434,
+ "step": 1460
+ },
+ {
+ "epoch": 3.28,
+ "learning_rate": 1.9858461390795648e-05,
+ "loss": 0.4471,
+ "step": 1470
+ },
+ {
+ "epoch": 3.3,
+ "learning_rate": 1.985541516656182e-05,
+ "loss": 0.3898,
+ "step": 1480
+ },
+ {
+ "epoch": 3.33,
+ "learning_rate": 1.985233674831908e-05,
+ "loss": 0.4364,
+ "step": 1490
+ },
+ {
+ "epoch": 3.35,
+ "learning_rate": 1.984922614612348e-05,
+ "loss": 0.3605,
+ "step": 1500
+ },
+ {
+ "epoch": 3.37,
+ "learning_rate": 1.9846083370136217e-05,
+ "loss": 0.3267,
+ "step": 1510
+ },
+ {
+ "epoch": 3.39,
+ "learning_rate": 1.984290843062358e-05,
+ "loss": 0.4983,
+ "step": 1520
+ },
+ {
+ "epoch": 3.42,
+ "learning_rate": 1.9839701337956922e-05,
+ "loss": 0.351,
+ "step": 1530
+ },
+ {
+ "epoch": 3.44,
+ "learning_rate": 1.9836462102612643e-05,
+ "loss": 0.2861,
+ "step": 1540
+ },
+ {
+ "epoch": 3.46,
+ "learning_rate": 1.9833190735172117e-05,
+ "loss": 0.396,
+ "step": 1550
+ },
+ {
+ "epoch": 3.48,
+ "learning_rate": 1.982988724632171e-05,
+ "loss": 0.4732,
+ "step": 1560
+ },
+ {
+ "epoch": 3.5,
+ "learning_rate": 1.9826551646852696e-05,
+ "loss": 0.3606,
+ "step": 1570
+ },
+ {
+ "epoch": 3.53,
+ "learning_rate": 1.982318394766124e-05,
+ "loss": 0.4976,
+ "step": 1580
+ },
+ {
+ "epoch": 3.55,
+ "learning_rate": 1.9819784159748394e-05,
+ "loss": 0.3811,
+ "step": 1590
+ },
+ {
+ "epoch": 3.57,
+ "learning_rate": 1.9816352294219995e-05,
+ "loss": 0.296,
+ "step": 1600
+ },
+ {
+ "epoch": 3.59,
+ "learning_rate": 1.981288836228669e-05,
+ "loss": 0.4615,
+ "step": 1610
+ },
+ {
+ "epoch": 3.62,
+ "learning_rate": 1.9809392375263865e-05,
+ "loss": 0.2965,
+ "step": 1620
+ },
+ {
+ "epoch": 3.64,
+ "learning_rate": 1.9805864344571625e-05,
+ "loss": 0.346,
+ "step": 1630
+ },
+ {
+ "epoch": 3.66,
+ "learning_rate": 1.980230428173474e-05,
+ "loss": 0.4113,
+ "step": 1640
+ },
+ {
+ "epoch": 3.68,
+ "learning_rate": 1.979871219838263e-05,
+ "loss": 0.3872,
+ "step": 1650
+ },
+ {
+ "epoch": 3.71,
+ "learning_rate": 1.9795088106249306e-05,
+ "loss": 0.3399,
+ "step": 1660
+ },
+ {
+ "epoch": 3.73,
+ "learning_rate": 1.9791432017173338e-05,
+ "loss": 0.3639,
+ "step": 1670
+ },
+ {
+ "epoch": 3.75,
+ "learning_rate": 1.978774394309782e-05,
+ "loss": 0.3155,
+ "step": 1680
+ },
+ {
+ "epoch": 3.77,
+ "learning_rate": 1.9784023896070336e-05,
+ "loss": 0.3681,
+ "step": 1690
+ },
+ {
+ "epoch": 3.79,
+ "learning_rate": 1.9780271888242904e-05,
+ "loss": 0.2871,
+ "step": 1700
+ },
+ {
+ "epoch": 3.82,
+ "learning_rate": 1.9776487931871958e-05,
+ "loss": 0.3521,
+ "step": 1710
+ },
+ {
+ "epoch": 3.84,
+ "learning_rate": 1.9772672039318278e-05,
+ "loss": 0.3396,
+ "step": 1720
+ },
+ {
+ "epoch": 3.86,
+ "learning_rate": 1.976882422304699e-05,
+ "loss": 0.5764,
+ "step": 1730
+ },
+ {
+ "epoch": 3.88,
+ "learning_rate": 1.976494449562748e-05,
+ "loss": 0.2398,
+ "step": 1740
+ },
+ {
+ "epoch": 3.91,
+ "learning_rate": 1.9761032869733397e-05,
+ "loss": 0.4567,
+ "step": 1750
+ },
+ {
+ "epoch": 3.93,
+ "learning_rate": 1.9757089358142573e-05,
+ "loss": 0.4277,
+ "step": 1760
+ },
+ {
+ "epoch": 3.95,
+ "learning_rate": 1.9753113973737016e-05,
+ "loss": 0.5593,
+ "step": 1770
+ },
+ {
+ "epoch": 3.97,
+ "learning_rate": 1.9749106729502833e-05,
+ "loss": 0.5047,
+ "step": 1780
+ },
+ {
+ "epoch": 4.0,
+ "learning_rate": 1.974506763853022e-05,
+ "loss": 0.3204,
+ "step": 1790
+ },
+ {
+ "epoch": 4.02,
+ "learning_rate": 1.9740996714013394e-05,
+ "loss": 0.2983,
+ "step": 1800
+ },
+ {
+ "epoch": 4.04,
+ "learning_rate": 1.9736893969250567e-05,
+ "loss": 0.2599,
+ "step": 1810
+ },
+ {
+ "epoch": 4.06,
+ "learning_rate": 1.97327594176439e-05,
+ "loss": 0.2662,
+ "step": 1820
+ },
+ {
+ "epoch": 4.08,
+ "learning_rate": 1.972859307269944e-05,
+ "loss": 0.195,
+ "step": 1830
+ },
+ {
+ "epoch": 4.11,
+ "learning_rate": 1.9724394948027102e-05,
+ "loss": 0.2209,
+ "step": 1840
+ },
+ {
+ "epoch": 4.13,
+ "learning_rate": 1.9720165057340616e-05,
+ "loss": 0.2792,
+ "step": 1850
+ },
+ {
+ "epoch": 4.15,
+ "learning_rate": 1.971590341445747e-05,
+ "loss": 0.3092,
+ "step": 1860
+ },
+ {
+ "epoch": 4.17,
+ "learning_rate": 1.9711610033298888e-05,
+ "loss": 0.2411,
+ "step": 1870
+ },
+ {
+ "epoch": 4.2,
+ "learning_rate": 1.9707284927889757e-05,
+ "loss": 0.2349,
+ "step": 1880
+ },
+ {
+ "epoch": 4.22,
+ "learning_rate": 1.970292811235861e-05,
+ "loss": 0.2136,
+ "step": 1890
+ },
+ {
+ "epoch": 4.24,
+ "learning_rate": 1.969853960093755e-05,
+ "loss": 0.3193,
+ "step": 1900
+ },
+ {
+ "epoch": 4.26,
+ "learning_rate": 1.969411940796223e-05,
+ "loss": 0.2336,
+ "step": 1910
+ },
+ {
+ "epoch": 4.29,
+ "learning_rate": 1.9689667547871788e-05,
+ "loss": 0.2454,
+ "step": 1920
+ },
+ {
+ "epoch": 4.31,
+ "learning_rate": 1.9685184035208814e-05,
+ "loss": 0.3064,
+ "step": 1930
+ },
+ {
+ "epoch": 4.33,
+ "learning_rate": 1.968066888461929e-05,
+ "loss": 0.2555,
+ "step": 1940
+ },
+ {
+ "epoch": 4.35,
+ "learning_rate": 1.9676122110852547e-05,
+ "loss": 0.341,
+ "step": 1950
+ },
+ {
+ "epoch": 4.38,
+ "learning_rate": 1.9671543728761226e-05,
+ "loss": 0.2628,
+ "step": 1960
+ },
+ {
+ "epoch": 4.4,
+ "learning_rate": 1.9666933753301203e-05,
+ "loss": 0.2474,
+ "step": 1970
+ },
+ {
+ "epoch": 4.42,
+ "learning_rate": 1.9662292199531575e-05,
+ "loss": 0.1997,
+ "step": 1980
+ },
+ {
+ "epoch": 4.44,
+ "learning_rate": 1.9657619082614588e-05,
+ "loss": 0.3044,
+ "step": 1990
+ },
+ {
+ "epoch": 4.46,
+ "learning_rate": 1.965291441781559e-05,
+ "loss": 0.2905,
+ "step": 2000
+ },
+ {
+ "epoch": 4.49,
+ "learning_rate": 1.964817822050299e-05,
+ "loss": 0.2705,
+ "step": 2010
+ },
+ {
+ "epoch": 4.51,
+ "learning_rate": 1.9643410506148196e-05,
+ "loss": 0.2476,
+ "step": 2020
+ },
+ {
+ "epoch": 4.53,
+ "learning_rate": 1.9638611290325576e-05,
+ "loss": 0.279,
+ "step": 2030
+ },
+ {
+ "epoch": 4.55,
+ "learning_rate": 1.96337805887124e-05,
+ "loss": 0.294,
+ "step": 2040
+ },
+ {
+ "epoch": 4.58,
+ "learning_rate": 1.9628918417088785e-05,
+ "loss": 0.2771,
+ "step": 2050
+ },
+ {
+ "epoch": 4.6,
+ "learning_rate": 1.9624024791337657e-05,
+ "loss": 0.2671,
+ "step": 2060
+ },
+ {
+ "epoch": 4.62,
+ "learning_rate": 1.9619099727444686e-05,
+ "loss": 0.2352,
+ "step": 2070
+ },
+ {
+ "epoch": 4.64,
+ "learning_rate": 1.961414324149824e-05,
+ "loss": 0.2896,
+ "step": 2080
+ },
+ {
+ "epoch": 4.67,
+ "learning_rate": 1.9609155349689338e-05,
+ "loss": 0.2349,
+ "step": 2090
+ },
+ {
+ "epoch": 4.69,
+ "learning_rate": 1.9604136068311577e-05,
+ "loss": 0.32,
+ "step": 2100
+ },
+ {
+ "epoch": 4.71,
+ "learning_rate": 1.95990854137611e-05,
+ "loss": 0.2177,
+ "step": 2110
+ },
+ {
+ "epoch": 4.73,
+ "learning_rate": 1.959400340253654e-05,
+ "loss": 0.2956,
+ "step": 2120
+ },
+ {
+ "epoch": 4.75,
+ "learning_rate": 1.9588890051238945e-05,
+ "loss": 0.3919,
+ "step": 2130
+ },
+ {
+ "epoch": 4.78,
+ "learning_rate": 1.9583745376571756e-05,
+ "loss": 0.277,
+ "step": 2140
+ },
+ {
+ "epoch": 4.8,
+ "learning_rate": 1.9578569395340727e-05,
+ "loss": 0.2673,
+ "step": 2150
+ },
+ {
+ "epoch": 4.82,
+ "learning_rate": 1.9573362124453884e-05,
+ "loss": 0.2823,
+ "step": 2160
+ },
+ {
+ "epoch": 4.84,
+ "learning_rate": 1.9568123580921453e-05,
+ "loss": 0.397,
+ "step": 2170
+ },
+ {
+ "epoch": 4.87,
+ "learning_rate": 1.9562853781855836e-05,
+ "loss": 0.2997,
+ "step": 2180
+ },
+ {
+ "epoch": 4.89,
+ "learning_rate": 1.9557552744471517e-05,
+ "loss": 0.2876,
+ "step": 2190
+ },
+ {
+ "epoch": 4.91,
+ "learning_rate": 1.9552220486085033e-05,
+ "loss": 0.2976,
+ "step": 2200
+ },
+ {
+ "epoch": 4.93,
+ "learning_rate": 1.954685702411491e-05,
+ "loss": 0.2552,
+ "step": 2210
+ },
+ {
+ "epoch": 4.96,
+ "learning_rate": 1.9541462376081594e-05,
+ "loss": 0.3736,
+ "step": 2220
+ },
+ {
+ "epoch": 4.98,
+ "learning_rate": 1.953603655960742e-05,
+ "loss": 0.3086,
+ "step": 2230
+ },
+ {
+ "epoch": 5.0,
+ "learning_rate": 1.9530579592416522e-05,
+ "loss": 0.2744,
+ "step": 2240
+ },
+ {
+ "epoch": 5.02,
+ "learning_rate": 1.9525091492334813e-05,
+ "loss": 0.1888,
+ "step": 2250
+ },
+ {
+ "epoch": 5.04,
+ "learning_rate": 1.951957227728988e-05,
+ "loss": 0.22,
+ "step": 2260
+ },
+ {
+ "epoch": 5.07,
+ "learning_rate": 1.9514021965310972e-05,
+ "loss": 0.1589,
+ "step": 2270
+ },
+ {
+ "epoch": 5.09,
+ "learning_rate": 1.950844057452891e-05,
+ "loss": 0.2024,
+ "step": 2280
+ },
+ {
+ "epoch": 5.11,
+ "learning_rate": 1.9502828123176042e-05,
+ "loss": 0.1429,
+ "step": 2290
+ },
+ {
+ "epoch": 5.13,
+ "learning_rate": 1.9497184629586176e-05,
+ "loss": 0.2006,
+ "step": 2300
+ },
+ {
+ "epoch": 5.16,
+ "learning_rate": 1.949151011219453e-05,
+ "loss": 0.2362,
+ "step": 2310
+ },
+ {
+ "epoch": 5.18,
+ "learning_rate": 1.9485804589537655e-05,
+ "loss": 0.2062,
+ "step": 2320
+ },
+ {
+ "epoch": 5.2,
+ "learning_rate": 1.9480068080253393e-05,
+ "loss": 0.2053,
+ "step": 2330
+ },
+ {
+ "epoch": 5.22,
+ "learning_rate": 1.9474300603080805e-05,
+ "loss": 0.2391,
+ "step": 2340
+ },
+ {
+ "epoch": 5.25,
+ "learning_rate": 1.9468502176860117e-05,
+ "loss": 0.2064,
+ "step": 2350
+ },
+ {
+ "epoch": 5.27,
+ "learning_rate": 1.9462672820532643e-05,
+ "loss": 0.1944,
+ "step": 2360
+ },
+ {
+ "epoch": 5.29,
+ "learning_rate": 1.9456812553140744e-05,
+ "loss": 0.1962,
+ "step": 2370
+ },
+ {
+ "epoch": 5.31,
+ "learning_rate": 1.945092139382776e-05,
+ "loss": 0.2846,
+ "step": 2380
+ },
+ {
+ "epoch": 5.33,
+ "learning_rate": 1.944499936183793e-05,
+ "loss": 0.2058,
+ "step": 2390
+ },
+ {
+ "epoch": 5.36,
+ "learning_rate": 1.9439046476516356e-05,
+ "loss": 0.1816,
+ "step": 2400
+ },
+ {
+ "epoch": 5.38,
+ "learning_rate": 1.9433062757308914e-05,
+ "loss": 0.1817,
+ "step": 2410
+ },
+ {
+ "epoch": 5.4,
+ "learning_rate": 1.9427048223762212e-05,
+ "loss": 0.2779,
+ "step": 2420
+ },
+ {
+ "epoch": 5.42,
+ "learning_rate": 1.9421002895523515e-05,
+ "loss": 0.2003,
+ "step": 2430
+ },
+ {
+ "epoch": 5.45,
+ "learning_rate": 1.941492679234068e-05,
+ "loss": 0.2632,
+ "step": 2440
+ },
+ {
+ "epoch": 5.47,
+ "learning_rate": 1.9408819934062098e-05,
+ "loss": 0.2191,
+ "step": 2450
+ },
+ {
+ "epoch": 5.49,
+ "learning_rate": 1.9402682340636625e-05,
+ "loss": 0.189,
+ "step": 2460
+ },
+ {
+ "epoch": 5.51,
+ "learning_rate": 1.9396514032113514e-05,
+ "loss": 0.1873,
+ "step": 2470
+ },
+ {
+ "epoch": 5.54,
+ "learning_rate": 1.9390315028642355e-05,
+ "loss": 0.2246,
+ "step": 2480
+ },
+ {
+ "epoch": 5.56,
+ "learning_rate": 1.9384085350473016e-05,
+ "loss": 0.2121,
+ "step": 2490
+ },
+ {
+ "epoch": 5.58,
+ "learning_rate": 1.9377825017955548e-05,
+ "loss": 0.3086,
+ "step": 2500
+ },
+ {
+ "epoch": 5.6,
+ "learning_rate": 1.9371534051540158e-05,
+ "loss": 0.2127,
+ "step": 2510
+ },
+ {
+ "epoch": 5.62,
+ "learning_rate": 1.9365212471777113e-05,
+ "loss": 0.2333,
+ "step": 2520
+ },
+ {
+ "epoch": 5.65,
+ "learning_rate": 1.935886029931668e-05,
+ "loss": 0.2058,
+ "step": 2530
+ },
+ {
+ "epoch": 5.67,
+ "learning_rate": 1.9352477554909067e-05,
+ "loss": 0.2178,
+ "step": 2540
+ },
+ {
+ "epoch": 5.69,
+ "learning_rate": 1.934606425940435e-05,
+ "loss": 0.2761,
+ "step": 2550
+ },
+ {
+ "epoch": 5.71,
+ "learning_rate": 1.93396204337524e-05,
+ "loss": 0.1708,
+ "step": 2560
+ },
+ {
+ "epoch": 5.74,
+ "learning_rate": 1.9333146099002826e-05,
+ "loss": 0.215,
+ "step": 2570
+ },
+ {
+ "epoch": 5.76,
+ "learning_rate": 1.932664127630488e-05,
+ "loss": 0.2466,
+ "step": 2580
+ },
+ {
+ "epoch": 5.78,
+ "learning_rate": 1.9320105986907433e-05,
+ "loss": 0.2224,
+ "step": 2590
+ },
+ {
+ "epoch": 5.8,
+ "learning_rate": 1.931354025215886e-05,
+ "loss": 0.2357,
+ "step": 2600
+ },
+ {
+ "epoch": 5.83,
+ "learning_rate": 1.9306944093507e-05,
+ "loss": 0.208,
+ "step": 2610
+ },
+ {
+ "epoch": 5.85,
+ "learning_rate": 1.930031753249907e-05,
+ "loss": 0.2043,
+ "step": 2620
+ },
+ {
+ "epoch": 5.87,
+ "learning_rate": 1.9293660590781603e-05,
+ "loss": 0.2304,
+ "step": 2630
+ },
+ {
+ "epoch": 5.89,
+ "learning_rate": 1.928697329010037e-05,
+ "loss": 0.1963,
+ "step": 2640
+ },
+ {
+ "epoch": 5.92,
+ "learning_rate": 1.9280255652300326e-05,
+ "loss": 0.1532,
+ "step": 2650
+ },
+ {
+ "epoch": 5.94,
+ "learning_rate": 1.9273507699325513e-05,
+ "loss": 0.2437,
+ "step": 2660
+ },
+ {
+ "epoch": 5.96,
+ "learning_rate": 1.9266729453219008e-05,
+ "loss": 0.2697,
+ "step": 2670
+ },
+ {
+ "epoch": 5.98,
+ "learning_rate": 1.925992093612284e-05,
+ "loss": 0.208,
+ "step": 2680
+ },
+ {
+ "epoch": 6.0,
+ "learning_rate": 1.925308217027792e-05,
+ "loss": 0.1917,
+ "step": 2690
+ },
+ {
+ "epoch": 6.03,
+ "learning_rate": 1.924621317802399e-05,
+ "loss": 0.1847,
+ "step": 2700
+ },
+ {
+ "epoch": 6.05,
+ "learning_rate": 1.9239313981799507e-05,
+ "loss": 0.1475,
+ "step": 2710
+ },
+ {
+ "epoch": 6.07,
+ "learning_rate": 1.92323846041416e-05,
+ "loss": 0.1604,
+ "step": 2720
+ },
+ {
+ "epoch": 6.09,
+ "learning_rate": 1.9225425067685995e-05,
+ "loss": 0.1875,
+ "step": 2730
+ },
+ {
+ "epoch": 6.12,
+ "learning_rate": 1.9218435395166933e-05,
+ "loss": 0.1698,
+ "step": 2740
+ },
+ {
+ "epoch": 6.14,
+ "learning_rate": 1.9211415609417097e-05,
+ "loss": 0.1728,
+ "step": 2750
+ },
+ {
+ "epoch": 6.16,
+ "learning_rate": 1.920436573336754e-05,
+ "loss": 0.1616,
+ "step": 2760
+ },
+ {
+ "epoch": 6.18,
+ "learning_rate": 1.919728579004761e-05,
+ "loss": 0.1575,
+ "step": 2770
+ },
+ {
+ "epoch": 6.21,
+ "learning_rate": 1.919017580258487e-05,
+ "loss": 0.1626,
+ "step": 2780
+ },
+ {
+ "epoch": 6.23,
+ "learning_rate": 1.918303579420503e-05,
+ "loss": 0.2367,
+ "step": 2790
+ },
+ {
+ "epoch": 6.25,
+ "learning_rate": 1.917586578823186e-05,
+ "loss": 0.2239,
+ "step": 2800
+ },
+ {
+ "epoch": 6.27,
+ "learning_rate": 1.916866580808714e-05,
+ "loss": 0.1825,
+ "step": 2810
+ },
+ {
+ "epoch": 6.29,
+ "learning_rate": 1.9161435877290538e-05,
+ "loss": 0.2076,
+ "step": 2820
+ },
+ {
+ "epoch": 6.32,
+ "learning_rate": 1.915417601945958e-05,
+ "loss": 0.2033,
+ "step": 2830
+ },
+ {
+ "epoch": 6.34,
+ "learning_rate": 1.9146886258309548e-05,
+ "loss": 0.1622,
+ "step": 2840
+ },
+ {
+ "epoch": 6.36,
+ "learning_rate": 1.9139566617653395e-05,
+ "loss": 0.1373,
+ "step": 2850
+ },
+ {
+ "epoch": 6.38,
+ "learning_rate": 1.9132217121401698e-05,
+ "loss": 0.1458,
+ "step": 2860
+ },
+ {
+ "epoch": 6.41,
+ "learning_rate": 1.912483779356255e-05,
+ "loss": 0.2153,
