first

run.sh

@@ -10,7 +10,7 @@ deepspeed --master_port 29501 run_speech_recognition_seq2seq_streaming.py \
 	--output_dir="./" \
 	--per_device_train_batch_size="128" \
 	--per_device_eval_batch_size="32" \
-	--gradient_accumulation_steps="1" \
+	--gradient_accumulation_steps="2" \
 	--logging_steps="25" \
 	--learning_rate="3e-6" \
 	--warmup_steps="1000" \

runs/Dec15_13-08-24_dante/1671106135.038802/events.out.tfevents.1671106135.dante.2061867.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c254b3aad4e61959e99616758fbf74354491cdd1c3b63210f12180739dfc0a00
+size 5888

runs/Dec15_13-08-24_dante/events.out.tfevents.1671106135.dante.2061867.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:898520556ecaae20e5603fa9f306d4da446d41d9d105140991107014217debb7
+size 4291