Upload README.md with huggingface_hub

README.md

@@ -12,6 +12,14 @@ metrics:
   - accuracy
   - f1
 pipeline_tag: text-classification
+language:
+  - en
+library_name: transformers
+widget:
+  - text: "We introduce a novel attention mechanism that reduces the quadratic complexity of transformers to linear time while preserving accuracy on long-context language modeling benchmarks."
+    example_title: ML paper abstract
+  - text: "We present a distributed consensus protocol that tolerates Byzantine faults with optimal message complexity in partially synchronous networks."
+    example_title: Systems paper abstract
 ---
 
 # Academic Paper Classifier
@@ -36,16 +44,21 @@ acceptance or funding allocation.
 
 ## Labels
 
-| Id | Label    | Description                       |
-|----|----------|-----------------------------------|
-| 0  | cs.AI    | Artificial Intelligence           |
-| 1  | cs.CL    | Computation and Language (NLP)    |
-| 2  | cs.CV    | Computer Vision                   |
-| 3  | cs.LG    | Machine Learning                  |
-| 4  | cs.NE    | Neural and Evolutionary Computing |
-| 5  | cs.RO    | Robotics                          |
-| 6  | math.ST  | Statistics Theory                 |
-| 7  | stat.ML  | Machine Learning (Statistics)     |
+The label space comes from [ccdv/arxiv-classification](https://huggingface.co/datasets/ccdv/arxiv-classification) (11 classes):
+
+| Id | Label   | Description                        |
+|----|---------|------------------------------------|
+| 0  | math.AC | Commutative Algebra                |
+| 1  | cs.CV   | Computer Vision                    |
+| 2  | cs.AI   | Artificial Intelligence            |
+| 3  | cs.SY   | Systems and Control                |
+| 4  | math.GR | Group Theory                       |
+| 5  | cs.CE   | Computational Engineering          |
+| 6  | cs.PL   | Programming Languages              |
+| 7  | cs.IT   | Information Theory                 |
+| 8  | cs.DS   | Data Structures and Algorithms     |
+| 9  | cs.NE   | Neural and Evolutionary Computing  |
+| 10 | math.ST | Statistics Theory                  |
 
 ## Training Procedure
 
@@ -77,8 +90,19 @@ language-understanding performance.
 
 ### Metrics
 
-The model is evaluated on accuracy, weighted F1, weighted precision, and
-weighted recall. The best checkpoint is selected by weighted F1.
+Measured on 1,500 held-out validation papers (run of 2026-07-12; 8,000
+training samples, 3 epochs, max_length 384, transformers 5.13):
+
+| Metric | Score |
+|--------|-------|
+| Accuracy | 0.829 |
+| F1 (weighted) | 0.827 |
+| Precision (weighted) | 0.828 |
+| Recall (weighted) | 0.829 |
+
+The best checkpoint is selected by weighted F1. Note the model saw the
+first 8,000 of ~28k training documents truncated to 384 tokens; training
+on the full corpus at 512 tokens should improve these numbers.
 
 ## How to Use
 
@@ -89,7 +113,7 @@ from transformers import pipeline
 
 classifier = pipeline(
     "text-classification",
-    model="gr8monk3ys/paper-classifier-model",
+    model="gr8monk3ys/paper-classifier",
 )
 
 abstract = (
@@ -100,14 +124,14 @@ abstract = (
 
 result = classifier(abstract)
 print(result)
-# [{'label': 'cs.CL', 'score': 0.95}]
+# [{'label': 'cs.NE', 'score': 0.61}]
 ```
 
 ### With the included inference script
 
 ```bash
 python inference.py \
-    --model_path gr8monk3ys/paper-classifier-model \
+    --model_path gr8monk3ys/paper-classifier \
     --abstract "We propose a convolutional neural network for image recognition..."
 ```
 
@@ -141,6 +165,6 @@ If you use this model in your research, please cite:
     title  = {Academic Paper Classifier},
     author = {Lorenzo Scaturchio},
     year   = {2025},
-    url    = {https://huggingface.co/gr8monk3ys/paper-classifier-model}
+    url    = {https://huggingface.co/gr8monk3ys/paper-classifier}
 }
 ```