+ "step": 2870
+ },
+ {
+ "epoch": 6.43,
+ "learning_rate": 1.9117428658241498e-05,
+ "loss": 0.1351,
+ "step": 2880
+ },
+ {
+ "epoch": 6.45,
+ "learning_rate": 1.9109989739641446e-05,
+ "loss": 0.1979,
+ "step": 2890
+ },
+ {
+ "epoch": 6.47,
+ "learning_rate": 1.9102521062062615e-05,
+ "loss": 0.1764,
+ "step": 2900
+ },
+ {
+ "epoch": 6.5,
+ "learning_rate": 1.909502264990241e-05,
+ "loss": 0.1984,
+ "step": 2910
+ },
+ {
+ "epoch": 6.52,
+ "learning_rate": 1.9087494527655383e-05,
+ "loss": 0.1612,
+ "step": 2920
+ },
+ {
+ "epoch": 6.54,
+ "learning_rate": 1.9079936719913138e-05,
+ "loss": 0.2061,
+ "step": 2930
+ },
+ {
+ "epoch": 6.56,
+ "learning_rate": 1.9072349251364238e-05,
+ "loss": 0.1196,
+ "step": 2940
+ },
+ {
+ "epoch": 6.58,
+ "learning_rate": 1.906473214679416e-05,
+ "loss": 0.1662,
+ "step": 2950
+ },
+ {
+ "epoch": 6.61,
+ "learning_rate": 1.9057085431085163e-05,
+ "loss": 0.2524,
+ "step": 2960
+ },
+ {
+ "epoch": 6.63,
+ "learning_rate": 1.904940912921626e-05,
+ "loss": 0.2487,
+ "step": 2970
+ },
+ {
+ "epoch": 6.65,
+ "learning_rate": 1.9041703266263095e-05,
+ "loss": 0.2368,
+ "step": 2980
+ },
+ {
+ "epoch": 6.67,
+ "learning_rate": 1.9033967867397883e-05,
+ "loss": 0.204,
+ "step": 2990
+ },
+ {
+ "epoch": 6.7,
+ "learning_rate": 1.902620295788932e-05,
+ "loss": 0.177,
+ "step": 3000
+ },
+ {
+ "epoch": 6.72,
+ "learning_rate": 1.9018408563102505e-05,
+ "loss": 0.2044,
+ "step": 3010
+ },
+ {
+ "epoch": 6.74,
+ "learning_rate": 1.901058470849885e-05,
+ "loss": 0.1489,
+ "step": 3020
+ },
+ {
+ "epoch": 6.76,
+ "learning_rate": 1.900273141963601e-05,
+ "loss": 0.1812,
+ "step": 3030
+ },
+ {
+ "epoch": 6.79,
+ "learning_rate": 1.899484872216778e-05,
+ "loss": 0.277,
+ "step": 3040
+ },
+ {
+ "epoch": 6.81,
+ "learning_rate": 1.8986936641844025e-05,
+ "loss": 0.1903,
+ "step": 3050
+ },
+ {
+ "epoch": 6.83,
+ "learning_rate": 1.8978995204510605e-05,
+ "loss": 0.1638,
+ "step": 3060
+ },
+ {
+ "epoch": 6.85,
+ "learning_rate": 1.897102443610926e-05,
+ "loss": 0.2071,
+ "step": 3070
+ },
+ {
+ "epoch": 6.88,
+ "learning_rate": 1.8963024362677557e-05,
+ "loss": 0.1562,
+ "step": 3080
+ },
+ {
+ "epoch": 6.9,
+ "learning_rate": 1.895499501034878e-05,
+ "loss": 0.1878,
+ "step": 3090
+ },
+ {
+ "epoch": 6.92,
+ "learning_rate": 1.8946936405351877e-05,
+ "loss": 0.1863,
+ "step": 3100
+ },
+ {
+ "epoch": 6.94,
+ "learning_rate": 1.893884857401133e-05,
+ "loss": 0.2107,
+ "step": 3110
+ },
+ {
+ "epoch": 6.96,
+ "learning_rate": 1.8930731542747108e-05,
+ "loss": 0.1832,
+ "step": 3120
+ },
+ {
+ "epoch": 6.99,
+ "learning_rate": 1.8922585338074556e-05,
+ "loss": 0.2089,
+ "step": 3130
+ },
+ {
+ "epoch": 7.01,
+ "learning_rate": 1.8914409986604327e-05,
+ "loss": 0.1695,
+ "step": 3140
+ },
+ {
+ "epoch": 7.03,
+ "learning_rate": 1.8906205515042272e-05,
+ "loss": 0.1289,
+ "step": 3150
+ },
+ {
+ "epoch": 7.05,
+ "learning_rate": 1.8897971950189385e-05,
+ "loss": 0.1292,
+ "step": 3160
+ },
+ {
+ "epoch": 7.08,
+ "learning_rate": 1.888970931894169e-05,
+ "loss": 0.122,
+ "step": 3170
+ },
+ {
+ "epoch": 7.1,
+ "learning_rate": 1.888141764829015e-05,
+ "loss": 0.1477,
+ "step": 3180
+ },
+ {
+ "epoch": 7.12,
+ "learning_rate": 1.8873096965320597e-05,
+ "loss": 0.1819,
+ "step": 3190
+ },
+ {
+ "epoch": 7.14,
+ "learning_rate": 1.886474729721364e-05,
+ "loss": 0.1542,
+ "step": 3200
+ },
+ {
+ "epoch": 7.17,
+ "learning_rate": 1.8856368671244565e-05,
+ "loss": 0.1291,
+ "step": 3210
+ },
+ {
+ "epoch": 7.19,
+ "learning_rate": 1.8847961114783254e-05,
+ "loss": 0.1943,
+ "step": 3220
+ },
+ {
+ "epoch": 7.21,
+ "learning_rate": 1.88395246552941e-05,
+ "loss": 0.1725,
+ "step": 3230
+ },
+ {
+ "epoch": 7.23,
+ "learning_rate": 1.8831059320335902e-05,
+ "loss": 0.2418,
+ "step": 3240
+ },
+ {
+ "epoch": 7.25,
+ "learning_rate": 1.882256513756179e-05,
+ "loss": 0.1416,
+ "step": 3250
+ },
+ {
+ "epoch": 7.28,
+ "learning_rate": 1.881404213471913e-05,
+ "loss": 0.1334,
+ "step": 3260
+ },
+ {
+ "epoch": 7.3,
+ "learning_rate": 1.8805490339649428e-05,
+ "loss": 0.1948,
+ "step": 3270
+ },
+ {
+ "epoch": 7.32,
+ "learning_rate": 1.879690978028825e-05,
+ "loss": 0.1445,
+ "step": 3280
+ },
+ {
+ "epoch": 7.34,
+ "learning_rate": 1.8788300484665118e-05,
+ "loss": 0.1854,
+ "step": 3290
+ },
+ {
+ "epoch": 7.37,
+ "learning_rate": 1.877966248090343e-05,
+ "loss": 0.1674,
+ "step": 3300
+ },
+ {
+ "epoch": 7.39,
+ "learning_rate": 1.8770995797220356e-05,
+ "loss": 0.15,
+ "step": 3310
+ },
+ {
+ "epoch": 7.41,
+ "learning_rate": 1.8762300461926766e-05,
+ "loss": 0.1475,
+ "step": 3320
+ },
+ {
+ "epoch": 7.43,
+ "learning_rate": 1.8753576503427107e-05,
+ "loss": 0.1303,
+ "step": 3330
+ },
+ {
+ "epoch": 7.46,
+ "learning_rate": 1.874482395021934e-05,
+ "loss": 0.1653,
+ "step": 3340
+ },
+ {
+ "epoch": 7.48,
+ "learning_rate": 1.8736042830894828e-05,
+ "loss": 0.158,
+ "step": 3350
+ },
+ {
+ "epoch": 7.5,
+ "learning_rate": 1.8727233174138254e-05,
+ "loss": 0.1433,
+ "step": 3360
+ },
+ {
+ "epoch": 7.52,
+ "learning_rate": 1.871839500872752e-05,
+ "loss": 0.1524,
+ "step": 3370
+ },
+ {
+ "epoch": 7.54,
+ "learning_rate": 1.8709528363533653e-05,
+ "loss": 0.1571,
+ "step": 3380
+ },
+ {
+ "epoch": 7.57,
+ "learning_rate": 1.8700633267520715e-05,
+ "loss": 0.1541,
+ "step": 3390
+ },
+ {
+ "epoch": 7.59,
+ "learning_rate": 1.8691709749745705e-05,
+ "loss": 0.1731,
+ "step": 3400
+ },
+ {
+ "epoch": 7.61,
+ "learning_rate": 1.8682757839358472e-05,
+ "loss": 0.138,
+ "step": 3410
+ },
+ {
+ "epoch": 7.63,
+ "learning_rate": 1.86737775656016e-05,
+ "loss": 0.165,
+ "step": 3420
+ },
+ {
+ "epoch": 7.66,
+ "learning_rate": 1.866476895781034e-05,
+ "loss": 0.1463,
+ "step": 3430
+ },
+ {
+ "epoch": 7.68,
+ "learning_rate": 1.8655732045412488e-05,
+ "loss": 0.1407,
+ "step": 3440
+ },
+ {
+ "epoch": 7.7,
+ "learning_rate": 1.8646666857928314e-05,
+ "loss": 0.1489,
+ "step": 3450
+ },
+ {
+ "epoch": 7.72,
+ "learning_rate": 1.8637573424970435e-05,
+ "loss": 0.1773,
+ "step": 3460
+ },
+ {
+ "epoch": 7.75,
+ "learning_rate": 1.862845177624375e-05,
+ "loss": 0.1575,
+ "step": 3470
+ },
+ {
+ "epoch": 7.77,
+ "learning_rate": 1.8619301941545323e-05,
+ "loss": 0.1614,
+ "step": 3480
+ },
+ {
+ "epoch": 7.79,
+ "learning_rate": 1.8610123950764288e-05,
+ "loss": 0.1401,
+ "step": 3490
+ },
+ {
+ "epoch": 7.81,
+ "learning_rate": 1.8600917833881765e-05,
+ "loss": 0.1524,
+ "step": 3500
+ },
+ {
+ "epoch": 7.83,
+ "learning_rate": 1.8591683620970737e-05,
+ "loss": 0.162,
+ "step": 3510
+ },
+ {
+ "epoch": 7.86,
+ "learning_rate": 1.858242134219598e-05,
+ "loss": 0.1616,
+ "step": 3520
+ },
+ {
+ "epoch": 7.88,
+ "learning_rate": 1.8573131027813945e-05,
+ "loss": 0.2043,
+ "step": 3530
+ },
+ {
+ "epoch": 7.9,
+ "learning_rate": 1.856381270817266e-05,
+ "loss": 0.1434,
+ "step": 3540
+ },
+ {
+ "epoch": 7.92,
+ "learning_rate": 1.8554466413711644e-05,
+ "loss": 0.2232,
+ "step": 3550
+ },
+ {
+ "epoch": 7.95,
+ "learning_rate": 1.8545092174961795e-05,
+ "loss": 0.2112,
+ "step": 3560
+ },
+ {
+ "epoch": 7.97,
+ "learning_rate": 1.85356900225453e-05,
+ "loss": 0.1861,
+ "step": 3570
+ },
+ {
+ "epoch": 7.99,
+ "learning_rate": 1.852625998717552e-05,
+ "loss": 0.2109,
+ "step": 3580
+ },
+ {
+ "epoch": 8.01,
+ "learning_rate": 1.8516802099656907e-05,
+ "loss": 0.1285,
+ "step": 3590
+ },
+ {
+ "epoch": 8.04,
+ "learning_rate": 1.8507316390884894e-05,
+ "loss": 0.1166,
+ "step": 3600
+ },
+ {
+ "epoch": 8.06,
+ "learning_rate": 1.84978028918458e-05,
+ "loss": 0.1238,
+ "step": 3610
+ },
+ {
+ "epoch": 8.08,
+ "learning_rate": 1.848826163361671e-05,
+ "loss": 0.0952,
+ "step": 3620
+ },
+ {
+ "epoch": 8.1,
+ "learning_rate": 1.8478692647365402e-05,
+ "loss": 0.1141,
+ "step": 3630
+ },
+ {
+ "epoch": 8.12,
+ "learning_rate": 1.846909596435023e-05,
+ "loss": 0.18,
+ "step": 3640
+ },
+ {
+ "epoch": 8.15,
+ "learning_rate": 1.845947161592002e-05,
+ "loss": 0.1475,
+ "step": 3650
+ },
+ {
+ "epoch": 8.17,
+ "learning_rate": 1.844981963351397e-05,
+ "loss": 0.1424,
+ "step": 3660
+ },
+ {
+ "epoch": 8.19,
+ "learning_rate": 1.8440140048661547e-05,
+ "loss": 0.14,
+ "step": 3670
+ },
+ {
+ "epoch": 8.21,
+ "learning_rate": 1.843043289298239e-05,
+ "loss": 0.1228,
+ "step": 3680
+ },
+ {
+ "epoch": 8.24,
+ "learning_rate": 1.8420698198186197e-05,
+ "loss": 0.1212,
+ "step": 3690
+ },
+ {
+ "epoch": 8.26,
+ "learning_rate": 1.841093599607263e-05,
+ "loss": 0.1263,
+ "step": 3700
+ },
+ {
+ "epoch": 8.28,
+ "learning_rate": 1.8401146318531204e-05,
+ "loss": 0.1271,
+ "step": 3710
+ },
+ {
+ "epoch": 8.3,
+ "learning_rate": 1.8391329197541186e-05,
+ "loss": 0.141,
+ "step": 3720
+ },
+ {
+ "epoch": 8.33,
+ "learning_rate": 1.838148466517149e-05,
+ "loss": 0.1824,
+ "step": 3730
+ },
+ {
+ "epoch": 8.35,
+ "learning_rate": 1.8371612753580583e-05,
+ "loss": 0.1272,
+ "step": 3740
+ },
+ {
+ "epoch": 8.37,
+ "learning_rate": 1.8361713495016354e-05,
+ "loss": 0.1308,
+ "step": 3750
+ },
+ {
+ "epoch": 8.39,
+ "learning_rate": 1.8351786921816037e-05,
+ "loss": 0.1611,
+ "step": 3760
+ },
+ {
+ "epoch": 8.42,
+ "learning_rate": 1.8341833066406083e-05,
+ "loss": 0.1242,
+ "step": 3770
+ },
+ {
+ "epoch": 8.44,
+ "learning_rate": 1.8331851961302075e-05,
+ "loss": 0.1504,
+ "step": 3780
+ },
+ {
+ "epoch": 8.46,
+ "learning_rate": 1.83218436391086e-05,
+ "loss": 0.1294,
+ "step": 3790
+ },
+ {
+ "epoch": 8.48,
+ "learning_rate": 1.8311808132519157e-05,
+ "loss": 0.1181,
+ "step": 3800
+ },
+ {
+ "epoch": 8.5,
+ "learning_rate": 1.830174547431605e-05,
+ "loss": 0.1479,
+ "step": 3810
+ },
+ {
+ "epoch": 8.53,
+ "learning_rate": 1.8291655697370276e-05,
+ "loss": 0.2064,
+ "step": 3820
+ },
+ {
+ "epoch": 8.55,
+ "learning_rate": 1.8281538834641416e-05,
+ "loss": 0.1564,
+ "step": 3830
+ },
+ {
+ "epoch": 8.57,
+ "learning_rate": 1.8271394919177528e-05,
+ "loss": 0.1583,
+ "step": 3840
+ },
+ {
+ "epoch": 8.59,
+ "learning_rate": 1.8261223984115052e-05,
+ "loss": 0.1678,
+ "step": 3850
+ },
+ {
+ "epoch": 8.62,
+ "learning_rate": 1.8251026062678673e-05,
+ "loss": 0.1941,
+ "step": 3860
+ },
+ {
+ "epoch": 8.64,
+ "learning_rate": 1.8240801188181257e-05,
+ "loss": 0.1633,
+ "step": 3870
+ },
+ {
+ "epoch": 8.66,
+ "learning_rate": 1.823054939402369e-05,
+ "loss": 0.1259,
+ "step": 3880
+ },
+ {
+ "epoch": 8.68,
+ "learning_rate": 1.8220270713694803e-05,
+ "loss": 0.1341,
+ "step": 3890
+ },
+ {
+ "epoch": 8.71,
+ "learning_rate": 1.8209965180771262e-05,
+ "loss": 0.1942,
+ "step": 3900
+ },
+ {
+ "epoch": 8.73,
+ "learning_rate": 1.8199632828917445e-05,
+ "loss": 0.1352,
+ "step": 3910
+ },
+ {
+ "epoch": 8.75,
+ "learning_rate": 1.8189273691885336e-05,
+ "loss": 0.15,
+ "step": 3920
+ },
+ {
+ "epoch": 8.77,
+ "learning_rate": 1.8178887803514415e-05,
+ "loss": 0.145,
+ "step": 3930
+ },
+ {
+ "epoch": 8.79,
+ "learning_rate": 1.8168475197731553e-05,
+ "loss": 0.1428,
+ "step": 3940
+ },
+ {
+ "epoch": 8.82,
+ "learning_rate": 1.81580359085509e-05,
+ "loss": 0.1569,
+ "step": 3950
+ },
+ {
+ "epoch": 8.84,
+ "learning_rate": 1.814756997007376e-05,
+ "loss": 0.1141,
+ "step": 3960
+ },
+ {
+ "epoch": 8.86,
+ "learning_rate": 1.8137077416488496e-05,
+ "loss": 0.1747,
+ "step": 3970
+ },
+ {
+ "epoch": 8.88,
+ "learning_rate": 1.8126558282070417e-05,
+ "loss": 0.1588,
+ "step": 3980
+ },
+ {
+ "epoch": 8.91,
+ "learning_rate": 1.8116012601181655e-05,
+ "loss": 0.1262,
+ "step": 3990
+ },
+ {
+ "epoch": 8.93,
+ "learning_rate": 1.810544040827107e-05,
+ "loss": 0.1482,
+ "step": 4000
+ },
+ {
+ "epoch": 8.95,
+ "learning_rate": 1.8094841737874108e-05,
+ "loss": 0.185,
+ "step": 4010
+ },
+ {
+ "epoch": 8.97,
+ "learning_rate": 1.8084216624612726e-05,
+ "loss": 0.1451,
+ "step": 4020
+ },
+ {
+ "epoch": 9.0,
+ "learning_rate": 1.8073565103195254e-05,
+ "loss": 0.1462,
+ "step": 4030
+ },
+ {
+ "epoch": 9.02,
+ "learning_rate": 1.8062887208416282e-05,
+ "loss": 0.1038,
+ "step": 4040
+ },
+ {
+ "epoch": 9.04,
+ "learning_rate": 1.8052182975156557e-05,
+ "loss": 0.1353,
+ "step": 4050
+ },
+ {
+ "epoch": 9.06,
+ "learning_rate": 1.8041452438382873e-05,
+ "loss": 0.1158,
+ "step": 4060
+ },
+ {
+ "epoch": 9.08,
+ "learning_rate": 1.8030695633147926e-05,
+ "loss": 0.1207,
+ "step": 4070
+ },
+ {
+ "epoch": 9.11,
+ "learning_rate": 1.801991259459024e-05,
+ "loss": 0.1248,
+ "step": 4080
+ },
+ {
+ "epoch": 9.13,
+ "learning_rate": 1.8009103357934024e-05,
+ "loss": 0.188,
+ "step": 4090
+ },
+ {
+ "epoch": 9.15,
+ "learning_rate": 1.7998267958489076e-05,
+ "loss": 0.1505,
+ "step": 4100
+ },
+ {
+ "epoch": 9.17,
+ "learning_rate": 1.7987406431650653e-05,
+ "loss": 0.1179,
+ "step": 4110
+ },
+ {
+ "epoch": 9.2,
+ "learning_rate": 1.797651881289935e-05,
+ "loss": 0.0997,
+ "step": 4120
+ },
+ {
+ "epoch": 9.22,
+ "learning_rate": 1.7965605137801015e-05,
+ "loss": 0.1266,
+ "step": 4130
+ },
+ {
+ "epoch": 9.24,
+ "learning_rate": 1.79546654420066e-05,
+ "loss": 0.1295,
+ "step": 4140
+ },
+ {
+ "epoch": 9.26,
+ "learning_rate": 1.7943699761252057e-05,
+ "loss": 0.1423,
+ "step": 4150
+ },
+ {
+ "epoch": 9.29,
+ "learning_rate": 1.7932708131358222e-05,
+ "loss": 0.1616,
+ "step": 4160
+ },
+ {
+ "epoch": 9.31,
+ "learning_rate": 1.7921690588230698e-05,
+ "loss": 0.099,
+ "step": 4170
+ },
+ {
+ "epoch": 9.33,
+ "learning_rate": 1.7910647167859744e-05,
+ "loss": 0.1278,
+ "step": 4180
+ },
+ {
+ "epoch": 9.35,
+ "learning_rate": 1.7899577906320135e-05,
+ "loss": 0.1232,
+ "step": 4190
+ },
+ {
+ "epoch": 9.38,
+ "learning_rate": 1.7888482839771074e-05,
+ "loss": 0.1216,
+ "step": 4200
+ },
+ {
+ "epoch": 9.4,
+ "learning_rate": 1.787736200445606e-05,
+ "loss": 0.1426,
+ "step": 4210
+ },
+ {
+ "epoch": 9.42,
+ "learning_rate": 1.786621543670275e-05,
+ "loss": 0.1266,
+ "step": 4220
+ },
+ {
+ "epoch": 9.44,
+ "learning_rate": 1.7855043172922883e-05,
+ "loss": 0.1054,
+ "step": 4230
+ },
+ {
+ "epoch": 9.46,
+ "learning_rate": 1.7843845249612122e-05,
+ "loss": 0.1219,
+ "step": 4240
+ },
+ {
+ "epoch": 9.49,
+ "learning_rate": 1.7832621703349956e-05,
+ "loss": 0.126,
+ "step": 4250
+ },
+ {
+ "epoch": 9.51,
+ "learning_rate": 1.7821372570799574e-05,
+ "loss": 0.1168,
+ "step": 4260
+ },
+ {
+ "epoch": 9.53,
+ "learning_rate": 1.781009788870775e-05,
+ "loss": 0.1324,
+ "step": 4270
+ },
+ {
+ "epoch": 9.55,
+ "learning_rate": 1.779879769390471e-05,
+ "loss": 0.2456,
