Educational Purpose Only: Cultural perspectives & historical research — not medical, financial, or legal advice. Editorial standards.

Skip to content
engineering

Kush v7 Release Notes

Kush v7 ships with a canonical prompt contract, cleaner training data, stronger eval gates, and a more stable deployment path.

hotep-llm kush release prompt-engineering evaluation
T

Tunde Adeyemi

Technology & Engineering Editor · · Updated

Tunde Adeyemi oversees technology and engineering content at Hotep Intelligence. As an ML engineer and open source advocate, he ensures technical content is accurate, accessible, and aligned with the project's mission of digital sovereignty. His expertise spans machine learning, self-hosted infrastructure, and the application of AI for community empowerment.

Editorially Reviewed

by Hotep Intelligence Editorial Team · Kemetic History, Holistic Wellness, ML Engineering

Our editorial standards →

Kush v7 Release Notes

Update (November 2026): Production now runs Kush v9.1 (hotep-llm-kush-v91), achieving 7/7 Exact Entity Recall and 7/7 Prime Agent E2E at 48.2 tok/s. Read the Kush v9.1 Release Notes.

Kush v7 was promoted to production in May 2026.

This version is not a cosmetic bump. It is a cleanup of the entire release path:

  • the runtime prompt now comes from one canonical contract
  • the training payload rejects duplicate rows, prompt bloat, and mixed-version leakage
  • the eval gate runs pairwise comparisons when a baseline is available
  • the load test checks the live model instead of assuming the compare model is loaded
  • the deployment path is aligned so the Mac mini does not keep multiple Kush variants warm at once

The goal is simple: make the model more consistent, easier to maintain, and harder to regress.

What changed

The biggest change is structural. Kush used to rely on prompt text scattered across runtime, persona fallback, export scripts, and training builders. That created drift. Different parts of the system were teaching slightly different versions of the same assistant.

Kush v7 replaces that with one short prompt contract. The runtime, the persona fallback, and the export path now all point at the same source of truth. That matters because prompt quality is not just about wording. It is about consistency.

We also tightened the corpus pipeline:

  • long system prompts are blocked
  • repeated system prompts in a single row are blocked
  • near-duplicates are removed
  • mixed-version leakage is blocked
  • low-signal rows are filtered out before export

This keeps the model from learning noise at scale.

Why it matters

Models drift when the data and the prompt stop agreeing with each other. The result is familiar:

  • inconsistent address forms
  • unnecessary repetition
  • weaker persona control
  • brittle behavior under load

Kush v7 addresses those problems by making the prompt contract short, structured, and testable. It also moves release decisions toward evidence:

  • corpus lint
  • paired evaluation
  • judge sampling
  • load testing

That is the difference between a model that looks good in a demo and a model that survives production traffic.

What to expect

You should see a cleaner voice, fewer duplicate phrases, and more predictable behavior under real use. The model should be easier to retrain because the release pipeline now gives us a clearer signal about what is working and what is not.

What comes next

The next step is not more prompt inflation. It is more signal:

  • higher-quality feedback from live users
  • better baseline coverage in evals
  • a tighter loop between production traces and retraining candidates

Release Gate Checklist Summary

Before any successor adapter replaces Kush v7 on our self-hosted zero-dollar production LLM stack, it must clear four independent promotion gates:

  1. Prompt Contract Parity: Identical system instructions across training export, evaluation harness, and runtime inference so model behavior never drifts between benchmark and production.
  2. Corpus Hygiene Audit: Zero duplicate assistant completions, zero rubric or evaluator tag leakage, and balanced category representation across history, alkaline health, finance, and sovereign AI architecture.
  3. Pairwise Blind Evaluation: Head-to-head judge comparisons against the live production baseline across core entity recall, tone consistency, and factual grounding.
  4. Live Concurrency Verification: Sub-second time-to-first-token and stable memory footprint under concurrent requests on local Apple Silicon and RTX inference nodes, documented further in our Kush v8.2 training pipeline hardening postmortem and engineering blog.

Frequently Asked Questions

What base architecture powers Kush v7? Kush v7 is a supervised fine-tune (QLoRA merged to quantized GGUF) built on Meta’s Llama 3.1 8B Instruct foundation model, served locally without external cloud API dependencies.

Why did Kush v7 unify the system prompt contract? Earlier builds maintained slightly different system prompts across training scripts, fallback handlers, and production servers. Consolidating to one canonical contract eliminated persona drift and improved benchmark reproducibility.

Kush v7 is the current release. The release process around it is now the product.

📚

Sources & References

  1. Grattafiori et al. (Meta AI)(2024). The Llama 3 Herd of ModelsJournal
  2. Edward J. Hu et al.(2022). LoRA: Low-Rank Adaptation of Large Language ModelsJournal
  3. Lianmin Zheng et al.(2023). Judging LLM-as-a-Judge with MT-Bench and Chatbot ArenaJournal

Want to explore this topic further?

Ask Hotep about engineering wisdom and get personalized guidance.

Full Guides & Hubs

Explore deeper on AskHotep.ai & Knowledge Hubs

Continue Your Journey