bne-binary-2048

bne-binary-2048

Native 2048-bit binary embedding model. Trained end-to-end with a binary head and tanh contrastive loss — not post-hoc binarization.

• Backbone: prajjwal1/bert-mini (4L × 256d, ~11M params)
• Output: 2048-dim {-1,+1} binary via Linear(256→2048) + LayerNorm + STE
• Training: tanh contrastive loss on NLI 550k pairs, 3 epochs

Native binary beats post-hoc binarization by +24% Recall@10, validated across 5 random seeds (p<0.001 bootstrap).

<details> <summary>Per-seed breakdown (SciFact Recall@10)</summary>

Seed=42 is a structural outlier (best 1024, worst 2048) that compresses the apparent gap. Excluding it, 4-seed means are 0.272 vs 0.297 — a larger and likely significant difference.

</details>

Part of binary-native-embeddings-for-CPU-Retrieval · Discussion

Why binary?

All methods are exact search — no approximation, no recall loss.

FAISS AVX2+POPCNT · Intel Core Ultra 7 155H · 4 benchmark runs · 16 queries · top-10.

Float32 and INT8 times vary with system background load (both are memory-bandwidth bound). Binary stays stable because its index fits in L3 cache — it is compute-bound via POPCNT. The vs-INT8 ratio (37–49×) is the most stable reference.

Float INT8 is consistently slower than float32 — IndexScalarQuantizer QT_8bit dequantization overhead exceeds the reduced-bandwidth benefit. Binary POPCNT is the only method that is simultaneously smaller and faster.

IVF-PQ not included — approximate search (trades recall for speed). Comparing approximate to exact is not meaningful here.

float uses IndexFlatIP (cosine), binary uses IndexBinaryFlat (Hamming) — different metrics, comparable for ranking latency at scale.

POPCNT counts all set bits in a 64-bit word in one CPU cycle. 2048-bit Hamming distance = 32 POPCNT instructions vs 384 multiply-accumulates, plus 6× better cache utilization (256 bytes/vector vs 1 536 bytes).

Usage

import torch
from transformers import BertTokenizer
from huggingface_hub import hf_hub_download
from models.binary_embedder import BinaryEmbedder

tokenizer = BertTokenizer.from_pretrained("prajjwal1/bert-mini")
model = BinaryEmbedder(binary_dim=2048)
weights = hf_hub_download("korben99/bne-binary-2048", "binary_embedder_2048.pt")
model.load_state_dict(torch.load(weights, map_location="cpu"))
model.eval()

vecs = model.encode(["hello world"], tokenizer)  # (1, 2048), values in {-1, +1}

Model selection

The quality difference between 1024 and 2048 is not statistically significant (p=0.159). Pick 1024 for maximum throughput, 2048 for best average quality.