+ "step": 4280
+ },
+ {
+ "epoch": 9.58,
+ "learning_rate": 1.7787472023304023e-05,
+ "loss": 0.1771,
+ "step": 4290
+ },
+ {
+ "epoch": 9.6,
+ "learning_rate": 1.7776120913902487e-05,
+ "loss": 0.1553,
+ "step": 4300
+ },
+ {
+ "epoch": 9.62,
+ "learning_rate": 1.7764744402779992e-05,
+ "loss": 0.105,
+ "step": 4310
+ },
+ {
+ "epoch": 9.64,
+ "learning_rate": 1.77533425270994e-05,
+ "loss": 0.1179,
+ "step": 4320
+ },
+ {
+ "epoch": 9.67,
+ "learning_rate": 1.7741915324106445e-05,
+ "loss": 0.1344,
+ "step": 4330
+ },
+ {
+ "epoch": 9.69,
+ "learning_rate": 1.7730462831129584e-05,
+ "loss": 0.1112,
+ "step": 4340
+ },
+ {
+ "epoch": 9.71,
+ "learning_rate": 1.771898508557989e-05,
+ "loss": 0.0973,
+ "step": 4350
+ },
+ {
+ "epoch": 9.73,
+ "learning_rate": 1.7707482124950923e-05,
+ "loss": 0.1479,
+ "step": 4360
+ },
+ {
+ "epoch": 9.75,
+ "learning_rate": 1.7695953986818625e-05,
+ "loss": 0.1479,
+ "step": 4370
+ },
+ {
+ "epoch": 9.78,
+ "learning_rate": 1.7684400708841165e-05,
+ "loss": 0.1963,
+ "step": 4380
+ },
+ {
+ "epoch": 9.8,
+ "learning_rate": 1.7672822328758852e-05,
+ "loss": 0.1501,
+ "step": 4390
+ },
+ {
+ "epoch": 9.82,
+ "learning_rate": 1.7661218884393977e-05,
+ "loss": 0.1355,
+ "step": 4400
+ },
+ {
+ "epoch": 9.84,
+ "learning_rate": 1.764959041365073e-05,
+ "loss": 0.1039,
+ "step": 4410
+ },
+ {
+ "epoch": 9.87,
+ "learning_rate": 1.7637936954515026e-05,
+ "loss": 0.1467,
+ "step": 4420
+ },
+ {
+ "epoch": 9.89,
+ "learning_rate": 1.7626258545054425e-05,
+ "loss": 0.1539,
+ "step": 4430
+ },
+ {
+ "epoch": 9.91,
+ "learning_rate": 1.7614555223417987e-05,
+ "loss": 0.1343,
+ "step": 4440
+ },
+ {
+ "epoch": 9.93,
+ "learning_rate": 1.7602827027836153e-05,
+ "loss": 0.1413,
+ "step": 4450
+ },
+ {
+ "epoch": 9.96,
+ "learning_rate": 1.7591073996620607e-05,
+ "loss": 0.1185,
+ "step": 4460
+ },
+ {
+ "epoch": 9.98,
+ "learning_rate": 1.757929616816418e-05,
+ "loss": 0.1615,
+ "step": 4470
+ },
+ {
+ "epoch": 10.0,
+ "learning_rate": 1.756749358094069e-05,
+ "loss": 0.1807,
+ "step": 4480
+ },
+ {
+ "epoch": 10.02,
+ "learning_rate": 1.755566627350484e-05,
+ "loss": 0.1144,
+ "step": 4490
+ },
+ {
+ "epoch": 10.04,
+ "learning_rate": 1.754381428449209e-05,
+ "loss": 0.1164,
+ "step": 4500
+ },
+ {
+ "epoch": 10.07,
+ "learning_rate": 1.7531937652618515e-05,
+ "loss": 0.1131,
+ "step": 4510
+ },
+ {
+ "epoch": 10.09,
+ "learning_rate": 1.7520036416680687e-05,
+ "loss": 0.1127,
+ "step": 4520
+ },
+ {
+ "epoch": 10.11,
+ "learning_rate": 1.7508110615555573e-05,
+ "loss": 0.1369,
+ "step": 4530
+ },
+ {
+ "epoch": 10.13,
+ "learning_rate": 1.749616028820036e-05,
+ "loss": 0.1321,
+ "step": 4540
+ },
+ {
+ "epoch": 10.16,
+ "learning_rate": 1.748418547365236e-05,
+ "loss": 0.1112,
+ "step": 4550
+ },
+ {
+ "epoch": 10.18,
+ "learning_rate": 1.7472186211028884e-05,
+ "loss": 0.1214,
+ "step": 4560
+ },
+ {
+ "epoch": 10.2,
+ "learning_rate": 1.7460162539527104e-05,
+ "loss": 0.0961,
+ "step": 4570
+ },
+ {
+ "epoch": 10.22,
+ "learning_rate": 1.7448114498423915e-05,
+ "loss": 0.121,
+ "step": 4580
+ },
+ {
+ "epoch": 10.25,
+ "learning_rate": 1.743604212707583e-05,
+ "loss": 0.1242,
+ "step": 4590
+ },
+ {
+ "epoch": 10.27,
+ "learning_rate": 1.7423945464918835e-05,
+ "loss": 0.192,
+ "step": 4600
+ },
+ {
+ "epoch": 10.29,
+ "learning_rate": 1.741182455146827e-05,
+ "loss": 0.1953,
+ "step": 4610
+ },
+ {
+ "epoch": 10.31,
+ "learning_rate": 1.739967942631869e-05,
+ "loss": 0.1081,
+ "step": 4620
+ },
+ {
+ "epoch": 10.33,
+ "learning_rate": 1.738751012914375e-05,
+ "loss": 0.1431,
+ "step": 4630
+ },
+ {
+ "epoch": 10.36,
+ "learning_rate": 1.7375316699696042e-05,
+ "loss": 0.1492,
+ "step": 4640
+ },
+ {
+ "epoch": 10.38,
+ "learning_rate": 1.736309917780702e-05,
+ "loss": 0.1207,
+ "step": 4650
+ },
+ {
+ "epoch": 10.4,
+ "learning_rate": 1.7350857603386816e-05,
+ "loss": 0.181,
+ "step": 4660
+ },
+ {
+ "epoch": 10.42,
+ "learning_rate": 1.733859201642415e-05,
+ "loss": 0.1227,
+ "step": 4670
+ },
+ {
+ "epoch": 10.45,
+ "learning_rate": 1.732630245698617e-05,
+ "loss": 0.1,
+ "step": 4680
+ },
+ {
+ "epoch": 10.47,
+ "learning_rate": 1.7313988965218337e-05,
+ "loss": 0.1262,
+ "step": 4690
+ },
+ {
+ "epoch": 10.49,
+ "learning_rate": 1.730165158134429e-05,
+ "loss": 0.1355,
+ "step": 4700
+ },
+ {
+ "epoch": 10.51,
+ "learning_rate": 1.7289290345665713e-05,
+ "loss": 0.1155,
+ "step": 4710
+ },
+ {
+ "epoch": 10.54,
+ "learning_rate": 1.7276905298562208e-05,
+ "loss": 0.1259,
+ "step": 4720
+ },
+ {
+ "epoch": 10.56,
+ "learning_rate": 1.7264496480491165e-05,
+ "loss": 0.1167,
+ "step": 4730
+ },
+ {
+ "epoch": 10.58,
+ "learning_rate": 1.7252063931987607e-05,
+ "loss": 0.1081,
+ "step": 4740
+ },
+ {
+ "epoch": 10.6,
+ "learning_rate": 1.7239607693664103e-05,
+ "loss": 0.1521,
+ "step": 4750
+ },
+ {
+ "epoch": 10.62,
+ "learning_rate": 1.7227127806210578e-05,
+ "loss": 0.1104,
+ "step": 4760
+ },
+ {
+ "epoch": 10.65,
+ "learning_rate": 1.7214624310394236e-05,
+ "loss": 0.1163,
+ "step": 4770
+ },
+ {
+ "epoch": 10.67,
+ "learning_rate": 1.7202097247059383e-05,
+ "loss": 0.1074,
+ "step": 4780
+ },
+ {
+ "epoch": 10.69,
+ "learning_rate": 1.7189546657127315e-05,
+ "loss": 0.1098,
+ "step": 4790
+ },
+ {
+ "epoch": 10.71,
+ "learning_rate": 1.717697258159619e-05,
+ "loss": 0.1083,
+ "step": 4800
+ },
+ {
+ "epoch": 10.74,
+ "learning_rate": 1.7164375061540877e-05,
+ "loss": 0.1613,
+ "step": 4810
+ },
+ {
+ "epoch": 10.76,
+ "learning_rate": 1.715175413811283e-05,
+ "loss": 0.1204,
+ "step": 4820
+ },
+ {
+ "epoch": 10.78,
+ "learning_rate": 1.7139109852539954e-05,
+ "loss": 0.118,
+ "step": 4830
+ },
+ {
+ "epoch": 10.8,
+ "learning_rate": 1.712644224612647e-05,
+ "loss": 0.1186,
+ "step": 4840
+ },
+ {
+ "epoch": 10.83,
+ "learning_rate": 1.7113751360252777e-05,
+ "loss": 0.1273,
+ "step": 4850
+ },
+ {
+ "epoch": 10.85,
+ "learning_rate": 1.7101037236375324e-05,
+ "loss": 0.1068,
+ "step": 4860
+ },
+ {
+ "epoch": 10.87,
+ "learning_rate": 1.708829991602647e-05,
+ "loss": 0.1393,
+ "step": 4870
+ },
+ {
+ "epoch": 10.89,
+ "learning_rate": 1.707553944081434e-05,
+ "loss": 0.1556,
+ "step": 4880
+ },
+ {
+ "epoch": 10.92,
+ "learning_rate": 1.7062755852422705e-05,
+ "loss": 0.1144,
+ "step": 4890
+ },
+ {
+ "epoch": 10.94,
+ "learning_rate": 1.7049949192610845e-05,
+ "loss": 0.1256,
+ "step": 4900
+ },
+ {
+ "epoch": 10.96,
+ "learning_rate": 1.7037119503213385e-05,
+ "loss": 0.1046,
+ "step": 4910
+ },
+ {
+ "epoch": 10.98,
+ "learning_rate": 1.7024266826140194e-05,
+ "loss": 0.1114,
+ "step": 4920
+ },
+ {
+ "epoch": 11.0,
+ "learning_rate": 1.701139120337624e-05,
+ "loss": 0.1371,
+ "step": 4930
+ },
+ {
+ "epoch": 11.03,
+ "learning_rate": 1.699849267698143e-05,
+ "loss": 0.1544,
+ "step": 4940
+ },
+ {
+ "epoch": 11.05,
+ "learning_rate": 1.698557128909049e-05,
+ "loss": 0.0907,
+ "step": 4950
+ },
+ {
+ "epoch": 11.07,
+ "learning_rate": 1.6972627081912848e-05,
+ "loss": 0.1117,
+ "step": 4960
+ },
+ {
+ "epoch": 11.09,
+ "learning_rate": 1.695966009773244e-05,
+ "loss": 0.1146,
+ "step": 4970
+ },
+ {
+ "epoch": 11.12,
+ "learning_rate": 1.6946670378907635e-05,
+ "loss": 0.1179,
+ "step": 4980
+ },
+ {
+ "epoch": 11.14,
+ "learning_rate": 1.6933657967871056e-05,
+ "loss": 0.1219,
+ "step": 4990
+ },
+ {
+ "epoch": 11.16,
+ "learning_rate": 1.6920622907129452e-05,
+ "loss": 0.1076,
+ "step": 5000
+ },
+ {
+ "epoch": 11.18,
+ "learning_rate": 1.690756523926356e-05,
+ "loss": 0.095,
+ "step": 5010
+ },
+ {
+ "epoch": 11.21,
+ "learning_rate": 1.6894485006927972e-05,
+ "loss": 0.1216,
+ "step": 5020
+ },
+ {
+ "epoch": 11.23,
+ "learning_rate": 1.688138225285098e-05,
+ "loss": 0.1063,
+ "step": 5030
+ },
+ {
+ "epoch": 11.25,
+ "learning_rate": 1.6868257019834464e-05,
+ "loss": 0.0976,
+ "step": 5040
+ },
+ {
+ "epoch": 11.27,
+ "learning_rate": 1.685510935075371e-05,
+ "loss": 0.0978,
+ "step": 5050
+ },
+ {
+ "epoch": 11.29,
+ "learning_rate": 1.684193928855731e-05,
+ "loss": 0.1092,
+ "step": 5060
+ },
+ {
+ "epoch": 11.32,
+ "learning_rate": 1.682874687626701e-05,
+ "loss": 0.2025,
+ "step": 5070
+ },
+ {
+ "epoch": 11.34,
+ "learning_rate": 1.6815532156977553e-05,
+ "loss": 0.1098,
+ "step": 5080
+ },
+ {
+ "epoch": 11.36,
+ "learning_rate": 1.6802295173856558e-05,
+ "loss": 0.1467,
+ "step": 5090
+ },
+ {
+ "epoch": 11.38,
+ "learning_rate": 1.678903597014437e-05,
+ "loss": 0.1702,
+ "step": 5100
+ },
+ {
+ "epoch": 11.41,
+ "learning_rate": 1.6775754589153913e-05,
+ "loss": 0.1874,
+ "step": 5110
+ },
+ {
+ "epoch": 11.43,
+ "learning_rate": 1.676245107427058e-05,
+ "loss": 0.1075,
+ "step": 5120
+ },
+ {
+ "epoch": 11.45,
+ "learning_rate": 1.6749125468952033e-05,
+ "loss": 0.1048,
+ "step": 5130
+ },
+ {
+ "epoch": 11.47,
+ "learning_rate": 1.673577781672812e-05,
+ "loss": 0.1113,
+ "step": 5140
+ },
+ {
+ "epoch": 11.5,
+ "learning_rate": 1.672240816120071e-05,
+ "loss": 0.1181,
+ "step": 5150
+ },
+ {
+ "epoch": 11.52,
+ "learning_rate": 1.670901654604353e-05,
+ "loss": 0.1084,
+ "step": 5160
+ },
+ {
+ "epoch": 11.54,
+ "learning_rate": 1.669560301500205e-05,
+ "loss": 0.1561,
+ "step": 5170
+ },
+ {
+ "epoch": 11.56,
+ "learning_rate": 1.668216761189334e-05,
+ "loss": 0.1396,
+ "step": 5180
+ },
+ {
+ "epoch": 11.58,
+ "learning_rate": 1.6668710380605902e-05,
+ "loss": 0.0881,
+ "step": 5190
+ },
+ {
+ "epoch": 11.61,
+ "learning_rate": 1.6655231365099556e-05,
+ "loss": 0.1349,
+ "step": 5200
+ },
+ {
+ "epoch": 11.63,
+ "learning_rate": 1.6641730609405276e-05,
+ "loss": 0.0992,
+ "step": 5210
+ },
+ {
+ "epoch": 11.65,
+ "learning_rate": 1.6628208157625055e-05,
+ "loss": 0.1394,
+ "step": 5220
+ },
+ {
+ "epoch": 11.67,
+ "learning_rate": 1.6614664053931757e-05,
+ "loss": 0.1257,
+ "step": 5230
+ },
+ {
+ "epoch": 11.7,
+ "learning_rate": 1.6601098342568978e-05,
+ "loss": 0.1163,
+ "step": 5240
+ },
+ {
+ "epoch": 11.72,
+ "learning_rate": 1.65875110678509e-05,
+ "loss": 0.1088,
+ "step": 5250
+ },
+ {
+ "epoch": 11.74,
+ "learning_rate": 1.6573902274162135e-05,
+ "loss": 0.1163,
+ "step": 5260
+ },
+ {
+ "epoch": 11.76,
+ "learning_rate": 1.6560272005957604e-05,
+ "loss": 0.1018,
+ "step": 5270
+ },
+ {
+ "epoch": 11.79,
+ "learning_rate": 1.6546620307762364e-05,
+ "loss": 0.1073,
+ "step": 5280
+ },
+ {
+ "epoch": 11.81,
+ "learning_rate": 1.6532947224171482e-05,
+ "loss": 0.1123,
+ "step": 5290
+ },
+ {
+ "epoch": 11.83,
+ "learning_rate": 1.6519252799849887e-05,
+ "loss": 0.1195,
+ "step": 5300
+ },
+ {
+ "epoch": 11.85,
+ "learning_rate": 1.650553707953221e-05,
+ "loss": 0.1076,
+ "step": 5310
+ },
+ {
+ "epoch": 11.88,
+ "learning_rate": 1.6491800108022657e-05,
+ "loss": 0.1002,
+ "step": 5320
+ },
+ {
+ "epoch": 11.9,
+ "learning_rate": 1.6478041930194848e-05,
+ "loss": 0.0984,
+ "step": 5330
+ },
+ {
+ "epoch": 11.92,
+ "learning_rate": 1.6464262590991683e-05,
+ "loss": 0.1267,
+ "step": 5340
+ },
+ {
+ "epoch": 11.94,
+ "learning_rate": 1.6450462135425187e-05,
+ "loss": 0.1021,
+ "step": 5350
+ },
+ {
+ "epoch": 11.96,
+ "learning_rate": 1.6436640608576354e-05,
+ "loss": 0.1182,
+ "step": 5360
+ },
+ {
+ "epoch": 11.99,
+ "learning_rate": 1.642279805559502e-05,
+ "loss": 0.1624,
+ "step": 5370
+ },
+ {
+ "epoch": 12.01,
+ "learning_rate": 1.6408934521699706e-05,
+ "loss": 0.0877,
+ "step": 5380
+ },
+ {
+ "epoch": 12.03,
+ "learning_rate": 1.639505005217747e-05,
+ "loss": 0.0887,
+ "step": 5390
+ },
+ {
+ "epoch": 12.05,
+ "learning_rate": 1.6381144692383754e-05,
+ "loss": 0.0979,
+ "step": 5400
+ },
+ {
+ "epoch": 12.08,
+ "learning_rate": 1.636721848774224e-05,
+ "loss": 0.1142,
+ "step": 5410
+ },
+ {
+ "epoch": 12.1,
+ "learning_rate": 1.635327148374471e-05,
+ "loss": 0.0897,
+ "step": 5420
+ },
+ {
+ "epoch": 12.12,
+ "learning_rate": 1.633930372595088e-05,
+ "loss": 0.1103,
+ "step": 5430
+ },
+ {
+ "epoch": 12.14,
+ "learning_rate": 1.6325315259988275e-05,
+ "loss": 0.11,
+ "step": 5440
+ },
+ {
+ "epoch": 12.17,
+ "learning_rate": 1.631130613155205e-05,
+ "loss": 0.1137,
+ "step": 5450
+ },
+ {
+ "epoch": 12.19,
+ "learning_rate": 1.6297276386404872e-05,
+ "loss": 0.1219,
+ "step": 5460
+ },
+ {
+ "epoch": 12.21,
+ "learning_rate": 1.628322607037674e-05,
+ "loss": 0.0938,
+ "step": 5470
+ },
+ {
+ "epoch": 12.23,
+ "learning_rate": 1.626915522936486e-05,
+ "loss": 0.1024,
+ "step": 5480
+ },
+ {
+ "epoch": 12.25,
+ "learning_rate": 1.6255063909333486e-05,
+ "loss": 0.1111,
+ "step": 5490
+ },
+ {
+ "epoch": 12.28,
+ "learning_rate": 1.6240952156313762e-05,
+ "loss": 0.0934,
+ "step": 5500
+ },
+ {
+ "epoch": 12.3,
+ "learning_rate": 1.622682001640359e-05,
+ "loss": 0.0959,
+ "step": 5510
+ },
+ {
+ "epoch": 12.32,
+ "learning_rate": 1.6212667535767456e-05,
+ "loss": 0.1176,
+ "step": 5520
+ },
+ {
+ "epoch": 12.34,
+ "learning_rate": 1.6198494760636303e-05,
+ "loss": 0.1401,
+ "step": 5530
+ },
+ {
+ "epoch": 12.37,
+ "learning_rate": 1.618430173730736e-05,
+ "loss": 0.1104,
+ "step": 5540
+ },
+ {
+ "epoch": 12.39,
+ "learning_rate": 1.617008851214401e-05,
+ "loss": 0.1204,
+ "step": 5550
+ },
+ {
+ "epoch": 12.41,
+ "learning_rate": 1.6155855131575614e-05,
+ "loss": 0.0961,
+ "step": 5560
+ },
+ {
+ "epoch": 12.43,
+ "learning_rate": 1.6141601642097382e-05,
+ "loss": 0.1269,
+ "step": 5570
+ },
+ {
+ "epoch": 12.46,
+ "learning_rate": 1.6127328090270213e-05,
+ "loss": 0.0944,
+ "step": 5580
+ },
+ {
+ "epoch": 12.48,
+ "learning_rate": 1.611303452272053e-05,
+ "loss": 0.0853,
+ "step": 5590
+ },
+ {
+ "epoch": 12.5,
+ "learning_rate": 1.609872098614017e-05,
+ "loss": 0.111,
+ "step": 5600
+ },
+ {
+ "epoch": 12.52,
+ "learning_rate": 1.608438752728616e-05,
+ "loss": 0.1084,
+ "step": 5610
+ },
+ {
+ "epoch": 12.54,
+ "learning_rate": 1.6070034192980638e-05,
+ "loss": 0.1068,
+ "step": 5620
+ },
+ {
+ "epoch": 12.57,
+ "learning_rate": 1.6055661030110655e-05,
+ "loss": 0.0844,
+ "step": 5630
+ },
+ {
+ "epoch": 12.59,
+ "learning_rate": 1.6041268085628042e-05,
+ "loss": 0.1077,
+ "step": 5640
+ },
+ {
+ "epoch": 12.61,
+ "learning_rate": 1.602685540654924e-05,
+ "loss": 0.1085,
+ "step": 5650
+ },
+ {
+ "epoch": 12.63,
+ "learning_rate": 1.6012423039955153e-05,
+ "loss": 0.0955,
+ "step": 5660
+ },
+ {
+ "epoch": 12.66,
+ "learning_rate": 1.5997971032991007e-05,
+ "loss": 0.164,
+ "step": 5670
+ },
+ {
+ "epoch": 12.68,
+ "learning_rate": 1.5983499432866187e-05,
