deploy-kimi-k3-on-rtx-pro-6000 SKILL.md skillA
deploy-kimi-k3-on-rtx-pro-6000 from soulmachine/skills is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add soulmachine/skills/deploy-kimi-k3-on-rtx-pro-6000@git:20260907.758eb45Pins git:20260907.758eb45 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- soulmachine/skills
- Path
deploy-kimi-k3-on-rtx-pro-6000/SKILL.md- Versions recorded
- 1
- Latest version
- git:20260907.758eb45, committed Sep 7, 2026
- Size
- 19,061 bytes
- Repository stars
- 3
- License
- MIT
Deploy and serve Moonshot Kimi-K3 (2.8T MoE, hybrid KDA+MLA attention, native vision via MoonViT-3d) from Unsloth's GGUF quantizations (unsloth/Kimi-K3-GGUF) via llama.cpp on a Linux server with 8x NVIDIA RTX PRO 6000 Blackwell Server Edition (96 GB, sm_120) GPUs. Unlike vLLM/SGLang's tensor-parallel all-in-VRAM serving, this is CPU+GPU hybrid inference (llama-server's --n-cpu-moe / --override-tensor): routed MoE experts split between GPU-resident (fast) and CPU-resident…
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Deploy Kimi-K3 (GGUF, llama.cpp) on 8x RTX PRO 6000 Blackwell Server Edition (sm_120) Serve **Kimi-K3** (2.8T MoE; hybrid Kimi Delta Attention + MLA; native MoonViT-3d vision; 1M context) from **Unsloth's GGUF quantizations**, via **llama.cpp built from source** — OpenAI- compatible API on a configurable port, **CPU+GPU hybrid** (not TP-across-VRAM), weights bind- mounted read-only from local NVMe. **Why llama.cpp, why GGUF, why hybrid — not vLLM/SGLang/TP:** at UD-Q2_K_XL the checkpoint is ~861 GB — larger than an 8x96 GB VRAM pool (768 GB) can hold even before KV cache, so a pure-VRAM tensor-parallel engine (the pattern `deploy-kimi-k26-on-rtx-pro-6000` uses) is a non-starter no matter how many GPUs you throw at it. Two ways to bridge that gap both get called "CPU offload," and only one of them works here: - **Layer-wise weight streaming** (vLLM `--cpu-offload-gb`, SGLang `--cpu-offload-gb` `--offload-group-size`) shuttles whole weight tensors GPU<->CPU every forward pass while the GPU still does the math — decode becomes PCIe-bound. At this model's scale (~800 GB of experts) it's unusable, not just slow. …
Read the whole file at its exact version.
How to install
mdr add soulmachine/skills/deploy-kimi-k3-on-rtx-pro-6000@sha256:b5130c1d72dc1f54The line at the top of this page pins the label git:20260907.758eb45, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_iselavtl76qb3ifi)
2 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 1 version of this file, committed on Sep 7, 2026.
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (19061 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
soulmachine/skills · 3 stars · license MIT · pushed 2026-10-31 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_iselavtl76qb3ifi GET https://markdownregistry.com/api/v1/resolve?ref=soulmachine/skills/deploy-kimi-k3-on-rtx-pro-6000 GET https://markdownregistry.com/api/v1/blob/b5130c1d72dc1f54d6879b21ee1ff2c51aaa87944e3608f12ddcbb24063d74e6
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.