New training version with all tokens labelled

README.md

@@ -1 +1,29 @@
-DistilBERT cased, fine-tuned for conll03 English.
+---
+language: en
+license: apache-2.0
+datasets:
+- conll2003
+---
+
+DistilBERT cased, fine-tuned for NER using the conll03 english dataset.
+
+## Versions
+
+Transformers version: 4.3.1
+Datasets version: 1.3.0
+
+## Training
+
+```
+$ run_ner.py \
+  --model_name_or_path distilbert-base-uncased \
+  --label_all_tokens True \
+  --return_entity_level_metrics True \
+  --dataset_name conll2003 \
+  --output_dir /tmp/distilbert-base-uncased-finetuned-conll2003 \
+  --do_train \
+  --do_eval
+```
+
+After training, we update the labels to match the NER specific labels from the
+dataset [conll2003](https://raw.githubusercontent.com/huggingface/datasets/1.3.0/datasets/conll2003/dataset_infos.json)

config.json

@@ -32,6 +32,7 @@
     "I-PER": 2,
     "O": 0
   },
+  "initializer_range": 0.02,
   "max_position_embeddings": 512,
   "model_type": "distilbert",
   "n_heads": 12,

eval_results_ner.txt

@@ -0,0 +1,24 @@
+eval_loss = 0.07699371129274368
+eval_LOC_precision = 0.9622327131352402
+eval_LOC_recall = 0.953232462173315
+eval_LOC_f1 = 0.9577114427860697
+eval_LOC_number = 3635
+eval_MISC_precision = 0.8739042481456507
+eval_MISC_recall = 0.8756756756756757
+eval_MISC_f1 = 0.8747890651366859
+eval_MISC_number = 1480
+eval_ORG_precision = 0.895021645021645
+eval_ORG_recall = 0.9182087342709104
+eval_ORG_f1 = 0.906466934599927
+eval_ORG_number = 2702
+eval_PER_precision = 0.9527067221891731
+eval_PER_recall = 0.9621507960348453
+eval_PER_f1 = 0.9574054700343745
+eval_PER_number = 3329
+eval_overall_precision = 0.9310928864325192
+eval_overall_recall = 0.9371074825049345
+eval_overall_f1 = 0.9340905025934537
+eval_overall_accuracy = 0.9834579384235004
+eval_runtime = 120.388
+eval_samples_per_second = 26.996
+epoch = 3.0

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84e7920a9300b9bc91267fd9842eb15211b9e683a5adf1d61fc966e0c8c9e928
+oid sha256:448297be01602ee7542a5c2ae05b578415acd5770854170d5e53fb282ed54dcf
 size 260832555

train_results.txt

@@ -0,0 +1,3 @@
+epoch = 3.0
+train_runtime = 6988.1143
+train_samples_per_second = 0.754

trainer_state.json

@@ -0,0 +1,83 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "global_step": 5268,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.525436598329537e-05,
+      "loss": 0.255,
+      "step": 500
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.050873196659074e-05,
+      "loss": 0.1219,
+      "step": 1000
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 3.5763097949886106e-05,
+      "loss": 0.0916,
+      "step": 1500
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 3.1017463933181475e-05,
+      "loss": 0.075,
+      "step": 2000
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 2.6271829916476843e-05,
+      "loss": 0.0497,
+      "step": 2500
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 2.152619589977221e-05,
+      "loss": 0.0495,
+      "step": 3000
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.678056188306758e-05,
+      "loss": 0.0417,
+      "step": 3500
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 1.2034927866362947e-05,
+      "loss": 0.0186,
+      "step": 4000
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 7.289293849658315e-06,
+      "loss": 0.023,
+      "step": 4500
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 2.5436598329536827e-06,
+      "loss": 0.0193,
+      "step": 5000
+    },
+    {
+      "epoch": 3.0,
+      "step": 5268,
+      "total_flos": 705655184326236,
+      "train_runtime": 6988.1143,
+      "train_samples_per_second": 0.754
+    }
+  ],
+  "max_steps": 5268,
+  "num_train_epochs": 3,
+  "total_flos": 705655184326236,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c30c38f69e6dfc85b6437f712c9c9e6530957b27791423275f6d42b0e4bbbce1
+size 2159