+ "loss": 0.1099,
+ "step": 5680
+ },
+ {
+ "epoch": 12.7,
+ "learning_rate": 1.596900828685407e-05,
+ "loss": 0.1086,
+ "step": 5690
+ },
+ {
+ "epoch": 12.72,
+ "learning_rate": 1.5954497642291897e-05,
+ "loss": 0.1172,
+ "step": 5700
+ },
+ {
+ "epoch": 12.75,
+ "learning_rate": 1.593996754658059e-05,
+ "loss": 0.0974,
+ "step": 5710
+ },
+ {
+ "epoch": 12.77,
+ "learning_rate": 1.5925418047184615e-05,
+ "loss": 0.1267,
+ "step": 5720
+ },
+ {
+ "epoch": 12.79,
+ "learning_rate": 1.591084919163183e-05,
+ "loss": 0.0986,
+ "step": 5730
+ },
+ {
+ "epoch": 12.81,
+ "learning_rate": 1.589626102751331e-05,
+ "loss": 0.1285,
+ "step": 5740
+ },
+ {
+ "epoch": 12.83,
+ "learning_rate": 1.588165360248321e-05,
+ "loss": 0.1259,
+ "step": 5750
+ },
+ {
+ "epoch": 12.86,
+ "learning_rate": 1.5867026964258614e-05,
+ "loss": 0.157,
+ "step": 5760
+ },
+ {
+ "epoch": 12.88,
+ "learning_rate": 1.5852381160619343e-05,
+ "loss": 0.2282,
+ "step": 5770
+ },
+ {
+ "epoch": 12.9,
+ "learning_rate": 1.5837716239407855e-05,
+ "loss": 0.1113,
+ "step": 5780
+ },
+ {
+ "epoch": 12.92,
+ "learning_rate": 1.582303224852903e-05,
+ "loss": 0.1462,
+ "step": 5790
+ },
+ {
+ "epoch": 12.95,
+ "learning_rate": 1.580832923595006e-05,
+ "loss": 0.1599,
+ "step": 5800
+ },
+ {
+ "epoch": 12.97,
+ "learning_rate": 1.5793607249700268e-05,
+ "loss": 0.111,
+ "step": 5810
+ },
+ {
+ "epoch": 12.99,
+ "learning_rate": 1.5778866337870952e-05,
+ "loss": 0.1049,
+ "step": 5820
+ },
+ {
+ "epoch": 13.01,
+ "learning_rate": 1.5764106548615244e-05,
+ "loss": 0.1054,
+ "step": 5830
+ },
+ {
+ "epoch": 13.04,
+ "learning_rate": 1.5749327930147932e-05,
+ "loss": 0.1013,
+ "step": 5840
+ },
+ {
+ "epoch": 13.06,
+ "learning_rate": 1.573453053074532e-05,
+ "loss": 0.1004,
+ "step": 5850
+ },
+ {
+ "epoch": 13.08,
+ "learning_rate": 1.571971439874505e-05,
+ "loss": 0.1305,
+ "step": 5860
+ },
+ {
+ "epoch": 13.1,
+ "learning_rate": 1.570487958254597e-05,
+ "loss": 0.0941,
+ "step": 5870
+ },
+ {
+ "epoch": 13.12,
+ "learning_rate": 1.569002613060796e-05,
+ "loss": 0.0892,
+ "step": 5880
+ },
+ {
+ "epoch": 13.15,
+ "learning_rate": 1.5675154091451765e-05,
+ "loss": 0.0935,
+ "step": 5890
+ },
+ {
+ "epoch": 13.17,
+ "learning_rate": 1.566026351365886e-05,
+ "loss": 0.1031,
+ "step": 5900
+ },
+ {
+ "epoch": 13.19,
+ "learning_rate": 1.5645354445871274e-05,
+ "loss": 0.12,
+ "step": 5910
+ },
+ {
+ "epoch": 13.21,
+ "learning_rate": 1.5630426936791433e-05,
+ "loss": 0.126,
+ "step": 5920
+ },
+ {
+ "epoch": 13.24,
+ "learning_rate": 1.5615481035182013e-05,
+ "loss": 0.1178,
+ "step": 5930
+ },
+ {
+ "epoch": 13.26,
+ "learning_rate": 1.5600516789865767e-05,
+ "loss": 0.094,
+ "step": 5940
+ },
+ {
+ "epoch": 13.28,
+ "learning_rate": 1.5585534249725362e-05,
+ "loss": 0.1057,
+ "step": 5950
+ },
+ {
+ "epoch": 13.3,
+ "learning_rate": 1.5570533463703233e-05,
+ "loss": 0.1176,
+ "step": 5960
+ },
+ {
+ "epoch": 13.33,
+ "learning_rate": 1.555551448080143e-05,
+ "loss": 0.1141,
+ "step": 5970
+ },
+ {
+ "epoch": 13.35,
+ "learning_rate": 1.5540477350081423e-05,
+ "loss": 0.1395,
+ "step": 5980
+ },
+ {
+ "epoch": 13.37,
+ "learning_rate": 1.5525422120663986e-05,
+ "loss": 0.0936,
+ "step": 5990
+ },
+ {
+ "epoch": 13.39,
+ "learning_rate": 1.5510348841728997e-05,
+ "loss": 0.079,
+ "step": 6000
+ },
+ {
+ "epoch": 13.42,
+ "learning_rate": 1.5495257562515308e-05,
+ "loss": 0.1107,
+ "step": 6010
+ },
+ {
+ "epoch": 13.44,
+ "learning_rate": 1.5480148332320562e-05,
+ "loss": 0.1274,
+ "step": 6020
+ },
+ {
+ "epoch": 13.46,
+ "learning_rate": 1.5465021200501046e-05,
+ "loss": 0.0954,
+ "step": 6030
+ },
+ {
+ "epoch": 13.48,
+ "learning_rate": 1.5449876216471525e-05,
+ "loss": 0.1024,
+ "step": 6040
+ },
+ {
+ "epoch": 13.5,
+ "learning_rate": 1.5434713429705078e-05,
+ "loss": 0.1373,
+ "step": 6050
+ },
+ {
+ "epoch": 13.53,
+ "learning_rate": 1.5419532889732943e-05,
+ "loss": 0.1223,
+ "step": 6060
+ },
+ {
+ "epoch": 13.55,
+ "learning_rate": 1.540433464614435e-05,
+ "loss": 0.0948,
+ "step": 6070
+ },
+ {
+ "epoch": 13.57,
+ "learning_rate": 1.5389118748586357e-05,
+ "loss": 0.1481,
+ "step": 6080
+ },
+ {
+ "epoch": 13.59,
+ "learning_rate": 1.537388524676369e-05,
+ "loss": 0.1186,
+ "step": 6090
+ },
+ {
+ "epoch": 13.62,
+ "learning_rate": 1.5358634190438592e-05,
+ "loss": 0.0923,
+ "step": 6100
+ },
+ {
+ "epoch": 13.64,
+ "learning_rate": 1.5343365629430638e-05,
+ "loss": 0.0974,
+ "step": 6110
+ },
+ {
+ "epoch": 13.66,
+ "learning_rate": 1.5328079613616592e-05,
+ "loss": 0.0923,
+ "step": 6120
+ },
+ {
+ "epoch": 13.68,
+ "learning_rate": 1.531277619293023e-05,
+ "loss": 0.0964,
+ "step": 6130
+ },
+ {
+ "epoch": 13.71,
+ "learning_rate": 1.5297455417362194e-05,
+ "loss": 0.1066,
+ "step": 6140
+ },
+ {
+ "epoch": 13.73,
+ "learning_rate": 1.52821173369598e-05,
+ "loss": 0.0963,
+ "step": 6150
+ },
+ {
+ "epoch": 13.75,
+ "learning_rate": 1.526676200182691e-05,
+ "loss": 0.1095,
+ "step": 6160
+ },
+ {
+ "epoch": 13.77,
+ "learning_rate": 1.5251389462123748e-05,
+ "loss": 0.0817,
+ "step": 6170
+ },
+ {
+ "epoch": 13.79,
+ "learning_rate": 1.5235999768066729e-05,
+ "loss": 0.1061,
+ "step": 6180
+ },
+ {
+ "epoch": 13.82,
+ "learning_rate": 1.5220592969928313e-05,
+ "loss": 0.108,
+ "step": 6190
+ },
+ {
+ "epoch": 13.84,
+ "learning_rate": 1.5205169118036831e-05,
+ "loss": 0.1021,
+ "step": 6200
+ },
+ {
+ "epoch": 13.86,
+ "learning_rate": 1.5189728262776325e-05,
+ "loss": 0.0855,
+ "step": 6210
+ },
+ {
+ "epoch": 13.88,
+ "learning_rate": 1.5174270454586375e-05,
+ "loss": 0.1021,
+ "step": 6220
+ },
+ {
+ "epoch": 13.91,
+ "learning_rate": 1.5158795743961942e-05,
+ "loss": 0.1141,
+ "step": 6230
+ },
+ {
+ "epoch": 13.93,
+ "learning_rate": 1.5143304181453204e-05,
+ "loss": 0.1101,
+ "step": 6240
+ },
+ {
+ "epoch": 13.95,
+ "learning_rate": 1.5127795817665389e-05,
+ "loss": 0.0898,
+ "step": 6250
+ },
+ {
+ "epoch": 13.97,
+ "learning_rate": 1.5112270703258602e-05,
+ "loss": 0.1091,
+ "step": 6260
+ },
+ {
+ "epoch": 14.0,
+ "learning_rate": 1.5096728888947669e-05,
+ "loss": 0.102,
+ "step": 6270
+ },
+ {
+ "epoch": 14.02,
+ "learning_rate": 1.508117042550197e-05,
+ "loss": 0.0954,
+ "step": 6280
+ },
+ {
+ "epoch": 14.04,
+ "learning_rate": 1.5065595363745272e-05,
+ "loss": 0.0801,
+ "step": 6290
+ },
+ {
+ "epoch": 14.06,
+ "learning_rate": 1.505000375455556e-05,
+ "loss": 0.0796,
+ "step": 6300
+ },
+ {
+ "epoch": 14.08,
+ "learning_rate": 1.503439564886487e-05,
+ "loss": 0.0913,
+ "step": 6310
+ },
+ {
+ "epoch": 14.11,
+ "learning_rate": 1.501877109765914e-05,
+ "loss": 0.1051,
+ "step": 6320
+ },
+ {
+ "epoch": 14.13,
+ "learning_rate": 1.5003130151978012e-05,
+ "loss": 0.0902,
+ "step": 6330
+ },
+ {
+ "epoch": 14.15,
+ "learning_rate": 1.4987472862914697e-05,
+ "loss": 0.1053,
+ "step": 6340
+ },
+ {
+ "epoch": 14.17,
+ "learning_rate": 1.4971799281615782e-05,
+ "loss": 0.097,
+ "step": 6350
+ },
+ {
+ "epoch": 14.2,
+ "learning_rate": 1.4956109459281083e-05,
+ "loss": 0.0678,
+ "step": 6360
+ },
+ {
+ "epoch": 14.22,
+ "learning_rate": 1.4940403447163467e-05,
+ "loss": 0.0964,
+ "step": 6370
+ },
+ {
+ "epoch": 14.24,
+ "learning_rate": 1.4924681296568689e-05,
+ "loss": 0.0963,
+ "step": 6380
+ },
+ {
+ "epoch": 14.26,
+ "learning_rate": 1.4908943058855213e-05,
+ "loss": 0.092,
+ "step": 6390
+ },
+ {
+ "epoch": 14.29,
+ "learning_rate": 1.4893188785434067e-05,
+ "loss": 0.0857,
+ "step": 6400
+ },
+ {
+ "epoch": 14.31,
+ "learning_rate": 1.4877418527768654e-05,
+ "loss": 0.1047,
+ "step": 6410
+ },
+ {
+ "epoch": 14.33,
+ "learning_rate": 1.4861632337374596e-05,
+ "loss": 0.1823,
+ "step": 6420
+ },
+ {
+ "epoch": 14.35,
+ "learning_rate": 1.4845830265819552e-05,
+ "loss": 0.1387,
+ "step": 6430
+ },
+ {
+ "epoch": 14.38,
+ "learning_rate": 1.483001236472307e-05,
+ "loss": 0.1315,
+ "step": 6440
+ },
+ {
+ "epoch": 14.4,
+ "learning_rate": 1.4814178685756405e-05,
+ "loss": 0.0956,
+ "step": 6450
+ },
+ {
+ "epoch": 14.42,
+ "learning_rate": 1.4798329280642345e-05,
+ "loss": 0.0904,
+ "step": 6460
+ },
+ {
+ "epoch": 14.44,
+ "learning_rate": 1.4782464201155057e-05,
+ "loss": 0.0915,
+ "step": 6470
+ },
+ {
+ "epoch": 14.46,
+ "learning_rate": 1.476658349911991e-05,
+ "loss": 0.1157,
+ "step": 6480
+ },
+ {
+ "epoch": 14.49,
+ "learning_rate": 1.4750687226413305e-05,
+ "loss": 0.1634,
+ "step": 6490
+ },
+ {
+ "epoch": 14.51,
+ "learning_rate": 1.4734775434962504e-05,
+ "loss": 0.1473,
+ "step": 6500
+ },
+ {
+ "epoch": 14.53,
+ "learning_rate": 1.471884817674546e-05,
+ "loss": 0.1026,
+ "step": 6510
+ },
+ {
+ "epoch": 14.55,
+ "learning_rate": 1.4702905503790668e-05,
+ "loss": 0.0913,
+ "step": 6520
+ },
+ {
+ "epoch": 14.58,
+ "learning_rate": 1.4686947468176955e-05,
+ "loss": 0.0684,
+ "step": 6530
+ },
+ {
+ "epoch": 14.6,
+ "learning_rate": 1.467097412203334e-05,
+ "loss": 0.1009,
+ "step": 6540
+ },
+ {
+ "epoch": 14.62,
+ "learning_rate": 1.4654985517538864e-05,
+ "loss": 0.14,
+ "step": 6550
+ },
+ {
+ "epoch": 14.64,
+ "learning_rate": 1.4638981706922401e-05,
+ "loss": 0.1611,
+ "step": 6560
+ },
+ {
+ "epoch": 14.67,
+ "learning_rate": 1.4622962742462503e-05,
+ "loss": 0.0993,
+ "step": 6570
+ },
+ {
+ "epoch": 14.69,
+ "learning_rate": 1.4606928676487223e-05,
+ "loss": 0.0919,
+ "step": 6580
+ },
+ {
+ "epoch": 14.71,
+ "learning_rate": 1.459087956137394e-05,
+ "loss": 0.0996,
+ "step": 6590
+ },
+ {
+ "epoch": 14.73,
+ "learning_rate": 1.4574815449549209e-05,
+ "loss": 0.0858,
+ "step": 6600
+ },
+ {
+ "epoch": 14.75,
+ "learning_rate": 1.4558736393488553e-05,
+ "loss": 0.0887,
+ "step": 6610
+ },
+ {
+ "epoch": 14.78,
+ "learning_rate": 1.4542642445716326e-05,
+ "loss": 0.0945,
+ "step": 6620
+ },
+ {
+ "epoch": 14.8,
+ "learning_rate": 1.4526533658805517e-05,
+ "loss": 0.0836,
+ "step": 6630
+ },
+ {
+ "epoch": 14.82,
+ "learning_rate": 1.4510410085377606e-05,
+ "loss": 0.0706,
+ "step": 6640
+ },
+ {
+ "epoch": 14.84,
+ "learning_rate": 1.4494271778102358e-05,
+ "loss": 0.101,
+ "step": 6650
+ },
+ {
+ "epoch": 14.87,
+ "learning_rate": 1.4478118789697675e-05,
+ "loss": 0.0984,
+ "step": 6660
+ },
+ {
+ "epoch": 14.89,
+ "learning_rate": 1.4461951172929419e-05,
+ "loss": 0.0862,
+ "step": 6670
+ },
+ {
+ "epoch": 14.91,
+ "learning_rate": 1.4445768980611233e-05,
+ "loss": 0.0985,
+ "step": 6680
+ },
+ {
+ "epoch": 14.93,
+ "learning_rate": 1.4429572265604375e-05,
+ "loss": 0.1011,
+ "step": 6690
+ },
+ {
+ "epoch": 14.96,
+ "learning_rate": 1.4413361080817545e-05,
+ "loss": 0.0866,
+ "step": 6700
+ },
+ {
+ "epoch": 14.98,
+ "learning_rate": 1.4397135479206705e-05,
+ "loss": 0.0966,
+ "step": 6710
+ },
+ {
+ "epoch": 15.0,
+ "learning_rate": 1.4380895513774922e-05,
+ "loss": 0.0861,
+ "step": 6720
+ },
+ {
+ "epoch": 15.02,
+ "learning_rate": 1.436464123757217e-05,
+ "loss": 0.0794,
+ "step": 6730
+ },
+ {
+ "epoch": 15.04,
+ "learning_rate": 1.4348372703695184e-05,
+ "loss": 0.1678,
+ "step": 6740
+ },
+ {
+ "epoch": 15.07,
+ "learning_rate": 1.4332089965287266e-05,
+ "loss": 0.0916,
+ "step": 6750
+ },
+ {
+ "epoch": 15.09,
+ "learning_rate": 1.431579307553812e-05,
+ "loss": 0.0875,
+ "step": 6760
+ },
+ {
+ "epoch": 15.11,
+ "learning_rate": 1.429948208768368e-05,
+ "loss": 0.0924,
+ "step": 6770
+ },
+ {
+ "epoch": 15.13,
+ "learning_rate": 1.4283157055005928e-05,
+ "loss": 0.0924,
+ "step": 6780
+ },
+ {
+ "epoch": 15.16,
+ "learning_rate": 1.4266818030832732e-05,
+ "loss": 0.111,
+ "step": 6790
+ },
+ {
+ "epoch": 15.18,
+ "learning_rate": 1.4250465068537664e-05,
+ "loss": 0.0983,
+ "step": 6800
+ },
+ {
+ "epoch": 15.2,
+ "learning_rate": 1.4234098221539818e-05,
+ "loss": 0.1122,
+ "step": 6810
+ },
+ {
+ "epoch": 15.22,
+ "learning_rate": 1.4217717543303657e-05,
+ "loss": 0.0798,
+ "step": 6820
+ },
+ {
+ "epoch": 15.25,
+ "learning_rate": 1.4201323087338816e-05,
+ "loss": 0.0831,
+ "step": 6830
+ },
+ {
+ "epoch": 15.27,
+ "learning_rate": 1.4184914907199942e-05,
+ "loss": 0.072,
+ "step": 6840
+ },
+ {
+ "epoch": 15.29,
+ "learning_rate": 1.4168493056486512e-05,
+ "loss": 0.0853,
+ "step": 6850
+ },
+ {
+ "epoch": 15.31,
+ "learning_rate": 1.4152057588842657e-05,
+ "loss": 0.0792,
+ "step": 6860
+ },
+ {
+ "epoch": 15.33,
+ "learning_rate": 1.4135608557956992e-05,
+ "loss": 0.09,
+ "step": 6870
+ },
+ {
+ "epoch": 15.36,
+ "learning_rate": 1.4119146017562441e-05,
+ "loss": 0.1251,
+ "step": 6880
+ },
+ {
+ "epoch": 15.38,
+ "learning_rate": 1.4102670021436059e-05,
+ "loss": 0.0906,
+ "step": 6890
+ },
+ {
+ "epoch": 15.4,
+ "learning_rate": 1.4086180623398842e-05,
+ "loss": 0.0977,
+ "step": 6900
+ },
+ {
+ "epoch": 15.42,
+ "learning_rate": 1.4069677877315587e-05,
+ "loss": 0.0855,
+ "step": 6910
+ },
+ {
+ "epoch": 15.45,
+ "learning_rate": 1.4053161837094675e-05,
+ "loss": 0.0897,
+ "step": 6920
+ },
+ {
+ "epoch": 15.47,
+ "learning_rate": 1.4036632556687927e-05,
+ "loss": 0.0905,
+ "step": 6930
+ },
+ {
+ "epoch": 15.49,
+ "learning_rate": 1.4020090090090408e-05,
+ "loss": 0.0939,
+ "step": 6940
+ },
+ {
+ "epoch": 15.51,
+ "learning_rate": 1.4003534491340259e-05,
+ "loss": 0.1042,
+ "step": 6950
+ },
+ {
+ "epoch": 15.54,
+ "learning_rate": 1.3986965814518521e-05,
+ "loss": 0.087,
+ "step": 6960
+ },
+ {
+ "epoch": 15.56,
+ "learning_rate": 1.3970384113748951e-05,
+ "loss": 0.1534,
+ "step": 6970
+ },
+ {
+ "epoch": 15.58,
+ "learning_rate": 1.3953789443197857e-05,
+ "loss": 0.1028,
+ "step": 6980
+ },
+ {
+ "epoch": 15.6,
+ "learning_rate": 1.3937181857073912e-05,
+ "loss": 0.0845,
+ "step": 6990
+ },
+ {
+ "epoch": 15.62,
+ "learning_rate": 1.3920561409627974e-05,
+ "loss": 0.0913,
+ "step": 7000
+ },
+ {
+ "epoch": 15.65,
+ "learning_rate": 1.3903928155152926e-05,
+ "loss": 0.0892,
+ "step": 7010
+ },
+ {
+ "epoch": 15.67,
+ "learning_rate": 1.3887282147983472e-05,
+ "loss": 0.0888,
+ "step": 7020
+ },
+ {
+ "epoch": 15.69,
+ "learning_rate": 1.3870623442495987e-05,
+ "loss": 0.0869,
+ "step": 7030
+ },
+ {
+ "epoch": 15.71,
+ "learning_rate": 1.3853952093108323e-05,
+ "loss": 0.0917,
+ "step": 7040
+ },
+ {
+ "epoch": 15.74,
+ "learning_rate": 1.3837268154279628e-05,
+ "loss": 0.1263,
+ "step": 7050
+ },
+ {
+ "epoch": 15.76,
+ "learning_rate": 1.3820571680510187e-05,
+ "loss": 0.0976,
+ "step": 7060
+ },
+ {
+ "epoch": 15.78,
+ "learning_rate": 1.3803862726341224e-05,
+ "loss": 0.098,
+ "step": 7070
+ },
+ {
