KucLab commited on
Commit
dd66da7
·
verified ·
1 Parent(s): 38eb83c

Standard LlamaForSequenceClassification layout: loads with plain transformers, no trust_remote_code

Browse files
Files changed (1) hide show
  1. README.md +22 -0
README.md CHANGED
@@ -177,6 +177,28 @@ Measured on an AMD Radeon RX 9060 XT (ROCm 7.1, torch 2.13.0):
177
 
178
  Training the whole model from scratch takes about 3.5 minutes on the same GPU.
179
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
180
  ## Usage
181
 
182
  Plain `transformers`, nothing else to install — `from_pretrained` downloads
 
177
 
178
  Training the whole model from scratch takes about 3.5 minutes on the same GPU.
179
 
180
+ ## Comparison with other models
181
+
182
+ Benchmarked against public safety classifiers on the **same held-out test split**
183
+ (6,496 rows, threshold 0.5, AMD Radeon RX 9060 XT):
184
+
185
+ | Model | Params | Checkpoint | F1 | Latency | Throughput (b=32) | Peak VRAM |
186
+ | --- | --- | --- | --- | --- | --- | --- |
187
+ | **Censer.1 (ours)** | 1.1M | 4.3 MB | **0.835** | 2.17 ms | 1,454 texts/s | 102 MB |
188
+ | unitary/unbiased-toxic-roberta | 124.7M | 498.7 MB | 0.741 | 15.54 ms | 62 texts/s | 928 MB |
189
+ | unitary/toxic-bert | 109.5M | 876.0 MB | 0.682 | 14.85 ms | 65 texts/s | 853 MB |
190
+ | martin-ha/toxic-comment-model | 67.0M | 267.9 MB | 0.567 | 7.98 ms | 124 texts/s | 691 MB |
191
+ | Hate-speech-CNERG/bert-base-uncased-hatexplain | 109.5M | 438.0 MB | 0.545 | 15.04 ms | 66 texts/s | 722 MB |
192
+ | Hate-speech-CNERG/dehatebert-mono-english | 167.4M | 669.5 MB | 0.253 | 15.36 ms | 64 texts/s | 960 MB |
193
+ | keyword baseline | — | — | 0.094 | 0.03 ms | 29,689 texts/s | — |
194
+
195
+ ![Comparison overview](figs/comparison_overview.png)
196
+ ![F1 per megabyte](figs/f1_vs_size.png)
197
+ ![Metric bars](figs/metrics_bars.png)
198
+
199
+ Censer.1 is ~60-150x smaller than the encoder baselines, 3-7x faster per text and
200
+ uses ~7-9x less VRAM — with the highest F1 on this split.
201
+
202
  ## Usage
203
 
204
  Plain `transformers`, nothing else to install — `from_pretrained` downloads