Training in progress, step 1000

config.json

@@ -34,7 +34,7 @@
   "num_mel_bins": 80,
   "pad_token_id": 50257,
   "scale_embedding": false,
-  "torch_dtype": "float32",
+  "torch_dtype": "float16",
   "transformers_version": "4.26.0.dev0",
   "use_cache": false,
   "vocab_size": 51865

ds_config.json

@@ -0,0 +1,48 @@
+{
+    "fp16": {
+        "enabled": "auto",
+        "loss_scale": 0,
+        "loss_scale_window": 1000,
+        "initial_scale_power": 16,
+        "hysteresis": 2,
+        "min_loss_scale": 1
+    },
+
+    "optimizer": {
+        "type": "AdamW",
+        "params": {
+            "lr": "auto",
+            "betas": "auto",
+            "eps": "auto",
+            "weight_decay": "auto"
+        }
+    },
+
+    "scheduler": {
+        "type": "WarmupLR",
+        "params": {
+            "warmup_min_lr": "auto",
+            "warmup_max_lr": "auto",
+            "warmup_num_steps": "auto"
+        }
+    },
+
+    "zero_optimization": {
+        "stage": 2,
+        "offload_optimizer": {
+            "device": "cpu",
+            "pin_memory": true
+        },
+        "allgather_partitions": true,
+        "allgather_bucket_size": 2e8,
+        "overlap_comm": true,
+        "reduce_scatter": true,
+        "reduce_bucket_size": 2e8,
+        "contiguous_gradients": true
+    },
+
+    "gradient_accumulation_steps": "auto",
+    "gradient_clipping": "auto",
+    "train_batch_size": "auto",
+    "train_micro_batch_size_per_gpu": "auto"
+}

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7df34743dc068a9be2cc27fae73d17c5f0d29f2998d120215b05ec7b4958778
-size 967102601
+oid sha256:fb8c44a25c6536055d608d7e16c64d9906e9db3c32a37721c88e1c8b5323d953
+size 483536061

run.sh

@@ -1,4 +1,5 @@
-python run_speech_recognition_seq2seq_streaming.py \
+deepspeed run_speech_recognition_seq2seq_streaming.py \
+	--deepspeed="ds_config.json" \
 	--model_name_or_path="openai/whisper-small" \
 	--dataset_name="facebook/voxpopuli" \
 	--dataset_config_name="hr" \
@@ -33,5 +34,6 @@ python run_speech_recognition_seq2seq_streaming.py \
 	--do_eval \
 	--predict_with_generate \
 	--do_normalize_eval \
+	--streaming \
 	--use_auth_token \
 	--push_to_hub

runs/Dec14_16-16-20_129-146-123-136/1671034661.2950988/events.out.tfevents.1671034661.129-146-123-136.70213.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae91cb64ba0714a7cb4345a2f4de8e94d34b7ae097a8889448f2037939c28256
+size 5881

runs/Dec14_16-16-20_129-146-123-136/events.out.tfevents.1671034661.129-146-123-136.70213.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4122ec6beea05038077efd7c087c5cd203f115fd6c9887a582c2e82be1105867
+size 10862

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d094671071c14483f70e12ca1ccf05d2ccca6c31f93d08886692391dd0706384
-size 3515
+oid sha256:4661efef1c7d1829413d2a909012154ff96d9155a348c80c9a204ac96b0688df
+size 4667