+ "epoch": 15.8,
+ "learning_rate": 1.3787141346354733e-05,
+ "loss": 0.0808,
+ "step": 7080
+ },
+ {
+ "epoch": 15.83,
+ "learning_rate": 1.3770407595173301e-05,
+ "loss": 0.0841,
+ "step": 7090
+ },
+ {
+ "epoch": 15.85,
+ "learning_rate": 1.375366152745992e-05,
+ "loss": 0.0999,
+ "step": 7100
+ },
+ {
+ "epoch": 15.87,
+ "learning_rate": 1.373690319791783e-05,
+ "loss": 0.0809,
+ "step": 7110
+ },
+ {
+ "epoch": 15.89,
+ "learning_rate": 1.3720132661290311e-05,
+ "loss": 0.0904,
+ "step": 7120
+ },
+ {
+ "epoch": 15.92,
+ "learning_rate": 1.3703349972360527e-05,
+ "loss": 0.0811,
+ "step": 7130
+ },
+ {
+ "epoch": 15.94,
+ "learning_rate": 1.3686555185951334e-05,
+ "loss": 0.1308,
+ "step": 7140
+ },
+ {
+ "epoch": 15.96,
+ "learning_rate": 1.3669748356925112e-05,
+ "loss": 0.1161,
+ "step": 7150
+ },
+ {
+ "epoch": 15.98,
+ "learning_rate": 1.3652929540183578e-05,
+ "loss": 0.1,
+ "step": 7160
+ },
+ {
+ "epoch": 16.0,
+ "learning_rate": 1.3636098790667605e-05,
+ "loss": 0.0953,
+ "step": 7170
+ },
+ {
+ "epoch": 16.03,
+ "learning_rate": 1.3619256163357046e-05,
+ "loss": 0.0795,
+ "step": 7180
+ },
+ {
+ "epoch": 16.05,
+ "learning_rate": 1.3602401713270566e-05,
+ "loss": 0.0803,
+ "step": 7190
+ },
+ {
+ "epoch": 16.07,
+ "learning_rate": 1.3585535495465432e-05,
+ "loss": 0.0722,
+ "step": 7200
+ },
+ {
+ "epoch": 16.09,
+ "learning_rate": 1.3568657565037365e-05,
+ "loss": 0.0865,
+ "step": 7210
+ },
+ {
+ "epoch": 16.12,
+ "learning_rate": 1.3551767977120341e-05,
+ "loss": 0.0709,
+ "step": 7220
+ },
+ {
+ "epoch": 16.14,
+ "learning_rate": 1.353486678688642e-05,
+ "loss": 0.1338,
+ "step": 7230
+ },
+ {
+ "epoch": 16.16,
+ "learning_rate": 1.351795404954556e-05,
+ "loss": 0.1025,
+ "step": 7240
+ },
+ {
+ "epoch": 16.18,
+ "learning_rate": 1.3501029820345446e-05,
+ "loss": 0.0765,
+ "step": 7250
+ },
+ {
+ "epoch": 16.21,
+ "learning_rate": 1.3484094154571286e-05,
+ "loss": 0.0908,
+ "step": 7260
+ },
+ {
+ "epoch": 16.23,
+ "learning_rate": 1.3467147107545668e-05,
+ "loss": 0.0688,
+ "step": 7270
+ },
+ {
+ "epoch": 16.25,
+ "learning_rate": 1.3450188734628344e-05,
+ "loss": 0.0963,
+ "step": 7280
+ },
+ {
+ "epoch": 16.27,
+ "learning_rate": 1.3433219091216069e-05,
+ "loss": 0.0725,
+ "step": 7290
+ },
+ {
+ "epoch": 16.29,
+ "learning_rate": 1.3416238232742414e-05,
+ "loss": 0.1163,
+ "step": 7300
+ },
+ {
+ "epoch": 16.32,
+ "learning_rate": 1.3399246214677583e-05,
+ "loss": 0.1239,
+ "step": 7310
+ },
+ {
+ "epoch": 16.34,
+ "learning_rate": 1.338224309252824e-05,
+ "loss": 0.0865,
+ "step": 7320
+ },
+ {
+ "epoch": 16.36,
+ "learning_rate": 1.3365228921837314e-05,
+ "loss": 0.1136,
+ "step": 7330
+ },
+ {
+ "epoch": 16.38,
+ "learning_rate": 1.3348203758183831e-05,
+ "loss": 0.1241,
+ "step": 7340
+ },
+ {
+ "epoch": 16.41,
+ "learning_rate": 1.3331167657182726e-05,
+ "loss": 0.0817,
+ "step": 7350
+ },
+ {
+ "epoch": 16.43,
+ "learning_rate": 1.3314120674484663e-05,
+ "loss": 0.0872,
+ "step": 7360
+ },
+ {
+ "epoch": 16.45,
+ "learning_rate": 1.3297062865775851e-05,
+ "loss": 0.0777,
+ "step": 7370
+ },
+ {
+ "epoch": 16.47,
+ "learning_rate": 1.327999428677786e-05,
+ "loss": 0.1022,
+ "step": 7380
+ },
+ {
+ "epoch": 16.5,
+ "learning_rate": 1.3262914993247454e-05,
+ "loss": 0.1083,
+ "step": 7390
+ },
+ {
+ "epoch": 16.52,
+ "learning_rate": 1.324582504097638e-05,
+ "loss": 0.0749,
+ "step": 7400
+ },
+ {
+ "epoch": 16.54,
+ "learning_rate": 1.3228724485791225e-05,
+ "loss": 0.1021,
+ "step": 7410
+ },
+ {
+ "epoch": 16.56,
+ "learning_rate": 1.321161338355319e-05,
+ "loss": 0.0674,
+ "step": 7420
+ },
+ {
+ "epoch": 16.58,
+ "learning_rate": 1.3194491790157947e-05,
+ "loss": 0.0884,
+ "step": 7430
+ },
+ {
+ "epoch": 16.61,
+ "learning_rate": 1.3177359761535427e-05,
+ "loss": 0.0825,
+ "step": 7440
+ },
+ {
+ "epoch": 16.63,
+ "learning_rate": 1.3160217353649652e-05,
+ "loss": 0.084,
+ "step": 7450
+ },
+ {
+ "epoch": 16.65,
+ "learning_rate": 1.3143064622498551e-05,
+ "loss": 0.0861,
+ "step": 7460
+ },
+ {
+ "epoch": 16.67,
+ "learning_rate": 1.312590162411378e-05,
+ "loss": 0.0937,
+ "step": 7470
+ },
+ {
+ "epoch": 16.7,
+ "learning_rate": 1.310872841456052e-05,
+ "loss": 0.09,
+ "step": 7480
+ },
+ {
+ "epoch": 16.72,
+ "learning_rate": 1.3091545049937322e-05,
+ "loss": 0.088,
+ "step": 7490
+ },
+ {
+ "epoch": 16.74,
+ "learning_rate": 1.3074351586375906e-05,
+ "loss": 0.0989,
+ "step": 7500
+ },
+ {
+ "epoch": 16.76,
+ "learning_rate": 1.305714808004098e-05,
+ "loss": 0.0783,
+ "step": 7510
+ },
+ {
+ "epoch": 16.79,
+ "learning_rate": 1.3039934587130056e-05,
+ "loss": 0.0762,
+ "step": 7520
+ },
+ {
+ "epoch": 16.81,
+ "learning_rate": 1.3022711163873272e-05,
+ "loss": 0.0866,
+ "step": 7530
+ },
+ {
+ "epoch": 16.83,
+ "learning_rate": 1.3005477866533202e-05,
+ "loss": 0.0948,
+ "step": 7540
+ },
+ {
+ "epoch": 16.85,
+ "learning_rate": 1.2988234751404683e-05,
+ "loss": 0.0912,
+ "step": 7550
+ },
+ {
+ "epoch": 16.88,
+ "learning_rate": 1.2970981874814613e-05,
+ "loss": 0.0894,
+ "step": 7560
+ },
+ {
+ "epoch": 16.9,
+ "learning_rate": 1.2953719293121775e-05,
+ "loss": 0.0986,
+ "step": 7570
+ },
+ {
+ "epoch": 16.92,
+ "learning_rate": 1.2936447062716668e-05,
+ "loss": 0.0845,
+ "step": 7580
+ },
+ {
+ "epoch": 16.94,
+ "learning_rate": 1.2919165240021303e-05,
+ "loss": 0.0997,
+ "step": 7590
+ },
+ {
+ "epoch": 16.96,
+ "learning_rate": 1.2901873881489021e-05,
+ "loss": 0.085,
+ "step": 7600
+ },
+ {
+ "epoch": 16.99,
+ "learning_rate": 1.288457304360432e-05,
+ "loss": 0.0903,
+ "step": 7610
+ },
+ {
+ "epoch": 17.01,
+ "learning_rate": 1.2867262782882662e-05,
+ "loss": 0.0711,
+ "step": 7620
+ },
+ {
+ "epoch": 17.03,
+ "learning_rate": 1.2849943155870284e-05,
+ "loss": 0.09,
+ "step": 7630
+ },
+ {
+ "epoch": 17.05,
+ "learning_rate": 1.2832614219144027e-05,
+ "loss": 0.0666,
+ "step": 7640
+ },
+ {
+ "epoch": 17.08,
+ "learning_rate": 1.2815276029311138e-05,
+ "loss": 0.0822,
+ "step": 7650
+ },
+ {
+ "epoch": 17.1,
+ "learning_rate": 1.2797928643009097e-05,
+ "loss": 0.0739,
+ "step": 7660
+ },
+ {
+ "epoch": 17.12,
+ "learning_rate": 1.2780572116905418e-05,
+ "loss": 0.069,
+ "step": 7670
+ },
+ {
+ "epoch": 17.14,
+ "learning_rate": 1.276320650769748e-05,
+ "loss": 0.066,
+ "step": 7680
+ },
+ {
+ "epoch": 17.17,
+ "learning_rate": 1.2745831872112318e-05,
+ "loss": 0.0673,
+ "step": 7690
+ },
+ {
+ "epoch": 17.19,
+ "learning_rate": 1.2728448266906468e-05,
+ "loss": 0.093,
+ "step": 7700
+ },
+ {
+ "epoch": 17.21,
+ "learning_rate": 1.2711055748865765e-05,
+ "loss": 0.0757,
+ "step": 7710
+ },
+ {
+ "epoch": 17.23,
+ "learning_rate": 1.2693654374805148e-05,
+ "loss": 0.0815,
+ "step": 7720
+ },
+ {
+ "epoch": 17.25,
+ "learning_rate": 1.2676244201568498e-05,
+ "loss": 0.0831,
+ "step": 7730
+ },
+ {
+ "epoch": 17.28,
+ "learning_rate": 1.2658825286028428e-05,
+ "loss": 0.0937,
+ "step": 7740
+ },
+ {
+ "epoch": 17.3,
+ "learning_rate": 1.2641397685086124e-05,
+ "loss": 0.0926,
+ "step": 7750
+ },
+ {
+ "epoch": 17.32,
+ "learning_rate": 1.2623961455671125e-05,
+ "loss": 0.0721,
+ "step": 7760
+ },
+ {
+ "epoch": 17.34,
+ "learning_rate": 1.2606516654741172e-05,
+ "loss": 0.1003,
+ "step": 7770
+ },
+ {
+ "epoch": 17.37,
+ "learning_rate": 1.2589063339281995e-05,
+ "loss": 0.0766,
+ "step": 7780
+ },
+ {
+ "epoch": 17.39,
+ "learning_rate": 1.257160156630715e-05,
+ "loss": 0.0893,
+ "step": 7790
+ },
+ {
+ "epoch": 17.41,
+ "learning_rate": 1.2554131392857812e-05,
+ "loss": 0.0852,
+ "step": 7800
+ },
+ {
+ "epoch": 17.43,
+ "learning_rate": 1.253665287600259e-05,
+ "loss": 0.0828,
+ "step": 7810
+ },
+ {
+ "epoch": 17.46,
+ "learning_rate": 1.2519166072837368e-05,
+ "loss": 0.0883,
+ "step": 7820
+ },
+ {
+ "epoch": 17.48,
+ "learning_rate": 1.250167104048508e-05,
+ "loss": 0.0864,
+ "step": 7830
+ },
+ {
+ "epoch": 17.5,
+ "learning_rate": 1.248416783609555e-05,
+ "loss": 0.1761,
+ "step": 7840
+ },
+ {
+ "epoch": 17.52,
+ "learning_rate": 1.2466656516845293e-05,
+ "loss": 0.0797,
+ "step": 7850
+ },
+ {
+ "epoch": 17.54,
+ "learning_rate": 1.244913713993734e-05,
+ "loss": 0.0712,
+ "step": 7860
+ },
+ {
+ "epoch": 17.57,
+ "learning_rate": 1.2431609762601036e-05,
+ "loss": 0.0782,
+ "step": 7870
+ },
+ {
+ "epoch": 17.59,
+ "learning_rate": 1.241407444209186e-05,
+ "loss": 0.0674,
+ "step": 7880
+ },
+ {
+ "epoch": 17.61,
+ "learning_rate": 1.2396531235691245e-05,
+ "loss": 0.0744,
+ "step": 7890
+ },
+ {
+ "epoch": 17.63,
+ "learning_rate": 1.2378980200706376e-05,
+ "loss": 0.0691,
+ "step": 7900
+ },
+ {
+ "epoch": 17.66,
+ "learning_rate": 1.236142139447002e-05,
+ "loss": 0.0903,
+ "step": 7910
+ },
+ {
+ "epoch": 17.68,
+ "learning_rate": 1.2343854874340324e-05,
+ "loss": 0.1275,
+ "step": 7920
+ },
+ {
+ "epoch": 17.7,
+ "learning_rate": 1.2326280697700632e-05,
+ "loss": 0.1268,
+ "step": 7930
+ },
+ {
+ "epoch": 17.72,
+ "learning_rate": 1.2308698921959306e-05,
+ "loss": 0.0829,
+ "step": 7940
+ },
+ {
+ "epoch": 17.75,
+ "learning_rate": 1.2291109604549525e-05,
+ "loss": 0.1007,
+ "step": 7950
+ },
+ {
+ "epoch": 17.77,
+ "learning_rate": 1.2273512802929107e-05,
+ "loss": 0.0812,
+ "step": 7960
+ },
+ {
+ "epoch": 17.79,
+ "learning_rate": 1.2255908574580311e-05,
+ "loss": 0.086,
+ "step": 7970
+ },
+ {
+ "epoch": 17.81,
+ "learning_rate": 1.2238296977009672e-05,
+ "loss": 0.0975,
+ "step": 7980
+ },
+ {
+ "epoch": 17.83,
+ "learning_rate": 1.2220678067747785e-05,
+ "loss": 0.1009,
+ "step": 7990
+ },
+ {
+ "epoch": 17.86,
+ "learning_rate": 1.2203051904349128e-05,
+ "loss": 0.1109,
+ "step": 8000
+ },
+ {
+ "epoch": 17.88,
+ "learning_rate": 1.2185418544391885e-05,
+ "loss": 0.0898,
+ "step": 8010
+ },
+ {
+ "epoch": 17.9,
+ "learning_rate": 1.2167778045477743e-05,
+ "loss": 0.0846,
+ "step": 8020
+ },
+ {
+ "epoch": 17.92,
+ "learning_rate": 1.215013046523171e-05,
+ "loss": 0.088,
+ "step": 8030
+ },
+ {
+ "epoch": 17.95,
+ "learning_rate": 1.2132475861301928e-05,
+ "loss": 0.072,
+ "step": 8040
+ },
+ {
+ "epoch": 17.97,
+ "learning_rate": 1.2114814291359476e-05,
+ "loss": 0.0749,
+ "step": 8050
+ },
+ {
+ "epoch": 17.99,
+ "learning_rate": 1.20971458130982e-05,
+ "loss": 0.0712,
+ "step": 8060
+ },
+ {
+ "epoch": 18.01,
+ "learning_rate": 1.20794704842345e-05,
+ "loss": 0.0611,
+ "step": 8070
+ },
+ {
+ "epoch": 18.04,
+ "learning_rate": 1.2061788362507168e-05,
+ "loss": 0.0664,
+ "step": 8080
+ },
+ {
+ "epoch": 18.06,
+ "learning_rate": 1.204409950567717e-05,
+ "loss": 0.0761,
+ "step": 8090
+ },
+ {
+ "epoch": 18.08,
+ "learning_rate": 1.2026403971527487e-05,
+ "loss": 0.0663,
+ "step": 8100
+ },
+ {
+ "epoch": 18.1,
+ "learning_rate": 1.2008701817862906e-05,
+ "loss": 0.0731,
+ "step": 8110
+ },
+ {
+ "epoch": 18.12,
+ "learning_rate": 1.1990993102509838e-05,
+ "loss": 0.1079,
+ "step": 8120
+ },
+ {
+ "epoch": 18.15,
+ "learning_rate": 1.1973277883316128e-05,
+ "loss": 0.0633,
+ "step": 8130
+ },
+ {
+ "epoch": 18.17,
+ "learning_rate": 1.1955556218150872e-05,
+ "loss": 0.074,
+ "step": 8140
+ },
+ {
+ "epoch": 18.19,
+ "learning_rate": 1.1937828164904216e-05,
+ "loss": 0.0796,
+ "step": 8150
+ },
+ {
+ "epoch": 18.21,
+ "learning_rate": 1.1920093781487175e-05,
+ "loss": 0.0736,
+ "step": 8160
+ },
+ {
+ "epoch": 18.24,
+ "learning_rate": 1.1902353125831441e-05,
+ "loss": 0.0649,
+ "step": 8170
+ },
+ {
+ "epoch": 18.26,
+ "learning_rate": 1.1884606255889203e-05,
+ "loss": 0.0794,
+ "step": 8180
+ },
+ {
+ "epoch": 18.28,
+ "learning_rate": 1.1866853229632942e-05,
+ "loss": 0.0691,
+ "step": 8190
+ },
+ {
+ "epoch": 18.3,
+ "learning_rate": 1.1849094105055248e-05,
+ "loss": 0.0682,
+ "step": 8200
+ },
+ {
+ "epoch": 18.33,
+ "learning_rate": 1.1831328940168638e-05,
+ "loss": 0.0804,
+ "step": 8210
+ },
+ {
+ "epoch": 18.35,
+ "learning_rate": 1.181355779300536e-05,
+ "loss": 0.0818,
+ "step": 8220
+ },
+ {
+ "epoch": 18.37,
+ "learning_rate": 1.1795780721617199e-05,
+ "loss": 0.0884,
+ "step": 8230
+ },
+ {
+ "epoch": 18.39,
+ "learning_rate": 1.1777997784075294e-05,
+ "loss": 0.0829,
+ "step": 8240
+ },
+ {
+ "epoch": 18.42,
+ "learning_rate": 1.176020903846995e-05,
+ "loss": 0.0806,
+ "step": 8250
+ },
+ {
+ "epoch": 18.44,
+ "learning_rate": 1.1742414542910444e-05,
+ "loss": 0.0745,
+ "step": 8260
+ },
+ {
+ "epoch": 18.46,
+ "learning_rate": 1.1724614355524832e-05,
+ "loss": 0.1045,
+ "step": 8270
+ },
+ {
+ "epoch": 18.48,
+ "learning_rate": 1.1706808534459768e-05,
+ "loss": 0.068,
+ "step": 8280
+ },
+ {
+ "epoch": 18.5,
+ "learning_rate": 1.16889971378803e-05,
+ "loss": 0.1344,
+ "step": 8290
+ },
+ {
+ "epoch": 18.53,
+ "learning_rate": 1.1671180223969705e-05,
+ "loss": 0.0995,
+ "step": 8300
+ },
+ {
+ "epoch": 18.55,
+ "learning_rate": 1.1653357850929268e-05,
+ "loss": 0.0694,
+ "step": 8310
+ },
+ {
+ "epoch": 18.57,
+ "learning_rate": 1.1635530076978115e-05,
+ "loss": 0.0817,
+ "step": 8320
+ },
+ {
+ "epoch": 18.59,
+ "learning_rate": 1.161769696035301e-05,
+ "loss": 0.113,
+ "step": 8330
+ },
+ {
+ "epoch": 18.62,
+ "learning_rate": 1.1599858559308175e-05,
+ "loss": 0.076,
+ "step": 8340
+ },
+ {
+ "epoch": 18.64,
+ "learning_rate": 1.158201493211509e-05,
+ "loss": 0.0825,
+ "step": 8350
+ },
+ {
+ "epoch": 18.66,
+ "learning_rate": 1.156416613706231e-05,
+ "loss": 0.0812,
+ "step": 8360
+ },
+ {
+ "epoch": 18.68,
+ "learning_rate": 1.1546312232455266e-05,
+ "loss": 0.1022,
+ "step": 8370
+ },
+ {
+ "epoch": 18.71,
+ "learning_rate": 1.152845327661609e-05,
+ "loss": 0.0755,
+ "step": 8380
+ },
+ {
+ "epoch": 18.73,
+ "learning_rate": 1.1510589327883406e-05,
+ "loss": 0.1011,
+ "step": 8390
+ },
+ {
+ "epoch": 18.75,
+ "learning_rate": 1.1492720444612148e-05,
+ "loss": 0.0782,
+ "step": 8400
+ },
+ {
+ "epoch": 18.77,
+ "learning_rate": 1.1474846685173374e-05,
+ "loss": 0.0872,
+ "step": 8410
+ },
+ {
+ "epoch": 18.79,
+ "learning_rate": 1.1456968107954066e-05,
+ "loss": 0.0794,
+ "step": 8420
+ },
+ {
+ "epoch": 18.82,
+ "learning_rate": 1.143908477135695e-05,
+ "loss": 0.0759,
+ "step": 8430
+ },
+ {
+ "epoch": 18.84,
+ "learning_rate": 1.1421196733800291e-05,
+ "loss": 0.074,
+ "step": 8440
+ },
+ {
+ "epoch": 18.86,
+ "learning_rate": 1.1403304053717719e-05,
+ "loss": 0.0872,
+ "step": 8450
+ },
+ {
+ "epoch": 18.88,
+ "learning_rate": 1.138540678955802e-05,
+ "loss": 0.0677,
+ "step": 8460
+ },
+ {
+ "epoch": 18.91,
+ "learning_rate": 1.1367504999784963e-05,
