ykae
/

monarch-bert-base-mnli

Text Classification

monarch-matrices

hardware-efficient

Eval Results (legacy)

text-embeddings-inference

Model card Files Files and versions

ykae commited on 15 days ago

Commit

22bae58

·

verified ·

1 Parent(s): da0c393

Update README.md

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -71,7 +71,7 @@ Measured on a single NVIDIA H100 using `torch.compile(mode="max-autotune")`.
 | **Compute (GFLOPs)** | 696.5 | **232.6** | 📉 **-66.6%** |
 | **Throughput (TPS)** | 7261 | **9029** | 🚀 **+24.3%** |
 | **Latency (Batch 32)** | 4.41 ms | **3.54 ms** | ⚡ **+24.6% Faster** |
-| **Accuracy (MNLI)** | 83.62% | **78.34%** | 📉 **-5.28% (towards 1% acc drop)**
 **Our compression scales predictably: as training data increases, the accuracy deficit converges toward zero.**

 | **Compute (GFLOPs)** | 696.5 | **232.6** | 📉 **-66.6%** |
 | **Throughput (TPS)** | 7261 | **9029** | 🚀 **+24.3%** |
 | **Latency (Batch 32)** | 4.41 ms | **3.54 ms** | ⚡ **+24.6% Faster** |
+| **Accuracy (MNLI)** | 83.62% | **78.34%** | 📉 **-5.28% (towards 0% acc drop)**
 **Our compression scales predictably: as training data increases, the accuracy deficit converges toward zero.**