Initial release: Fine-tuned NLLB for Dzongkha-English

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 5.0,
-    "total_flos": 2708880752640000.0,
-    "train_loss": 1.667306510925293,
-    "train_runtime": 2329.7908,
-    "train_samples_per_second": 4.292,
-    "train_steps_per_second": 0.537
+    "total_flos": 1.35444037632e+16,
+    "train_loss": 0.05036833358764648,
+    "train_runtime": 6278.8779,
+    "train_samples_per_second": 7.963,
+    "train_steps_per_second": 0.995
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 5.0,
-    "total_flos": 2708880752640000.0,
-    "train_loss": 1.667306510925293,
-    "train_runtime": 2329.7908,
-    "train_samples_per_second": 4.292,
-    "train_steps_per_second": 0.537
+    "total_flos": 1.35444037632e+16,
+    "train_loss": 0.05036833358764648,
+    "train_runtime": 6278.8779,
+    "train_samples_per_second": 7.963,
+    "train_steps_per_second": 0.995
 }

trainer_state.json

@@ -1,85 +1,155 @@
 {
-  "best_global_step": 1250,
-  "best_metric": 63.12399055783562,
-  "best_model_checkpoint": "nllb-600m-dz-en-checkpoints/checkpoint-1250",
+  "best_global_step": 6250,
+  "best_metric": 59.5126771383472,
+  "best_model_checkpoint": "nllb-200-600M-dzo-eng-checkpoints/checkpoint-6250",
   "epoch": 5.0,
   "eval_steps": 500,
-  "global_step": 1250,
+  "global_step": 6250,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.19286614656448364,
+      "learning_rate": 2.76048e-05,
+      "loss": 0.0794,
+      "step": 500
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.27002865076065063,
+      "learning_rate": 2.52048e-05,
+      "loss": 0.0765,
+      "step": 1000
+    },
     {
       "epoch": 1.0,
-      "eval_bleu": 58.0755267273345,
-      "eval_loss": 3.2226617336273193,
-      "eval_runtime": 88.1796,
-      "eval_samples_per_second": 5.67,
-      "eval_steps_per_second": 1.418,
-      "step": 250
+      "eval_bleu": 58.03728575479501,
+      "eval_loss": 0.07460231333971024,
+      "eval_runtime": 167.4219,
+      "eval_samples_per_second": 5.973,
+      "eval_steps_per_second": 1.493,
+      "step": 1250
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 0.3723973035812378,
+      "learning_rate": 2.28048e-05,
+      "loss": 0.0644,
+      "step": 1500
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.20845282077789307,
+      "learning_rate": 2.04048e-05,
+      "loss": 0.0576,
+      "step": 2000
     },
     {
       "epoch": 2.0,
-      "grad_norm": 5.708385944366455,
-      "learning_rate": 1.8096e-05,
-      "loss": 4.0188,
-      "step": 500
+      "grad_norm": 0.3116896450519562,
+      "learning_rate": 1.80048e-05,
+      "loss": 0.0576,
+      "step": 2500
     },
     {
       "epoch": 2.0,
-      "eval_bleu": 59.852719093079365,
-      "eval_loss": 0.36037033796310425,
-      "eval_runtime": 100.2346,
-      "eval_samples_per_second": 4.988,
-      "eval_steps_per_second": 1.247,
-      "step": 500
+      "eval_bleu": 58.57463505915474,
+      "eval_loss": 0.07282423228025436,
+      "eval_runtime": 168.9336,
+      "eval_samples_per_second": 5.919,
+      "eval_steps_per_second": 1.48,
+      "step": 2500
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 0.2609412372112274,
+      "learning_rate": 1.56048e-05,
+      "loss": 0.0452,
+      "step": 3000
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 0.16577185690402985,
+      "learning_rate": 1.3204800000000001e-05,
+      "loss": 0.0465,
+      "step": 3500
     },
     {
       "epoch": 3.0,
-      "eval_bleu": 62.2628950594224,
-      "eval_loss": 0.10864810645580292,
-      "eval_runtime": 100.1968,
-      "eval_samples_per_second": 4.99,
-      "eval_steps_per_second": 1.248,
-      "step": 750
+      "eval_bleu": 59.30988278861181,
+      "eval_loss": 0.07350268214941025,
+      "eval_runtime": 171.8963,
+      "eval_samples_per_second": 5.817,
+      "eval_steps_per_second": 1.454,
+      "step": 3750
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.1930246204137802,
+      "learning_rate": 1.08048e-05,
+      "loss": 0.0426,
+      "step": 4000
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 0.3314656913280487,
+      "learning_rate": 8.404800000000001e-06,
+      "loss": 0.0381,
+      "step": 4500
     },
     {
       "epoch": 4.0,
-      "grad_norm": 0.4872799217700958,
-      "learning_rate": 6.096e-06,
-      "loss": 0.1201,
-      "step": 1000
+      "grad_norm": 0.26522204279899597,
+      "learning_rate": 6.0048000000000005e-06,
+      "loss": 0.0381,
+      "step": 5000
     },
     {
       "epoch": 4.0,
-      "eval_bleu": 63.06198866502594,
-      "eval_loss": 0.10180553793907166,
-      "eval_runtime": 100.7637,
-      "eval_samples_per_second": 4.962,
-      "eval_steps_per_second": 1.241,
-      "step": 1000
+      "eval_bleu": 59.249316493031536,
+      "eval_loss": 0.07584889233112335,
+      "eval_runtime": 168.0339,
+      "eval_samples_per_second": 5.951,
+      "eval_steps_per_second": 1.488,
+      "step": 5000
+    },
+    {
+      "epoch": 4.4,
+      "grad_norm": 0.17317089438438416,
+      "learning_rate": 3.6048e-06,
+      "loss": 0.0337,
+      "step": 5500
+    },
+    {
+      "epoch": 4.8,
+      "grad_norm": 0.32834669947624207,
+      "learning_rate": 1.2048e-06,
+      "loss": 0.033,
+      "step": 6000
     },
     {
       "epoch": 5.0,
-      "eval_bleu": 63.12399055783562,
-      "eval_loss": 0.10181548446416855,
-      "eval_runtime": 111.015,
-      "eval_samples_per_second": 4.504,
-      "eval_steps_per_second": 1.126,
-      "step": 1250
+      "eval_bleu": 59.5126771383472,
+      "eval_loss": 0.07744310051202774,
+      "eval_runtime": 168.6113,
+      "eval_samples_per_second": 5.931,
+      "eval_steps_per_second": 1.483,
+      "step": 6250
     },
     {
       "epoch": 5.0,
-      "step": 1250,
-      "total_flos": 2708880752640000.0,
-      "train_loss": 1.667306510925293,
-      "train_runtime": 2329.7908,
-      "train_samples_per_second": 4.292,
-      "train_steps_per_second": 0.537
+      "step": 6250,
+      "total_flos": 1.35444037632e+16,
+      "train_loss": 0.05036833358764648,
+      "train_runtime": 6278.8779,
+      "train_samples_per_second": 7.963,
+      "train_steps_per_second": 0.995
     }
   ],
   "logging_steps": 500,
-  "max_steps": 1250,
+  "max_steps": 6250,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 5,
   "save_steps": 500,
@@ -95,7 +165,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 2708880752640000.0,
+  "total_flos": 1.35444037632e+16,
   "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null