+ "loss": 0.0744,
+ "step": 8470
+ },
+ {
+ "epoch": 18.93,
+ "learning_rate": 1.1349598742877097e-05,
+ "loss": 0.0703,
+ "step": 8480
+ },
+ {
+ "epoch": 18.95,
+ "learning_rate": 1.1331688077327563e-05,
+ "loss": 0.0726,
+ "step": 8490
+ },
+ {
+ "epoch": 18.97,
+ "learning_rate": 1.1313773061643905e-05,
+ "loss": 0.0727,
+ "step": 8500
+ },
+ {
+ "epoch": 19.0,
+ "learning_rate": 1.1295853754347876e-05,
+ "loss": 0.0903,
+ "step": 8510
+ },
+ {
+ "epoch": 19.02,
+ "learning_rate": 1.1277930213975249e-05,
+ "loss": 0.0665,
+ "step": 8520
+ },
+ {
+ "epoch": 19.04,
+ "learning_rate": 1.1260002499075617e-05,
+ "loss": 0.0773,
+ "step": 8530
+ },
+ {
+ "epoch": 19.06,
+ "learning_rate": 1.1242070668212227e-05,
+ "loss": 0.0666,
+ "step": 8540
+ },
+ {
+ "epoch": 19.08,
+ "learning_rate": 1.1224134779961758e-05,
+ "loss": 0.066,
+ "step": 8550
+ },
+ {
+ "epoch": 19.11,
+ "learning_rate": 1.1206194892914142e-05,
+ "loss": 0.0651,
+ "step": 8560
+ },
+ {
+ "epoch": 19.13,
+ "learning_rate": 1.1188251065672382e-05,
+ "loss": 0.0779,
+ "step": 8570
+ },
+ {
+ "epoch": 19.15,
+ "learning_rate": 1.117030335685235e-05,
+ "loss": 0.082,
+ "step": 8580
+ },
+ {
+ "epoch": 19.17,
+ "learning_rate": 1.1152351825082588e-05,
+ "loss": 0.1009,
+ "step": 8590
+ },
+ {
+ "epoch": 19.2,
+ "learning_rate": 1.1134396529004143e-05,
+ "loss": 0.0726,
+ "step": 8600
+ },
+ {
+ "epoch": 19.22,
+ "learning_rate": 1.1116437527270343e-05,
+ "loss": 0.0733,
+ "step": 8610
+ },
+ {
+ "epoch": 19.24,
+ "learning_rate": 1.109847487854663e-05,
+ "loss": 0.0736,
+ "step": 8620
+ },
+ {
+ "epoch": 19.26,
+ "learning_rate": 1.1080508641510357e-05,
+ "loss": 0.0801,
+ "step": 8630
+ },
+ {
+ "epoch": 19.29,
+ "learning_rate": 1.1062538874850597e-05,
+ "loss": 0.0721,
+ "step": 8640
+ },
+ {
+ "epoch": 19.31,
+ "learning_rate": 1.1044565637267957e-05,
+ "loss": 0.07,
+ "step": 8650
+ },
+ {
+ "epoch": 19.33,
+ "learning_rate": 1.1026588987474379e-05,
+ "loss": 0.074,
+ "step": 8660
+ },
+ {
+ "epoch": 19.35,
+ "learning_rate": 1.100860898419295e-05,
+ "loss": 0.0613,
+ "step": 8670
+ },
+ {
+ "epoch": 19.38,
+ "learning_rate": 1.0990625686157714e-05,
+ "loss": 0.0902,
+ "step": 8680
+ },
+ {
+ "epoch": 19.4,
+ "learning_rate": 1.097263915211348e-05,
+ "loss": 0.0792,
+ "step": 8690
+ },
+ {
+ "epoch": 19.42,
+ "learning_rate": 1.0954649440815625e-05,
+ "loss": 0.059,
+ "step": 8700
+ },
+ {
+ "epoch": 19.44,
+ "learning_rate": 1.0936656611029901e-05,
+ "loss": 0.0846,
+ "step": 8710
+ },
+ {
+ "epoch": 19.46,
+ "learning_rate": 1.091866072153226e-05,
+ "loss": 0.0788,
+ "step": 8720
+ },
+ {
+ "epoch": 19.49,
+ "learning_rate": 1.090066183110863e-05,
+ "loss": 0.0726,
+ "step": 8730
+ },
+ {
+ "epoch": 19.51,
+ "learning_rate": 1.0882659998554759e-05,
+ "loss": 0.0739,
+ "step": 8740
+ },
+ {
+ "epoch": 19.53,
+ "learning_rate": 1.0864655282675997e-05,
+ "loss": 0.0741,
+ "step": 8750
+ },
+ {
+ "epoch": 19.55,
+ "learning_rate": 1.0846647742287116e-05,
+ "loss": 0.0718,
+ "step": 8760
+ },
+ {
+ "epoch": 19.58,
+ "learning_rate": 1.0828637436212111e-05,
+ "loss": 0.0617,
+ "step": 8770
+ },
+ {
+ "epoch": 19.6,
+ "learning_rate": 1.0810624423284012e-05,
+ "loss": 0.0975,
+ "step": 8780
+ },
+ {
+ "epoch": 19.62,
+ "learning_rate": 1.07926087623447e-05,
+ "loss": 0.0578,
+ "step": 8790
+ },
+ {
+ "epoch": 19.64,
+ "learning_rate": 1.0774590512244694e-05,
+ "loss": 0.1115,
+ "step": 8800
+ },
+ {
+ "epoch": 19.67,
+ "learning_rate": 1.0756569731842978e-05,
+ "loss": 0.0672,
+ "step": 8810
+ },
+ {
+ "epoch": 19.69,
+ "learning_rate": 1.07385464800068e-05,
+ "loss": 0.0745,
+ "step": 8820
+ },
+ {
+ "epoch": 19.71,
+ "learning_rate": 1.0720520815611476e-05,
+ "loss": 0.0707,
+ "step": 8830
+ },
+ {
+ "epoch": 19.73,
+ "learning_rate": 1.0702492797540214e-05,
+ "loss": 0.0779,
+ "step": 8840
+ },
+ {
+ "epoch": 19.75,
+ "learning_rate": 1.06844624846839e-05,
+ "loss": 0.075,
+ "step": 8850
+ },
+ {
+ "epoch": 19.78,
+ "learning_rate": 1.0666429935940925e-05,
+ "loss": 0.0603,
+ "step": 8860
+ },
+ {
+ "epoch": 19.8,
+ "learning_rate": 1.0648395210216975e-05,
+ "loss": 0.0877,
+ "step": 8870
+ },
+ {
+ "epoch": 19.82,
+ "learning_rate": 1.0630358366424856e-05,
+ "loss": 0.0771,
+ "step": 8880
+ },
+ {
+ "epoch": 19.84,
+ "learning_rate": 1.0612319463484286e-05,
+ "loss": 0.0716,
+ "step": 8890
+ },
+ {
+ "epoch": 19.87,
+ "learning_rate": 1.0594278560321713e-05,
+ "loss": 0.0879,
+ "step": 8900
+ },
+ {
+ "epoch": 19.89,
+ "learning_rate": 1.0576235715870119e-05,
+ "loss": 0.0771,
+ "step": 8910
+ },
+ {
+ "epoch": 19.91,
+ "learning_rate": 1.0558190989068822e-05,
+ "loss": 0.0703,
+ "step": 8920
+ },
+ {
+ "epoch": 19.93,
+ "learning_rate": 1.0540144438863302e-05,
+ "loss": 0.09,
+ "step": 8930
+ },
+ {
+ "epoch": 19.96,
+ "learning_rate": 1.052209612420498e-05,
+ "loss": 0.0703,
+ "step": 8940
+ },
+ {
+ "epoch": 19.98,
+ "learning_rate": 1.050404610405105e-05,
+ "loss": 0.083,
+ "step": 8950
+ },
+ {
+ "epoch": 20.0,
+ "learning_rate": 1.0485994437364278e-05,
+ "loss": 0.0824,
+ "step": 8960
+ },
+ {
+ "epoch": 20.02,
+ "learning_rate": 1.0467941183112801e-05,
+ "loss": 0.0616,
+ "step": 8970
+ },
+ {
+ "epoch": 20.04,
+ "learning_rate": 1.0449886400269952e-05,
+ "loss": 0.0507,
+ "step": 8980
+ },
+ {
+ "epoch": 20.07,
+ "learning_rate": 1.0431830147814049e-05,
+ "loss": 0.0626,
+ "step": 8990
+ },
+ {
+ "epoch": 20.09,
+ "learning_rate": 1.0413772484728211e-05,
+ "loss": 0.0551,
+ "step": 9000
+ },
+ {
+ "epoch": 20.11,
+ "learning_rate": 1.0395713470000173e-05,
+ "loss": 0.095,
+ "step": 9010
+ },
+ {
+ "epoch": 20.13,
+ "learning_rate": 1.0377653162622076e-05,
+ "loss": 0.0536,
+ "step": 9020
+ },
+ {
+ "epoch": 20.16,
+ "learning_rate": 1.0359591621590292e-05,
+ "loss": 0.0752,
+ "step": 9030
+ },
+ {
+ "epoch": 20.18,
+ "learning_rate": 1.034152890590521e-05,
+ "loss": 0.064,
+ "step": 9040
+ },
+ {
+ "epoch": 20.2,
+ "learning_rate": 1.0323465074571078e-05,
+ "loss": 0.0674,
+ "step": 9050
+ },
+ {
+ "epoch": 20.22,
+ "learning_rate": 1.0305400186595764e-05,
+ "loss": 0.0656,
+ "step": 9060
+ },
+ {
+ "epoch": 20.25,
+ "learning_rate": 1.0287334300990602e-05,
+ "loss": 0.0599,
+ "step": 9070
+ },
+ {
+ "epoch": 20.27,
+ "learning_rate": 1.026926747677018e-05,
+ "loss": 0.0881,
+ "step": 9080
+ },
+ {
+ "epoch": 20.29,
+ "learning_rate": 1.025119977295216e-05,
+ "loss": 0.0647,
+ "step": 9090
+ },
+ {
+ "epoch": 20.31,
+ "learning_rate": 1.0233131248557067e-05,
+ "loss": 0.07,
+ "step": 9100
+ },
+ {
+ "epoch": 20.33,
+ "learning_rate": 1.0215061962608111e-05,
+ "loss": 0.063,
+ "step": 9110
+ },
+ {
+ "epoch": 20.36,
+ "learning_rate": 1.0196991974130986e-05,
+ "loss": 0.0667,
+ "step": 9120
+ },
+ {
+ "epoch": 20.38,
+ "learning_rate": 1.017892134215369e-05,
+ "loss": 0.0667,
+ "step": 9130
+ },
+ {
+ "epoch": 20.4,
+ "learning_rate": 1.0160850125706314e-05,
+ "loss": 0.1344,
+ "step": 9140
+ },
+ {
+ "epoch": 20.42,
+ "learning_rate": 1.0142778383820861e-05,
+ "loss": 0.0714,
+ "step": 9150
+ },
+ {
+ "epoch": 20.45,
+ "learning_rate": 1.0124706175531054e-05,
+ "loss": 0.0604,
+ "step": 9160
+ },
+ {
+ "epoch": 20.47,
+ "learning_rate": 1.0106633559872135e-05,
+ "loss": 0.072,
+ "step": 9170
+ },
+ {
+ "epoch": 20.49,
+ "learning_rate": 1.0088560595880676e-05,
+ "loss": 0.0694,
+ "step": 9180
+ },
+ {
+ "epoch": 20.51,
+ "learning_rate": 1.0070487342594392e-05,
+ "loss": 0.1011,
+ "step": 9190
+ },
+ {
+ "epoch": 20.54,
+ "learning_rate": 1.005241385905194e-05,
+ "loss": 0.0657,
+ "step": 9200
+ },
+ {
+ "epoch": 20.56,
+ "learning_rate": 1.0034340204292728e-05,
+ "loss": 0.07,
+ "step": 9210
+ },
+ {
+ "epoch": 20.58,
+ "learning_rate": 1.0016266437356727e-05,
+ "loss": 0.0675,
+ "step": 9220
+ },
+ {
+ "epoch": 20.6,
+ "learning_rate": 9.998192617284271e-06,
+ "loss": 0.0648,
+ "step": 9230
+ },
+ {
+ "epoch": 20.62,
+ "learning_rate": 9.980118803115867e-06,
+ "loss": 0.0719,
+ "step": 9240
+ },
+ {
+ "epoch": 20.65,
+ "learning_rate": 9.962045053892004e-06,
+ "loss": 0.0905,
+ "step": 9250
+ },
+ {
+ "epoch": 20.67,
+ "learning_rate": 9.94397142865296e-06,
+ "loss": 0.0637,
+ "step": 9260
+ },
+ {
+ "epoch": 20.69,
+ "learning_rate": 9.925897986438613e-06,
+ "loss": 0.1299,
+ "step": 9270
+ },
+ {
+ "epoch": 20.71,
+ "learning_rate": 9.907824786288226e-06,
+ "loss": 0.0733,
+ "step": 9280
+ },
+ {
+ "epoch": 20.74,
+ "learning_rate": 9.889751887240296e-06,
+ "loss": 0.075,
+ "step": 9290
+ },
+ {
+ "epoch": 20.76,
+ "learning_rate": 9.87167934833231e-06,
+ "loss": 0.0997,
+ "step": 9300
+ },
+ {
+ "epoch": 20.78,
+ "learning_rate": 9.853607228600602e-06,
+ "loss": 0.0735,
+ "step": 9310
+ },
+ {
+ "epoch": 20.8,
+ "learning_rate": 9.835535587080118e-06,
+ "loss": 0.0721,
+ "step": 9320
+ },
+ {
+ "epoch": 20.83,
+ "learning_rate": 9.817464482804257e-06,
+ "loss": 0.0745,
+ "step": 9330
+ },
+ {
+ "epoch": 20.85,
+ "learning_rate": 9.799393974804651e-06,
+ "loss": 0.067,
+ "step": 9340
+ },
+ {
+ "epoch": 20.87,
+ "learning_rate": 9.781324122110993e-06,
+ "loss": 0.0751,
+ "step": 9350
+ },
+ {
+ "epoch": 20.89,
+ "learning_rate": 9.763254983750829e-06,
+ "loss": 0.0648,
+ "step": 9360
+ },
+ {
+ "epoch": 20.92,
+ "learning_rate": 9.745186618749373e-06,
+ "loss": 0.0675,
+ "step": 9370
+ },
+ {
+ "epoch": 20.94,
+ "learning_rate": 9.727119086129321e-06,
+ "loss": 0.0932,
+ "step": 9380
+ },
+ {
+ "epoch": 20.96,
+ "learning_rate": 9.709052444910636e-06,
+ "loss": 0.0592,
+ "step": 9390
+ },
+ {
+ "epoch": 20.98,
+ "learning_rate": 9.690986754110378e-06,
+ "loss": 0.0739,
+ "step": 9400
+ },
+ {
+ "epoch": 21.0,
+ "learning_rate": 9.6729220727425e-06,
+ "loss": 0.0635,
+ "step": 9410
+ },
+ {
+ "epoch": 21.03,
+ "learning_rate": 9.654858459817663e-06,
+ "loss": 0.0569,
+ "step": 9420
+ },
+ {
+ "epoch": 21.05,
+ "learning_rate": 9.636795974343023e-06,
+ "loss": 0.0555,
+ "step": 9430
+ },
+ {
+ "epoch": 21.07,
+ "learning_rate": 9.61873467532207e-06,
+ "loss": 0.0483,
+ "step": 9440
+ },
+ {
+ "epoch": 21.09,
+ "learning_rate": 9.600674621754406e-06,
+ "loss": 0.0511,
+ "step": 9450
+ },
+ {
+ "epoch": 21.12,
+ "learning_rate": 9.582615872635578e-06,
+ "loss": 0.0762,
+ "step": 9460
+ },
+ {
+ "epoch": 21.14,
+ "learning_rate": 9.564558486956853e-06,
+ "loss": 0.0705,
+ "step": 9470
+ },
+ {
+ "epoch": 21.16,
+ "learning_rate": 9.546502523705057e-06,
+ "loss": 0.0427,
+ "step": 9480
+ },
+ {
+ "epoch": 21.18,
+ "learning_rate": 9.528448041862375e-06,
+ "loss": 0.0709,
+ "step": 9490
+ },
+ {
+ "epoch": 21.21,
+ "learning_rate": 9.510395100406136e-06,
+ "loss": 0.0548,
+ "step": 9500
+ },
+ {
+ "epoch": 21.23,
+ "learning_rate": 9.492343758308651e-06,
+ "loss": 0.0764,
+ "step": 9510
+ },
+ {
+ "epoch": 21.25,
+ "learning_rate": 9.474294074536996e-06,
+ "loss": 0.0789,
+ "step": 9520
+ },
+ {
+ "epoch": 21.27,
+ "learning_rate": 9.456246108052844e-06,
+ "loss": 0.0708,
+ "step": 9530
+ },
+ {
+ "epoch": 21.29,
+ "learning_rate": 9.438199917812241e-06,
+ "loss": 0.0647,
+ "step": 9540
+ },
+ {
+ "epoch": 21.32,
+ "learning_rate": 9.420155562765443e-06,
+ "loss": 0.0583,
+ "step": 9550
+ },
+ {
+ "epoch": 21.34,
+ "learning_rate": 9.402113101856705e-06,
+ "loss": 0.0564,
+ "step": 9560
+ },
+ {
+ "epoch": 21.36,
+ "learning_rate": 9.384072594024103e-06,
+ "loss": 0.0851,
+ "step": 9570
+ },
+ {
+ "epoch": 21.38,
+ "learning_rate": 9.366034098199317e-06,
+ "loss": 0.0993,
+ "step": 9580
+ },
+ {
+ "epoch": 21.41,
+ "learning_rate": 9.347997673307473e-06,
+ "loss": 0.067,
+ "step": 9590
+ },
+ {
+ "epoch": 21.43,
+ "learning_rate": 9.329963378266919e-06,
+ "loss": 0.095,
+ "step": 9600
+ },
+ {
+ "epoch": 21.45,
+ "learning_rate": 9.31193127198905e-06,
+ "loss": 0.0569,
+ "step": 9610
+ },
+ {
+ "epoch": 21.47,
+ "learning_rate": 9.293901413378116e-06,
+ "loss": 0.0558,
+ "step": 9620
+ },
+ {
+ "epoch": 21.5,
+ "learning_rate": 9.275873861331012e-06,
+ "loss": 0.0627,
+ "step": 9630
+ },
+ {
+ "epoch": 21.52,
+ "learning_rate": 9.257848674737112e-06,
+ "loss": 0.0745,
+ "step": 9640
+ },
+ {
+ "epoch": 21.54,
+ "learning_rate": 9.239825912478054e-06,
+ "loss": 0.0547,
+ "step": 9650
+ },
+ {
+ "epoch": 21.56,
+ "learning_rate": 9.221805633427564e-06,
+ "loss": 0.0792,
+ "step": 9660
+ },
+ {
+ "epoch": 21.58,
+ "learning_rate": 9.203787896451246e-06,
+ "loss": 0.0602,
+ "step": 9670
+ },
+ {
+ "epoch": 21.61,
+ "learning_rate": 9.185772760406408e-06,
+ "loss": 0.0566,
+ "step": 9680
+ },
+ {
+ "epoch": 21.63,
+ "learning_rate": 9.167760284141859e-06,
+ "loss": 0.0572,
+ "step": 9690
+ },
+ {
+ "epoch": 21.65,
+ "learning_rate": 9.149750526497725e-06,
+ "loss": 0.0576,
+ "step": 9700
+ },
+ {
+ "epoch": 21.67,
+ "learning_rate": 9.131743546305235e-06,
+ "loss": 0.0584,
+ "step": 9710
+ },
+ {
+ "epoch": 21.7,
+ "learning_rate": 9.113739402386566e-06,
+ "loss": 0.0759,
+ "step": 9720
+ },
+ {
+ "epoch": 21.72,
+ "learning_rate": 9.095738153554624e-06,
+ "loss": 0.0495,
+ "step": 9730
+ },
+ {
+ "epoch": 21.74,
+ "learning_rate": 9.077739858612843e-06,
+ "loss": 0.0691,
+ "step": 9740
+ },
+ {
+ "epoch": 21.76,
+ "learning_rate": 9.059744576355027e-06,
+ "loss": 0.0618,
+ "step": 9750
+ },
+ {
+ "epoch": 21.79,
+ "learning_rate": 9.041752365565125e-06,
+ "loss": 0.1139,
+ "step": 9760
+ },
+ {
+ "epoch": 21.81,
+ "learning_rate": 9.023763285017065e-06,
+ "loss": 0.0635,
+ "step": 9770
+ },
+ {
+ "epoch": 21.83,
+ "learning_rate": 9.005777393474534e-06,
+ "loss": 0.0819,
+ "step": 9780
+ },
+ {
+ "epoch": 21.85,
+ "learning_rate": 8.987794749690819e-06,
+ "loss": 0.0575,
+ "step": 9790
+ },
+ {
+ "epoch": 21.88,
+ "learning_rate": 8.969815412408583e-06,
+ "loss": 0.0621,
+ "step": 9800
+ },
+ {
+ "epoch": 21.9,
+ "learning_rate": 8.951839440359701e-06,
+ "loss": 0.0766,
+ "step": 9810
+ },
+ {
+ "epoch": 21.92,
+ "learning_rate": 8.93386689226504e-06,
+ "loss": 0.0692,
+ "step": 9820
+ },
+ {
+ "epoch": 21.94,
+ "learning_rate": 8.915897826834295e-06,
+ "loss": 0.0595,
+ "step": 9830
+ },
+ {
+ "epoch": 21.96,
+ "learning_rate": 8.89793230276578e-06,
+ "loss": 0.0918,
+ "step": 9840
+ },
+ {
+ "epoch": 21.99,
+ "learning_rate": 8.879970378746238e-06,
+ "loss": 0.074,
+ "step": 9850
+ },
+ {
+ "epoch": 22.01,
+ "learning_rate": 8.862012113450662e-06,
+ "loss": 0.0765,
+ "step": 9860
+ },
+ {
+ "epoch": 22.03,
+ "learning_rate": 8.844057565542074e-06,
