RowRed commited on
Commit
74c0320
·
verified ·
1 Parent(s): f285c7f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -2
README.md CHANGED
@@ -25,7 +25,6 @@ It classifies prompts into **3 difficulty tiers** with a single classification h
25
 
26
  - **Model type:** Text Classification (3‑tier, single head)
27
  - **Language:** English
28
- - **Backbone:** microsoft/deberta-v3-xsmall
29
  - **License:** Apache‑2.0
30
  - **Finetuned from model:** microsoft/deberta-v3-xsmall
31
  - **Training data:** `RowRed/prompts-24000-en` (L0 downsampled to 2,000, L1/L2/L3 kept fully)
@@ -88,7 +87,7 @@ This model is a research prototype.
88
  - **Max sequence length:** 256 tokens
89
  - **Epochs:** 12 max (Early stopping patience = 4 on F1-macro; stopped at epoch 7, best checkpoint at epoch 3)
90
  - **Class balancing:** WeightedRandomSampler only; class weights in loss removed to avoid double-counting
91
- - **Head:** 256-dim, dropout=0.1 (simpler head for less overfit)
92
  - **Precision:** FP16 mixed precision
93
 
94
  ## Evaluation Results
 
25
 
26
  - **Model type:** Text Classification (3‑tier, single head)
27
  - **Language:** English
 
28
  - **License:** Apache‑2.0
29
  - **Finetuned from model:** microsoft/deberta-v3-xsmall
30
  - **Training data:** `RowRed/prompts-24000-en` (L0 downsampled to 2,000, L1/L2/L3 kept fully)
 
87
  - **Max sequence length:** 256 tokens
88
  - **Epochs:** 12 max (Early stopping patience = 4 on F1-macro; stopped at epoch 7, best checkpoint at epoch 3)
89
  - **Class balancing:** WeightedRandomSampler only; class weights in loss removed to avoid double-counting
90
+ - **Head:** 256-dim, dropout=0.1
91
  - **Precision:** FP16 mixed precision
92
 
93
  ## Evaluation Results