new attempt

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "./",
+  "_name_or_path": "./config.json",
   "architectures": [
     "XLMRobertaForMaskedLM"
   ],
@@ -21,7 +21,7 @@
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
   "torch_dtype": "float32",
-  "transformers_version": "4.23.1",
+  "transformers_version": "4.24.0",
   "type_vocab_size": 1,
   "use_cache": true,
   "vocab_size": 250002

events.out.tfevents.1672819976.t1v-n-0853dee6-w-3.291836.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4189a298dc3ab9be3250fd74db70fac076b8e11d308c2afba23e19d047a123a8
+size 22598

events.out.tfevents.1672820882.t1v-n-0853dee6-w-3.322000.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f7294f94d22c80b866f0559f6606ac68f6f6f386918848c1f7f4ba68e161340
+size 7637

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a04d8032d9ff18b1b727c98ee3bea82de1d908107d257d94486dba8650f5680
+size 1113187999

run.sh

@@ -1,6 +1,5 @@
 
 python run_mlm_flax_stream.py \
-    --output_dir="./" \
     --hub_model_id="NbAiLab/nb-roberta-base-scandi" \
     --model_name_or_path="xlm-roberta-base" \
     --config_name="./config.json" \
@@ -12,6 +11,7 @@ python run_mlm_flax_stream.py \
     --per_device_eval_batch_size="16" \
     --learning_rate="4e-4" \
     --warmup_steps="1000" \
+    --use_auth_token \
     --overwrite_output_dir \
     --num_train_steps="10000" \
     --adam_beta1="0.9" \
@@ -19,5 +19,6 @@ python run_mlm_flax_stream.py \
     --logging_steps="50" \
     --save_steps="50" \
     --eval_steps="50" \
+    push_to_--output_dir="./" \
     --dtype="bfloat16" \
     --push_to_hub

run_mlm_flax_stream.py

@@ -655,13 +655,18 @@ if __name__ == "__main__":
                     )
                     print(f"Saving weights and logs of step {step+1}. \nThe result is saved to {training_args.output_folder} by worker {jax.process_index()}.")
                 except:
-
+                    breakpoint()
                     model.save_pretrained(
                         training_args.output_dir,
                         params=params
                     )
                     print("Problems pushing this to the hub. The bug should be fixed.")
-
+            else:
+                model.save_pretrained(
+                    training_args.output_dir,
+                    params=params
+                )
+            
         # update tqdm bar
         steps.update(1)