+ "loss": 0.0531,
+ "step": 9870
+ },
+ {
+ "epoch": 22.05,
+ "learning_rate": 8.826106793671376e-06,
+ "loss": 0.0569,
+ "step": 9880
+ },
+ {
+ "epoch": 22.08,
+ "learning_rate": 8.808159856477115e-06,
+ "loss": 0.054,
+ "step": 9890
+ },
+ {
+ "epoch": 22.1,
+ "learning_rate": 8.790216812585327e-06,
+ "loss": 0.0655,
+ "step": 9900
+ },
+ {
+ "epoch": 22.12,
+ "learning_rate": 8.772277720609312e-06,
+ "loss": 0.0521,
+ "step": 9910
+ },
+ {
+ "epoch": 22.14,
+ "learning_rate": 8.754342639149486e-06,
+ "loss": 0.066,
+ "step": 9920
+ },
+ {
+ "epoch": 22.17,
+ "learning_rate": 8.736411626793139e-06,
+ "loss": 0.0822,
+ "step": 9930
+ },
+ {
+ "epoch": 22.19,
+ "learning_rate": 8.718484742114285e-06,
+ "loss": 0.0576,
+ "step": 9940
+ },
+ {
+ "epoch": 22.21,
+ "learning_rate": 8.700562043673448e-06,
+ "loss": 0.0585,
+ "step": 9950
+ },
+ {
+ "epoch": 22.23,
+ "learning_rate": 8.682643590017474e-06,
+ "loss": 0.0546,
+ "step": 9960
+ },
+ {
+ "epoch": 22.25,
+ "learning_rate": 8.664729439679354e-06,
+ "loss": 0.0921,
+ "step": 9970
+ },
+ {
+ "epoch": 22.28,
+ "learning_rate": 8.646819651178008e-06,
+ "loss": 0.0628,
+ "step": 9980
+ },
+ {
+ "epoch": 22.3,
+ "learning_rate": 8.628914283018119e-06,
+ "loss": 0.0784,
+ "step": 9990
+ },
+ {
+ "epoch": 22.32,
+ "learning_rate": 8.61101339368992e-06,
+ "loss": 0.0729,
+ "step": 10000
+ },
+ {
+ "epoch": 22.34,
+ "learning_rate": 8.593117041669024e-06,
+ "loss": 0.0491,
+ "step": 10010
+ },
+ {
+ "epoch": 22.37,
+ "learning_rate": 8.57522528541621e-06,
+ "loss": 0.0643,
+ "step": 10020
+ },
+ {
+ "epoch": 22.39,
+ "learning_rate": 8.55733818337726e-06,
+ "loss": 0.0614,
+ "step": 10030
+ },
+ {
+ "epoch": 22.41,
+ "learning_rate": 8.539455793982737e-06,
+ "loss": 0.065,
+ "step": 10040
+ },
+ {
+ "epoch": 22.43,
+ "learning_rate": 8.521578175647823e-06,
+ "loss": 0.0823,
+ "step": 10050
+ },
+ {
+ "epoch": 22.46,
+ "learning_rate": 8.503705386772098e-06,
+ "loss": 0.0591,
+ "step": 10060
+ },
+ {
+ "epoch": 22.48,
+ "learning_rate": 8.485837485739384e-06,
+ "loss": 0.0503,
+ "step": 10070
+ },
+ {
+ "epoch": 22.5,
+ "learning_rate": 8.467974530917524e-06,
+ "loss": 0.0549,
+ "step": 10080
+ },
+ {
+ "epoch": 22.52,
+ "learning_rate": 8.450116580658208e-06,
+ "loss": 0.0587,
+ "step": 10090
+ },
+ {
+ "epoch": 22.54,
+ "learning_rate": 8.432263693296783e-06,
+ "loss": 0.0557,
+ "step": 10100
+ },
+ {
+ "epoch": 22.57,
+ "learning_rate": 8.414415927152042e-06,
+ "loss": 0.0772,
+ "step": 10110
+ },
+ {
+ "epoch": 22.59,
+ "learning_rate": 8.396573340526069e-06,
+ "loss": 0.0604,
+ "step": 10120
+ },
+ {
+ "epoch": 22.61,
+ "learning_rate": 8.37873599170401e-06,
+ "loss": 0.0641,
+ "step": 10130
+ },
+ {
+ "epoch": 22.63,
+ "learning_rate": 8.360903938953914e-06,
+ "loss": 0.0577,
+ "step": 10140
+ },
+ {
+ "epoch": 22.66,
+ "learning_rate": 8.343077240526522e-06,
+ "loss": 0.0742,
+ "step": 10150
+ },
+ {
+ "epoch": 22.68,
+ "learning_rate": 8.325255954655093e-06,
+ "loss": 0.0563,
+ "step": 10160
+ },
+ {
+ "epoch": 22.7,
+ "learning_rate": 8.307440139555192e-06,
+ "loss": 0.0672,
+ "step": 10170
+ },
+ {
+ "epoch": 22.72,
+ "learning_rate": 8.289629853424526e-06,
+ "loss": 0.071,
+ "step": 10180
+ },
+ {
+ "epoch": 22.75,
+ "learning_rate": 8.271825154442732e-06,
+ "loss": 0.0648,
+ "step": 10190
+ },
+ {
+ "epoch": 22.77,
+ "learning_rate": 8.2540261007712e-06,
+ "loss": 0.0585,
+ "step": 10200
+ },
+ {
+ "epoch": 22.79,
+ "learning_rate": 8.236232750552881e-06,
+ "loss": 0.0604,
+ "step": 10210
+ },
+ {
+ "epoch": 22.81,
+ "learning_rate": 8.218445161912088e-06,
+ "loss": 0.0506,
+ "step": 10220
+ },
+ {
+ "epoch": 22.83,
+ "learning_rate": 8.20066339295432e-06,
+ "loss": 0.0474,
+ "step": 10230
+ },
+ {
+ "epoch": 22.86,
+ "learning_rate": 8.182887501766059e-06,
+ "loss": 0.0513,
+ "step": 10240
+ },
+ {
+ "epoch": 22.88,
+ "learning_rate": 8.165117546414595e-06,
+ "loss": 0.0754,
+ "step": 10250
+ },
+ {
+ "epoch": 22.9,
+ "learning_rate": 8.147353584947818e-06,
+ "loss": 0.0691,
+ "step": 10260
+ },
+ {
+ "epoch": 22.92,
+ "learning_rate": 8.129595675394045e-06,
+ "loss": 0.0599,
+ "step": 10270
+ },
+ {
+ "epoch": 22.95,
+ "learning_rate": 8.11184387576182e-06,
+ "loss": 0.0625,
+ "step": 10280
+ },
+ {
+ "epoch": 22.97,
+ "learning_rate": 8.094098244039734e-06,
+ "loss": 0.0613,
+ "step": 10290
+ },
+ {
+ "epoch": 22.99,
+ "learning_rate": 8.076358838196216e-06,
+ "loss": 0.0683,
+ "step": 10300
+ },
+ {
+ "epoch": 23.01,
+ "learning_rate": 8.058625716179375e-06,
+ "loss": 0.0809,
+ "step": 10310
+ },
+ {
+ "epoch": 23.04,
+ "learning_rate": 8.04089893591678e-06,
+ "loss": 0.0757,
+ "step": 10320
+ },
+ {
+ "epoch": 23.06,
+ "learning_rate": 8.023178555315291e-06,
+ "loss": 0.0543,
+ "step": 10330
+ },
+ {
+ "epoch": 23.08,
+ "learning_rate": 8.005464632260862e-06,
+ "loss": 0.042,
+ "step": 10340
+ },
+ {
+ "epoch": 23.1,
+ "learning_rate": 7.987757224618346e-06,
+ "loss": 0.05,
+ "step": 10350
+ },
+ {
+ "epoch": 23.12,
+ "learning_rate": 7.970056390231323e-06,
+ "loss": 0.0933,
+ "step": 10360
+ },
+ {
+ "epoch": 23.15,
+ "learning_rate": 7.952362186921889e-06,
+ "loss": 0.0619,
+ "step": 10370
+ },
+ {
+ "epoch": 23.17,
+ "learning_rate": 7.934674672490488e-06,
+ "loss": 0.0677,
+ "step": 10380
+ },
+ {
+ "epoch": 23.19,
+ "learning_rate": 7.916993904715708e-06,
+ "loss": 0.0579,
+ "step": 10390
+ },
+ {
+ "epoch": 23.21,
+ "learning_rate": 7.899319941354107e-06,
+ "loss": 0.0573,
+ "step": 10400
+ },
+ {
+ "epoch": 23.24,
+ "learning_rate": 7.881652840140001e-06,
+ "loss": 0.0481,
+ "step": 10410
+ },
+ {
+ "epoch": 23.26,
+ "learning_rate": 7.863992658785302e-06,
+ "loss": 0.0543,
+ "step": 10420
+ },
+ {
+ "epoch": 23.28,
+ "learning_rate": 7.846339454979312e-06,
+ "loss": 0.0541,
+ "step": 10430
+ },
+ {
+ "epoch": 23.3,
+ "learning_rate": 7.828693286388542e-06,
+ "loss": 0.0524,
+ "step": 10440
+ },
+ {
+ "epoch": 23.33,
+ "learning_rate": 7.811054210656526e-06,
+ "loss": 0.0484,
+ "step": 10450
+ },
+ {
+ "epoch": 23.35,
+ "learning_rate": 7.793422285403614e-06,
+ "loss": 0.052,
+ "step": 10460
+ },
+ {
+ "epoch": 23.37,
+ "learning_rate": 7.775797568226816e-06,
+ "loss": 0.0752,
+ "step": 10470
+ },
+ {
+ "epoch": 23.39,
+ "learning_rate": 7.758180116699578e-06,
+ "loss": 0.0462,
+ "step": 10480
+ },
+ {
+ "epoch": 23.42,
+ "learning_rate": 7.74056998837163e-06,
+ "loss": 0.0552,
+ "step": 10490
+ },
+ {
+ "epoch": 23.44,
+ "learning_rate": 7.722967240768761e-06,
+ "loss": 0.0595,
+ "step": 10500
+ },
+ {
+ "epoch": 23.46,
+ "learning_rate": 7.705371931392668e-06,
+ "loss": 0.0595,
+ "step": 10510
+ },
+ {
+ "epoch": 23.48,
+ "learning_rate": 7.687784117720736e-06,
+ "loss": 0.057,
+ "step": 10520
+ },
+ {
+ "epoch": 23.5,
+ "learning_rate": 7.670203857205877e-06,
+ "loss": 0.0494,
+ "step": 10530
+ },
+ {
+ "epoch": 23.53,
+ "learning_rate": 7.652631207276311e-06,
+ "loss": 0.0594,
+ "step": 10540
+ },
+ {
+ "epoch": 23.55,
+ "learning_rate": 7.635066225335417e-06,
+ "loss": 0.0508,
+ "step": 10550
+ },
+ {
+ "epoch": 23.57,
+ "learning_rate": 7.617508968761519e-06,
+ "loss": 0.0537,
+ "step": 10560
+ },
+ {
+ "epoch": 23.59,
+ "learning_rate": 7.599959494907695e-06,
+ "loss": 0.0798,
+ "step": 10570
+ },
+ {
+ "epoch": 23.62,
+ "learning_rate": 7.582417861101614e-06,
+ "loss": 0.0538,
+ "step": 10580
+ },
+ {
+ "epoch": 23.64,
+ "learning_rate": 7.564884124645325e-06,
+ "loss": 0.0662,
+ "step": 10590
+ },
+ {
+ "epoch": 23.66,
+ "learning_rate": 7.547358342815089e-06,
+ "loss": 0.0714,
+ "step": 10600
+ },
+ {
+ "epoch": 23.68,
+ "learning_rate": 7.5298405728611645e-06,
+ "loss": 0.0526,
+ "step": 10610
+ },
+ {
+ "epoch": 23.71,
+ "learning_rate": 7.512330872007659e-06,
+ "loss": 0.051,
+ "step": 10620
+ },
+ {
+ "epoch": 23.73,
+ "learning_rate": 7.494829297452306e-06,
+ "loss": 0.0575,
+ "step": 10630
+ },
+ {
+ "epoch": 23.75,
+ "learning_rate": 7.4773359063663045e-06,
+ "loss": 0.0607,
+ "step": 10640
+ },
+ {
+ "epoch": 23.77,
+ "learning_rate": 7.459850755894108e-06,
+ "loss": 0.0552,
+ "step": 10650
+ },
+ {
+ "epoch": 23.79,
+ "learning_rate": 7.442373903153266e-06,
+ "loss": 0.0618,
+ "step": 10660
+ },
+ {
+ "epoch": 23.82,
+ "learning_rate": 7.424905405234209e-06,
+ "loss": 0.0834,
+ "step": 10670
+ },
+ {
+ "epoch": 23.84,
+ "learning_rate": 7.407445319200083e-06,
+ "loss": 0.0586,
+ "step": 10680
+ },
+ {
+ "epoch": 23.86,
+ "learning_rate": 7.38999370208656e-06,
+ "loss": 0.0574,
+ "step": 10690
+ },
+ {
+ "epoch": 23.88,
+ "learning_rate": 7.37255061090163e-06,
+ "loss": 0.0682,
+ "step": 10700
+ },
+ {
+ "epoch": 23.91,
+ "learning_rate": 7.355116102625451e-06,
+ "loss": 0.0704,
+ "step": 10710
+ },
+ {
+ "epoch": 23.93,
+ "learning_rate": 7.337690234210132e-06,
+ "loss": 0.0435,
+ "step": 10720
+ },
+ {
+ "epoch": 23.95,
+ "learning_rate": 7.320273062579568e-06,
+ "loss": 0.0501,
+ "step": 10730
+ },
+ {
+ "epoch": 23.97,
+ "learning_rate": 7.3028646446292295e-06,
+ "loss": 0.0602,
+ "step": 10740
+ },
+ {
+ "epoch": 24.0,
+ "learning_rate": 7.28546503722601e-06,
+ "loss": 0.0527,
+ "step": 10750
+ },
+ {
+ "epoch": 24.02,
+ "learning_rate": 7.268074297208008e-06,
+ "loss": 0.0412,
+ "step": 10760
+ },
+ {
+ "epoch": 24.04,
+ "learning_rate": 7.250692481384366e-06,
+ "loss": 0.0526,
+ "step": 10770
+ },
+ {
+ "epoch": 24.06,
+ "learning_rate": 7.233319646535067e-06,
+ "loss": 0.0406,
+ "step": 10780
+ },
+ {
+ "epoch": 24.08,
+ "learning_rate": 7.21595584941076e-06,
+ "loss": 0.049,
+ "step": 10790
+ },
+ {
+ "epoch": 24.11,
+ "learning_rate": 7.198601146732573e-06,
+ "loss": 0.0504,
+ "step": 10800
+ },
+ {
+ "epoch": 24.13,
+ "learning_rate": 7.181255595191919e-06,
+ "loss": 0.0586,
+ "step": 10810
+ },
+ {
+ "epoch": 24.15,
+ "learning_rate": 7.1639192514503265e-06,
+ "loss": 0.0658,
+ "step": 10820
+ },
+ {
+ "epoch": 24.17,
+ "learning_rate": 7.146592172139234e-06,
+ "loss": 0.0464,
+ "step": 10830
+ },
+ {
+ "epoch": 24.2,
+ "learning_rate": 7.129274413859832e-06,
+ "loss": 0.0603,
+ "step": 10840
+ },
+ {
+ "epoch": 24.22,
+ "learning_rate": 7.111966033182845e-06,
+ "loss": 0.0467,
+ "step": 10850
+ },
+ {
+ "epoch": 24.24,
+ "learning_rate": 7.094667086648381e-06,
+ "loss": 0.0486,
+ "step": 10860
+ },
+ {
+ "epoch": 24.26,
+ "learning_rate": 7.077377630765716e-06,
+ "loss": 0.0546,
+ "step": 10870
+ },
+ {
+ "epoch": 24.29,
+ "learning_rate": 7.060097722013137e-06,
+ "loss": 0.0587,
+ "step": 10880
+ },
+ {
+ "epoch": 24.31,
+ "learning_rate": 7.042827416837728e-06,
+ "loss": 0.0725,
+ "step": 10890
+ },
+ {
+ "epoch": 24.33,
+ "learning_rate": 7.025566771655219e-06,
+ "loss": 0.051,
+ "step": 10900
+ },
+ {
+ "epoch": 24.35,
+ "learning_rate": 7.00831584284977e-06,
+ "loss": 0.0508,
+ "step": 10910
+ },
+ {
+ "epoch": 24.38,
+ "learning_rate": 6.991074686773809e-06,
+ "loss": 0.0443,
+ "step": 10920
+ },
+ {
+ "epoch": 24.4,
+ "learning_rate": 6.973843359747845e-06,
+ "loss": 0.0453,
+ "step": 10930
+ },
+ {
+ "epoch": 24.42,
+ "learning_rate": 6.95662191806026e-06,
+ "loss": 0.0571,
+ "step": 10940
+ },
+ {
+ "epoch": 24.44,
+ "learning_rate": 6.939410417967168e-06,
+ "loss": 0.0497,
+ "step": 10950
+ },
+ {
+ "epoch": 24.46,
+ "learning_rate": 6.922208915692186e-06,
+ "loss": 0.0501,
+ "step": 10960
+ },
+ {
+ "epoch": 24.49,
+ "learning_rate": 6.905017467426291e-06,
+ "loss": 0.0545,
+ "step": 10970
+ },
+ {
+ "epoch": 24.51,
+ "learning_rate": 6.887836129327602e-06,
+ "loss": 0.0485,
+ "step": 10980
+ },
+ {
+ "epoch": 24.53,
+ "learning_rate": 6.870664957521225e-06,
+ "loss": 0.045,
+ "step": 10990
+ },
+ {
+ "epoch": 24.55,
+ "learning_rate": 6.85350400809904e-06,
+ "loss": 0.0552,
+ "step": 11000
+ }
+ ],
+ "max_steps": 17920,
+ "num_train_epochs": 40,
+ "total_flos": 6.438863431217971e+16,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/s3_en/training_args.bin b/s3_en/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..88dfd11ff232be624af99251704f91ec2c2ff481
--- /dev/null
+++ b/s3_en/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0a937952efa576dfebccb0461cf6205acf50feb0e554806aee932e1428712af
+size 6264
diff --git a/s3_en/zero_to_fp32.py b/s3_en/zero_to_fp32.py
new file mode 100644
index 0000000000000000000000000000000000000000..0e759146cadd92ddfefab3680146c2bd6a2b5c04
--- /dev/null
+++ b/s3_en/zero_to_fp32.py
@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+# python zero_to_fp32.py . output_dir/
+# or
+# python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+ buffers: dict()
+ param_shapes: dict()
+ shared_params: list
+ ds_version: int
+ frozen_param_shapes: dict()
+ frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+ return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+ '''
+ alist.sort(key=natural_keys) sorts in human order
+ http://nedbatchelder.com/blog/200712/human_sorting.html
+ (See Toothy's implementation in the comments)
+ '''
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+ if not os.path.isdir(checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+ # there should be only one file
+ if zero_stage <= 2:
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+ elif zero_stage == 3:
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+ if not os.path.exists(file):
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+ return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+ # XXX: need to test that this simple glob rule works for multi-node setup too
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+ if len(ckpt_files) == 0:
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+ return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+ zero_model_states = []
+ for file in files:
+ state_dict = torch.load(file, map_location=device, weights_only=False)
+
+ if BUFFER_NAMES not in state_dict:
+ raise ValueError(f"{file} is not a model state checkpoint")
+ buffer_names = state_dict[BUFFER_NAMES]
+ if debug:
+ print("Found buffers:", buffer_names)
+
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+ param_shapes = state_dict[PARAM_SHAPES]
+
+ # collect parameters that are included in param_shapes
+ param_names = []
+ for s in param_shapes:
+ for name in s.keys():
+ param_names.append(name)
+
+ # update with frozen parameters
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+ if frozen_param_shapes is not None:
+ if debug:
