End of training

all_results.json

@@ -1,12 +1,12 @@
 {
-    "epoch": 22.22,
-    "eval_loss": 1.2309296131134033,
-    "eval_runtime": 450.2608,
-    "eval_samples_per_second": 1.137,
-    "eval_steps_per_second": 0.142,
+    "epoch": 100.0,
+    "eval_loss": 1.2280755043029785,
+    "eval_runtime": 477.4146,
+    "eval_samples_per_second": 1.072,
+    "eval_steps_per_second": 0.067,
     "eval_wer": 56.651029055690074,
-    "train_loss": 0.001749273296445608,
-    "train_runtime": 576.5843,
-    "train_samples_per_second": 16.65,
-    "train_steps_per_second": 1.041
+    "train_loss": 0.0004244298700775419,
+    "train_runtime": 836.7871,
+    "train_samples_per_second": 53.538,
+    "train_steps_per_second": 0.837
 }

eval_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 22.22,
-    "eval_loss": 1.2309296131134033,
-    "eval_runtime": 450.2608,
-    "eval_samples_per_second": 1.137,
-    "eval_steps_per_second": 0.142,
+    "epoch": 100.0,
+    "eval_loss": 1.2280755043029785,
+    "eval_runtime": 477.4146,
+    "eval_samples_per_second": 1.072,
+    "eval_steps_per_second": 0.067,
     "eval_wer": 56.651029055690074
 }

run.sh

@@ -6,13 +6,13 @@ python run_speech_recognition_seq2seq_streaming.py \
       --train_split_name="train+validation" \
       --eval_split_name="test" \
       --model_index_name="Whisper Small Pashto" \
-      --max_steps="700" \
+      --max_steps="800" \
       --output_dir="./" \
-      --per_device_train_batch_size="32" \
-      --per_device_eval_batch_size="16" \
-      --gradient_accumulation_steps="2" \
+      --per_device_train_batch_size="64" \
+      --per_device_eval_batch_size="32" \
+      --gradient_accumulation_steps="1" \
       --logging_steps="10" \
-      --learning_rate="1e-5" \
+      --learning_rate="3e-7" \
       --warmup_steps="500" \
       --evaluation_strategy="steps" \
       --eval_steps="100" \

runs/Dec16_18-23-51_129-146-104-29/events.out.tfevents.1671216516.129-146-104-29.139583.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d3c061057271338c2c2400cf2b6ac3a6a129116f28cbfaffde6d31165819119
+size 358

train_results.json

@@ -1,7 +1,7 @@
 {
-    "epoch": 22.22,
-    "train_loss": 0.001749273296445608,
-    "train_runtime": 576.5843,
-    "train_samples_per_second": 16.65,
-    "train_steps_per_second": 1.041
+    "epoch": 100.0,
+    "train_loss": 0.0004244298700775419,
+    "train_runtime": 836.7871,
+    "train_samples_per_second": 53.538,
+    "train_steps_per_second": 0.837
 }

trainer_state.json

@@ -1,8 +1,8 @@
 {
   "best_metric": 56.651029055690074,
   "best_model_checkpoint": "./checkpoint-600",
-  "epoch": 22.22222222222222,
-  "global_step": 600,
+  "epoch": 100.0,
+  "global_step": 700,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -314,18 +314,87 @@
       "step": 600
     },
     {
-      "epoch": 22.22,
-      "step": 600,
-      "total_flos": 2.76407096426496e+18,
-      "train_loss": 0.001749273296445608,
-      "train_runtime": 576.5843,
-      "train_samples_per_second": 16.65,
-      "train_steps_per_second": 1.041
+      "epoch": 87.14,
+      "learning_rate": 4.7e-06,
+      "loss": 0.0038,
+      "step": 610
+    },
+    {
+      "epoch": 88.57,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 0.0033,
+      "step": 620
+    },
+    {
+      "epoch": 90.0,
+      "learning_rate": 3.7e-06,
+      "loss": 0.0033,
+      "step": 630
+    },
+    {
+      "epoch": 91.43,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 0.0031,
+      "step": 640
+    },
+    {
+      "epoch": 92.86,
+      "learning_rate": 2.7000000000000004e-06,
+      "loss": 0.0029,
+      "step": 650
+    },
+    {
+      "epoch": 94.29,
+      "learning_rate": 2.2e-06,
+      "loss": 0.0028,
+      "step": 660
+    },
+    {
+      "epoch": 95.71,
+      "learning_rate": 1.7000000000000002e-06,
+      "loss": 0.0027,
+      "step": 670
+    },
+    {
+      "epoch": 97.14,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 0.0026,
+      "step": 680
+    },
+    {
+      "epoch": 98.57,
+      "learning_rate": 7.000000000000001e-07,
+      "loss": 0.0026,
+      "step": 690
+    },
+    {
+      "epoch": 100.0,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 0.0026,
+      "step": 700
+    },
+    {
+      "epoch": 100.0,
+      "eval_loss": 1.2580721378326416,
+      "eval_runtime": 471.9052,
+      "eval_samples_per_second": 1.085,
+      "eval_steps_per_second": 0.068,
+      "eval_wer": 56.847760290556906,
+      "step": 700
+    },
+    {
+      "epoch": 100.0,
+      "step": 700,
+      "total_flos": 4.53742824923136e+18,
+      "train_loss": 0.0004244298700775419,
+      "train_runtime": 836.7871,
+      "train_samples_per_second": 53.538,
+      "train_steps_per_second": 0.837
     }
   ],
-  "max_steps": 600,
-  "num_train_epochs": 23,
-  "total_flos": 2.76407096426496e+18,
+  "max_steps": 700,
+  "num_train_epochs": 100,
+  "total_flos": 4.53742824923136e+18,
   "trial_name": null,
   "trial_params": null
 }