sip786's picture
docs: honest-positioning — near-lossless + verifiable reconstruction (no 'lossless'/'bit-identical to original')
9ba110d verified
metadata
license: other
license_name: busl-1.1
license_link: https://github.com/sipsalabs/ultracompress/blob/main/LICENSE
tags:
  - ultracompress
base_model: microsoft/Phi-3-mini-4k-instruct

phi-3-mini-4k-instruct-uc-v3-bpw5

Quick start — try Sipsa near-lossless 5-bit in 30 seconds

pip install ultracompress
uc try sipsa-qwen3-0.6b      # 30s demo (free, no signup, no GPU)

This calls Sipsa's free inference API on a compressed model — see what 1.004x PPL ratio looks like in practice.

To run this model via the API: uc try sipsa-phi-3-mini-4k

Browse the full catalog of try-able models: uc catalog

Need this model in production?

Phase 0 POC: $5K / 5 business days / SHA-256 reproducible-reconstruction audit (deterministic decode to the validated artifact).

Contact: founder@sipsalabs.com | sipsalabs.com/poc

Near-lossless 5-bit compression (~1% perplexity; lossy) of microsoft/Phi-3-mini-4k-instruct — produced by Sipsa Labs (UltraCompress).

  • Reproducible, cryptographically verifiable reconstruction: a deterministic decode to the SHA-256-pinned validated artifact (not bit-identical to the original bf16 model).
  • Independently perplexity-verified end-to-end.
  • License: BUSL-1.1 + Additional Use Grant.
pip install ultracompress

Documentation & verification: https://sipsalabs.com · https://github.com/sipsalabs/ultracompress

Patents pending.