What changed
The current public PLI Labs GitHub organization has one public repo: ProprietaryLegal/pli-labs. The June 18 upload turns private operational research into public-safe guidance without exposing hostnames, internal paths, private prompts, client material, endpoints, secrets, or raw failure runbooks.
Why it matters
Law firms interested in private AI need realistic infrastructure answers and auditable model releases. The public upload connects the model side and the hardware side: MiniMax-M2.7 legal REAP candidates on Hugging Face, public model cards on GitHub, a B70 local-serving lane, and V100 research for older owned GPU fleets.
GitHub upload map
| Commit | Public upload | What it added |
|---|---|---|
| be010c5 | PLI Labs documentation | Public wiki, legal REAP thesis, release principles, tools, and repo scaffolding. |
| f3cb350 | V100 research | Sanitized V100 master plan, findings, benchmark summaries, reports, and public testing standard. |
| ed2a0c6 | B70 research | Four-card Arc Pro B70 stack findings, llama.cpp SYCL success, XPU guardrails, benchmark results, and public copy. |
| 1d490bb | Formal model cards | MODEL_CARD pages for the MiniMax-M2.7 legal REAP 0.35 and 0.45 candidates. |
| 3ccb3e3 | HF docs and owned-hardware profiles | Hugging Face card source, five release links, B70 successful launch profile, and V100 successful launch profiles. |
Hugging Face release map
- MiniMax-M2.7 Legal REAP 0.35 - conservative Transformers/safetensors candidate.
- MiniMax-M2.7 Legal REAP 0.45 - more aggressive Transformers/safetensors candidate.
- MiniMax-M2.7 Legal REAP 0.35 GGUF - GGUF release page for local-serving evaluation.
- MiniMax-M2.7 Legal REAP 0.45 GGUF - GGUF release page for the more aggressive candidate.
- MiniMax-M2.7 REAP 172B-A10B Q4_K_M GGUF - B70-friendly GGUF artifact for large local-serving research.
New public GitHub sections
- PLI Labs repository README
- Legal REAP thesis
- REAP release notes
- Hugging Face release-card source
- B70 research package
- V100 public master plan
- V100 successful launch profiles
- PLI Labs tooling principles
Plain-English takeaway
The update is not a generic AI announcement. It is a public release system for private legal AI: document the model, publish the model card, link the Hugging Face repo, explain the hardware lane, disclose the measured benchmark shape, and keep private client and deployment material out of the public record.