Sipsa Labs, Inc.

Sipsa Labs builds SIP, the operating system for the physical world.

UltraCompress is retired. Its public materials preserve historical research; new API access and paid pilots are no longer offered.

The research snapshot below is historical. Results apply to the versions and evaluation conditions recorded at the time.

Historical snapshot: v0.6.22 22 architectures shipped; 20 PPL-verified (19 transformer + 1 SSM, with comparator-note caveat) BUSL-1.1 + Additional Use Grant

Historical 5-bit compression research

22 architectures shipped, 20 PPL-verified end-to-end (19 transformer + 1 SSM with comparator-note caveat). Dense + Mixture-of-Experts + state-space, 0.6B to 405B parameters. These 5-bit artifacts use lossy weight quantization. Reconstructing a validated quantized artifact is distinct from recovering the original bf16 weights without loss. The public CLI checks structure and computes fingerprints; it does not reconstruct weights or automatically compare with a trusted reference.

Tightest verified PPL ratios at 5 bpw

(perplexity ratio = compressed PPL / bf16 baseline PPL; FineWeb-edu held-out tail; seq_len = 1024; n = 30-50; seed = 42)

Model	Params	Type	PPL ratio	Notes
Phi-3-mini-4k-instruct	3.8B	dense	1.00262×	seq_len=128 caveat
Mixtral-8x7B	47B	MoE	1.00368×	tightest MoE result
Qwen3-1.7B-Base	1.7B	dense	1.00401×	small-decoder record
Qwen3-14B	14B	dense	1.00403×	14B-class record
Yi-1.5-9B	8.8B	dense	1.00414×	>8B record
Qwen3-8B	8B	dense	1.00440×	8B-class record
Mistral-7B-v0.3	7B	dense	1.00548×	a strong dense 7B-class result
Hermes-3-Llama-3.1-405B	405B	dense	1.0066×	historical 405B-class 5-bit result
Qwen3-0.6B	0.6B	dense	1.0069×
OLMo-2-0425-1B	1B	dense	1.0073×
SmolLM2-1.7B-Instruct	1.7B	dense	1.0075×
Mamba-2.8B	2.8B	SSM	1.0059×	state-space model (uses SSM-compatible comparator; canonical transformer streaming reconstruction is architecture-incompatible with SSMs)
Llama-3.1-8B	8B	dense	1.0125×	standard eval

Historical source: github.com/sipsalabs/ultracompress · recorded evaluations: historical registry

Inspect pack structure and compute fingerprints

pip install "ultracompress==0.6.27"
hf download SipsaLabs/qwen3-8b-uc-v3-bpw5 --local-dir ./qwen3-8b
uc verify ./qwen3-8b

Expected result category: basic pack-structure checks and, unless hashing is skipped, a computed fingerprint. This is not reconstruction verification or an automatic comparison with a trusted reference. Without an API key, uc try prints a recorded reference response, not a live inference result.

Historical API and services

New API access and paid pilots are no longer offered. Historical materials remain in the public repository and model collection.

---

License + IP

• PyPI ultracompress v0.6+ under BUSL-1.1 with Additional Use Grant — free for sub-$1M ARR companies + research + individuals. Auto-converts to Apache 2.0 four years after each release.
• v0.5.x stays Apache-2.0 forever on legacy/0.5.x.
• Codec internals patent-protected. Continuations through 2027.

---

Contact

• Project inquiries → founder@sipsalabs.com
• Patents / licensing → legal@sipsalabs.com
• Press / media → press@sipsalabs.com
• Security disclosure → security@sipsalabs.com
• General → hello@sipsalabs.com

sipsalabs.com · github.com/sipsalabs · @SipsaLabs on X

Public CLI documentation: github.com/sipsalabs/ultracompress · Selective Disclosure Charter — what we publish vs what we keep internal: see github.com/sipsalabs/ultracompress