Upload README.md with huggingface_hub

README.md

@@ -1,87 +1,44 @@
 ---
 license: apache-2.0
-tags:
-  - generated_from_trainer
-  - token-classification
-  - ner
-  - nlp
-datasets:
-  - conll2003
-language:
-  - en
-pipeline_tag: token-classification
+language: en
 library_name: transformers
+pipeline_tag: token-classification
 base_model: bert-base-uncased
-model-index:
-  - name: harpertokenNER
-    results:
-      - task:
-          type: token-classification
-          name: Named Entity Recognition (NER)
-        dataset:
-          name: CoNLL-2003
-          type: conll2003
-        metrics:
-          - name: Validation Loss
-            type: loss
-            value: 0.0474
+datasets:
+- conll2003
+tags:
+- bert
+- conll2003
+- ner
+- token-classification
 widget:
-  - text: "Apple is buying a U.K. startup for $1 billion"
+- text: "Apple is buying a U.K. startup for $1 billion"
 ---
 
-> **Renamed:** previously `harpertoken/harpertokenNER`. The old ID still redirects.
-
-# harpertokenNER
-
-This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on the [CoNLL-2003](https://huggingface.co/datasets/eriktks/conll2003) dataset. It achieves a validation loss of **0.0474** on the evaluation set.
-
-## Model Description
-
-This is a token classification model fine-tuned for **Named Entity Recognition (NER)** on the CoNLL-2003 dataset, built on the `bert-base-uncased` architecture. It identifies entities like people, organizations, and locations in text. Optimized for CPU use. Uploaded by [harpertoken](https://huggingface.co/harpertoken).
-
-## Intended Uses & Limitations
-
-### Intended Uses
-- Extracting named entities from unstructured text (e.g., news articles, reports)
-- Powering NLP pipelines on CPU-based systems
-- Research or lightweight production use
-
-### Limitations
-- Trained on English text from CoNLL-2003, so it may not generalize well to other languages or domains
-- Uses `bert-base-uncased` tokenization (lowercase-only), potentially missing case-sensitive nuances
-- Optimized for NER; additional tuning needed for other token-classification tasks
+> Previously published as `harpertoken/harpertokenNER`. The old identifier still redirects here.
 
-## Training and Evaluation Data
+# name
 
-The model was trained and evaluated on the *CoNLL-2003 dataset*, a standard NER benchmark. It features annotated English news articles with entities like persons, organizations, and locations, split into training, validation, and test sets. Metrics here reflect the evaluation subset.
+A `bert-base-uncased` encoder with a token-classification head, fine-tuned on CoNLL-2003 for named entity recognition. It assigns a label to each token and reports the spans it considers to be people, organisations, locations and miscellaneous entities. Training ran for three epochs at a learning rate of 2e-5 with batch size eight and a linear schedule with 500 warmup steps, reaching a validation loss of 0.0474.
 
-## Training Procedure
+One thing is worth knowing before you use it. The label map in `config.json` is generic — `LABEL_0` through `LABEL_8` — rather than the CoNLL entity names. The nine classes are the CoNLL scheme, and the conventional ordering is `O`, `B-PER`, `I-PER`, `B-ORG`, `I-ORG`, `B-LOC`, `I-LOC`, `B-MISC`, `I-MISC`, but that mapping is not recorded in the repository, so predictions come back as `LABEL_3` rather than `B-ORG`. Anyone relying on `pipeline` output will need to supply their own mapping, and until one is written down the ordering should be treated as unverified.
 
-### Training Hyperparameters
+The repository also carries `pytorch_model.bin` alongside `model.safetensors` with the same parameters, a `training_args.bin`, and a `runs/` directory of TensorBoard event files. None of these are read at inference time.
 
-The following hyperparameters were used during training:
-- **learning_rate**: 2e-05
-- **train_batch_size**: 8
-- **eval_batch_size**: 8
-- **seed**: 42
-- **optimizer**: Adam with betas=(0.9,0.999) and epsilon=1e-08
-- **lr_scheduler_type**: linear
-- **lr_scheduler_warmup_steps**: 500
-- **num_epochs**: 3
+## Usage
 
-### Training Results
+```python
+from transformers import pipeline
 
-| Training Loss | Epoch | Step | Validation Loss |
-|:-------------:|:-----:|:----:|:---------------:|
-| 0.048         | 1.0   | 1756 | 0.0531          |
-| 0.0251        | 2.0   | 3512 | 0.0473          |
-| 0.016         | 3.0   | 5268 | 0.0474          |
+ner = pipeline("token-classification", model="harpertoken/name")
+for entity in ner("Apple is buying a U.K. startup for $1 billion"):
+    print(entity["word"], entity["entity"])
+```
 
-### Framework Versions
+## Limitations
 
-- **Transformers**: 4.28.1
-- **PyTorch**: 2.0.1
-- **Datasets**: 1.18.3
-- **Tokenizers**: 0.13.3
+CoNLL-2003 consists of English newswire from 2003, and entity boundaries in that data reflect contemporary editorial conventions. The model should not be expected to transfer to other languages, to literary or conversational prose, or to domains with unfamiliar entity types, and its results will depend heavily on the unresolved label mapping above. Case is discarded by the uncased vocabulary, which costs it the ability to distinguish `Jordan` the person from `jordan` the country.
 
+## Attribution
 
+BERT follows Devlin et al., *BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding* (2019). CoNLL-2003 is described in CoNLL-2003 shared task papers from Nanyeong Bang et al. and Tjong Kim Sang et al.