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+ param_names += list(frozen_param_shapes.keys())
+
+ # handle shared params
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+ ds_version = state_dict.get(DS_VERSION, None)
+
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+ z_model_state = zero_model_state(buffers=buffers,
+ param_shapes=param_shapes,
+ shared_params=shared_params,
+ ds_version=ds_version,
+ frozen_param_shapes=frozen_param_shapes,
+ frozen_param_fragments=frozen_param_fragments)
+ zero_model_states.append(z_model_state)
+
+ return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+ total_files = len(files)
+ state_dicts = []
+ for f in tqdm(files, desc='Loading checkpoint shards'):
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+ # and also handle the case where it was already removed by another helper script
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+ state_dicts.append(state_dict)
+
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
+ # use the max of the partition_count to get the dp world_size.
+
+ if type(world_size) is list:
+ world_size = max(world_size)
+
+ if world_size != total_files:
+ raise ValueError(
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+ )
+
+ # the groups are named differently in each stage
+ if zero_stage <= 2:
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+ elif zero_stage == 3:
+ fp32_groups_key = FP32_FLAT_GROUPS
+ else:
+ raise ValueError(f"unknown zero stage {zero_stage}")
+
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+ return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+ """
+ Returns fp32 state_dict reconstructed from ds checkpoint
+
+ Args:
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+ """
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+ optim_files = get_optim_files(ds_checkpoint_dir)
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+ model_files = get_model_state_files(ds_checkpoint_dir)
+
+ zero_model_states = parse_model_states(model_files)
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+ if zero_stage <= 2:
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+ elif zero_stage == 3:
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+ if debug:
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ state_dict[name] = frozen_param_fragments[name]
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+ attr = getattr(obj, fn, None)
+ return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+
+ # Reconstruction protocol:
+ #
+ # XXX: document this
+
+ if debug:
+ for i in range(world_size):
+ for j in range(len(fp32_flat_groups[0])):
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+ # XXX: memory usage doubles here (zero2)
+ num_param_groups = len(fp32_flat_groups[0])
+ merged_single_partition_of_fp32_groups = []
+ for i in range(num_param_groups):
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+ avail_numel = sum(
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+ if debug:
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+ # not asserting if there is a mismatch due to possible padding
+ print(f"Have {avail_numel} numels to process.")
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ total_numel = 0
+ total_params = 0
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+ offset = 0
+ avail_numel = full_single_fp32_vector.numel()
+ for name, shape in shapes.items():
+
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+ total_numel += unpartitioned_numel
+ total_params += 1
+
+ if debug:
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+ offset += unpartitioned_numel
+
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+ # live optimizer object, so we are checking that the numbers are within the right range
+ align_to = 2 * world_size
+
+ def zero2_align(x):
+ return align_to * math.ceil(x / align_to)
+
+ if debug:
+ print(f"original offset={offset}, avail_numel={avail_numel}")
+
+ offset = zero2_align(offset)
+ avail_numel = zero2_align(avail_numel)
+
+ if debug:
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+ remainder = unpartitioned_numel % world_size
+ padding_numel = (world_size - remainder) if remainder else 0
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+ return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+ return
+
+ if debug:
+ for i in range(world_size):
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+ wanted_params = len(frozen_param_shapes)
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+ print(f'Frozen params: Have {avail_numel} numels to process.')
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+ total_params = 0
+ total_numel = 0
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
+ total_params += 1
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+ """
+ A pseudo tensor that collects partitioned weights.
+ It is more memory efficient when there are multiple groups.
+ """
+
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+ self.flat_groups = flat_groups
+ self.flat_groups_offset = flat_groups_offset
+ self.offset = offset
+ self.partitioned_numel = partitioned_numel
+ self.shape = shape
+ self.dtype = self.flat_groups[0][0].dtype
+
+ def contiguous(self):
+ """
+ Merge partitioned weights from flat_groups into a single tensor.
+ """
+ end_idx = self.offset + self.partitioned_numel
+ world_size = len(self.flat_groups)
+ pad_flat_param_chunks = []
+
+ for rank_i in range(world_size):
+ # for each rank, we need to collect weights from related group/groups
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
+ start_group_id = None
+ end_group_id = None
+ for group_id in range(len(self.flat_groups_offset)):
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+ start_group_id = group_id
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+ end_group_id = group_id
+ break
+ # collect weights from related group/groups
+ for group_id in range(start_group_id, end_group_id + 1):
+ flat_tensor = flat_groups_at_rank_i[group_id]
+ start_offset = self.offset - self.flat_groups_offset[group_id]
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+ # collect weights from all ranks
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+ return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+ param_shapes = zero_model_states[0].param_shapes
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+ # param, re-consolidating each param, while dealing with padding if any
+
+ # merge list of dicts, preserving order
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+ if debug:
+ for i in range(world_size):
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+ wanted_params = len(param_shapes)
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+ # not asserting if there is a mismatch due to possible padding
+ avail_numel = fp32_flat_groups[0].numel() * world_size
+ print(f"Trainable params: Have {avail_numel} numels to process.")
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+ # params
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+ # out-of-core computing solution
+ offset = 0
+ total_numel = 0
+ total_params = 0
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+ unpartitioned_numel = shape.numel()
+ total_numel += unpartitioned_numel
+ total_params += 1
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+ if debug:
+ print(
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+ )
+
+ # memory efficient tensor
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+ state_dict[name] = tensor
+ offset += partitioned_numel
+
+ offset *= world_size
+
+ # Sanity check
+ if offset != avail_numel:
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+ exclude_frozen_parameters):
+ state_dict = OrderedDict()
+
+ # buffers
+ buffers = zero_model_states[0].buffers
+ state_dict.update(buffers)
+ if debug:
+ print(f"added {len(buffers)} buffers")
+
+ if not exclude_frozen_parameters:
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+ # recover shared parameters
+ for pair in zero_model_states[0].shared_params:
+ if pair[1] in state_dict:
+ state_dict[pair[0]] = state_dict[pair[1]]
+
+ return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+ """
+ Convert state_dict of GatheredTensor to torch tensor
+ """
+ torch_state_dict = {}
+ converted_tensors = {}
+ for name, tensor in state_dict.items():
+ tensor_id = id(tensor)
+ if tensor_id in converted_tensors: # shared tensors
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+ torch_state_dict[name] = shared_tensor
+ else:
+ converted_tensors[tensor_id] = name
+ if return_empty_tensor:
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+ else:
+ torch_state_dict[name] = tensor.contiguous()
+ return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag=None,
+ exclude_frozen_parameters=False,
+ lazy_mode=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+ via a model hub.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+ Returns:
+ - pytorch ``state_dict``
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ # do the training and checkpoint saving
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+ model = model.cpu() # move to cpu
+ model.load_state_dict(state_dict)
+ # submit to model hub or save the model to share with others
+
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
+ application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+ the checkpoint. Or you can load state_dict in lazy mode ::
+
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+ for name, lazy_tensor in state_dict.item():
+ tensor = lazy_tensor.contiguous() # to cpu
+ print(name, tensor)
+ # del tensor to release memory if it no longer in use
+ """
+ if tag is None:
+ latest_path = os.path.join(checkpoint_dir, 'latest')
+ if os.path.isfile(latest_path):
+ with open(latest_path, 'r') as fd:
+ tag = fd.read().strip()
+ else:
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+ if not os.path.isdir(ds_checkpoint_dir):
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+ if lazy_mode:
+ return state_dict
+ else:
+ return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+ output_dir,
+ max_shard_size="5GB",
+ safe_serialization=False,
+ tag=None,
+ exclude_frozen_parameters=False):
+ """
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+ Args:
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+ - ``exclude_frozen_parameters``: exclude frozen parameters
+ """
+
+ # Dependency pre-check
+ if safe_serialization:
+ try:
+ from safetensors.torch import save_file
+ except ImportError:
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
+ raise
+ if max_shard_size is not None:
+ try:
+ from huggingface_hub import split_torch_state_dict_into_shards
+ except ImportError:
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+ raise
+
+ # Convert zero checkpoint to state_dict
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+ tag,
+ exclude_frozen_parameters,
+ lazy_mode=True)
+
+ # Shard the model if it is too big.
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+ if max_shard_size is not None:
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+ # an memory-efficient approach for sharding
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+ filename_pattern=filename_pattern,
+ max_shard_size=max_shard_size)
+ else:
+ from collections import namedtuple
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+ state_dict_split = StateDictSplit(is_sharded=False,
+ filename_to_tensors={weights_name: list(state_dict.keys())})
+
+ # Save the model by shard
+ os.makedirs(output_dir, exist_ok=True)
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+ shard_state_dict = to_torch_tensor(shard_state_dict)
+ output_path = os.path.join(output_dir, shard_file)
+ if safe_serialization:
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+ else:
+ torch.save(shard_state_dict, output_path)
+ # release the memory of current shard
+ for tensor_name in list(shard_state_dict.keys()):
+ del state_dict[tensor_name]
+ del shard_state_dict[tensor_name]
+ del shard_state_dict
+ gc.collect()
+
+ # Save index if sharded
+ if state_dict_split.is_sharded:
+ index = {
+ "metadata": state_dict_split.metadata,
+ "weight_map": state_dict_split.tensor_to_filename,
+ }
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+ save_index_file = os.path.join(output_dir, save_index_file)
+ with open(save_index_file, "w", encoding="utf-8") as f:
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+ f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+ """
+ 1. Put the provided model to cpu
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+ 3. Load it into the provided model
+
+ Args:
+ - ``model``: the model object to update
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+ Returns:
+ - ``model`: modified model
+
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+ conveniently placed for you in the checkpoint folder.
+
+ A typical usage might be ::
+
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+ # submit to model hub or save the model to share with others
+
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+ """
+ logger.info(f"Extracting fp32 weights")
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+ logger.info(f"Overwriting model with fp32 weights")
+ model = model.cpu()
+ model.load_state_dict(state_dict, strict=False)
+
+ return model
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser()
+ parser.add_argument("checkpoint_dir",
+ type=str,
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+ parser.add_argument("output_dir",
+ type=str,
+ help="directory to the pytorch fp32 state_dict output files"
+ "(e.g. path/checkpoint-12-output/)")
+ parser.add_argument(
+ "--max_shard_size",
+ type=str,
+ default="5GB",
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+ "without CPU OOM issues.")
+ parser.add_argument(
+ "--safe_serialization",
+ default=False,
+ action='store_true',
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+ parser.add_argument("-t",
+ "--tag",
+ type=str,
+ default=None,
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+ args = parser.parse_args()
+
+ debug = args.debug
+
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+ args.output_dir,
+ max_shard_size=args.max_shard_size,
+ safe_serialization=args.safe_serialization,
+ tag=args.tag,
+ exclude_frozen_parameters=args.exclude_frozen_parameters)