tahamajs/bert-comp-wanda-sst2

BERT-Base (wanda) — NeuralOps Compressed & LoRA-Healed Model Zoo

This repository provides a physically compressed and LoRA-healed checkpoint of google-bert/bert-base-uncased fine-tuned on GLUE SST-2, compressed using the wanda (WandaCompressor) engine from the NeuralOps Library at Pareto-selected rank k=32 (1.00x compression ratio, -0.0% parameter reduction).

---

Compression & Architecture Profile

WandaCompressor: Wanda Compressor: Pruning by Weights and Activations. Prunes weights according to the product of weight magnitudes and input activation norms.

Property	Specification
Registry Key (COMPRESSOR_REGISTRY)	wanda
NeuralOps Class & Family	WandaCompressor (pruning)
Source Implementation	neuralops/compression/pruning/wanda.py
Base Architecture	google-bert/bert-base-uncased (109,483,778 parameters, 12 layers, $D=768$)
Showcase Rank (k)	32 (LoRA Recovery Applied: False)
Compressed Parameter Count	109,483,778 (-0.00% reduction, 1.000x physical compression ratio)
Teacher Baseline SST-2 Acc	91.63%
Showcase Final SST-2 Acc	87.84% (Zero-Shot: 87.84%, Parity Delta: -3.79%)

---

Multi-Rank Pareto Evaluation (k=256, 128, 64, 32)

Evaluated physically on NVIDIA GeForce RTX 3090 (24GB) on the GLUE SST-2 validation split (872 samples). Whenever zero-shot compression drops accuracy by > 2.0% alongside > 1.10x reduction, NeuralOps applies targeted r=8 LoRA healing (50 steps, lr=2e-4, 45s budget):

Rank (k)	Physical Ratio	Compressed Params (Reduction)	Zero-Shot Acc	Healed Acc	Parity vs Teacher (91.63%)
256	1.00x	109,483,778 (-0.0%)	87.84%	87.84%	-3.79%
128	1.00x	109,483,778 (-0.0%)	87.84%	87.84%	-3.79%
64	1.00x	109,483,778 (-0.0%)	87.84%	87.84%	-3.79%
32 ⭐	1.00x	109,483,778 (-0.0%)	87.84%	87.84%	-3.79%

(⭐ indicates the serialized checkpoint packaged in this Hugging Face repository.)

---

Packaged Artifacts

• pytorch_model.bin: Full serialized compressed BERT state dictionary (with low-rank factors / quantization buffers / LoRA healing adapters baked in).
• config.json: Model configuration tagged with compression_method: "wanda" and compression_rank: 32.
• compression_summary.json: Complete machine-readable telemetry across all evaluated ranks [256, 128, 64, 32].
• Tokenizer Assets: Standard bert-base-uncased vocabulary (vocab.txt, tokenizer.json, tokenizer_config.json).

---

Quick Start (Reconstruction & Inference)

import torch
from huggingface_hub import hf_hub_download
from transformers import AutoTokenizer, BertForSequenceClassification
from neuralops.core.compression_pipeline import compress_model

repo_id = "tahamajs/bert-comp-wanda-sst2"
tokenizer = AutoTokenizer.from_pretrained(repo_id)

# Reconstruct compressed NeuralOps topology and load serialized compressed + healed weights
model = BertForSequenceClassification.from_pretrained("google-bert/bert-base-uncased", num_labels=2)
model = compress_model(model, {"compressors": [{"method": "wanda", "rank": 32}]})

ckpt_path = hf_hub_download(repo_id=repo_id, filename="pytorch_model.bin")
state_dict = torch.load(ckpt_path, map_location="cpu")
model.load_state_dict(state_dict, strict=False)
model.eval()

inputs = tokenizer("NeuralOps achieves strong compression without sacrificing downstream accuracy.", return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits
    pred = logits.argmax(dim=-1).item()
print("Predicted SST-2 Sentiment (0=Negative, 1=Positive):", pred)

---

Citation

@software{neuralops2026compressorzoo,
  title   = {NeuralOps: Multi-Rank Compressed BERT-SST2 Zoo with Pareto LoRA Recovery},
  author  = {Majlesi, Taha},
  year    = {2026},
  url     = {https://github.com/tahamajs/NeuralOps}
}