add readme

pii/ner/README.md

@@ -0,0 +1,14 @@
+# Fine-tuning Bigcode-Encoder on an NER task  for PII detection
+
+To run the training on all the dataset `bigcode/pii-full-ds`, use the following command: 
+```bash
+python -m torch.distributed.launch \
+        --nproc_per_node number_of_gpus train.py \
+        --dataset_name bigcode/pii-full-ds \
+        --debug \
+        --learning_rate 2e-5 \
+        --train_batch_size 8 \
+        --bf16 \
+        --add_not_curated
+```
+Note that we use a global batch size of 64 (8*8 GPUs). To use only curated dataset remove the flag `--add_not_curated`.