Skip to content

MLNode 3.0.14 & Minimax m2.7#1260

Open
gmorgachev wants to merge 4 commits into
mainfrom
mlnode-v3.0.14
Open

MLNode 3.0.14 & Minimax m2.7#1260
gmorgachev wants to merge 4 commits into
mainfrom
mlnode-v3.0.14

Conversation

@gmorgachev

Copy link
Copy Markdown
Contributor

No description provided.

tcharchian added a commit to gonka-ai/gonka-docs that referenced this pull request May 27, 2026
…gs (#1143)

Follow-up to gonka-ai/gonka#1260 (MLNode 3.0.14 & MiniMax M2.7).

- Add MiniMaxAI/MiniMax-M2.7 to the Supported models table and the
  Proposed Hardware Configuration table (4xA100 / 4xH100 / 2xH200 /
  2xB200, ~320 GB VRAM).
- Add a "Reference deploy configs in the repo" tip pointing to every
  node-config-*.json in deploy/join/ for Qwen, Kimi, and MiniMax, so
  hosts can copy a ready file instead of writing one from scratch.
- Add new tabs in the "Edit Inference Node Description" section,
  mirroring the JSON files shipped in gonka/deploy/join/:
    * Qwen — 8xB200
    * Kimi — 8xH200 (FLASHMLA, tp=8)
    * MiniMax — 4xA100 (marlin MoE + VLLM_USE_FLASHINFER_MOE_FP8=0)
    * MiniMax — 4xH100 (FLASHINFER + fp8 kv-cache)
    * MiniMax — 2xH200 (FLASHINFER + fp8 kv-cache; matches the
      configuration used to record MiniMax PoC golden vectors)
    * MiniMax — 2xB200 (FLASHINFER_TRTLLM MoE + fp8 kv-cache)
- Add a MiniMax M2.7 tab in the Pre-download Model Weights section
  with the huggingface-cli command and the MLNode 3.0.14 / A100 env
  var note.
- Add a MiniMax delegation example in the "Optional: PoC delegation
  and refusal" section.

All vLLM argument sets are taken verbatim from the JSON files in
gonka/deploy/join/ so docs and repo stay in sync.

Co-authored-by: Cursor <cursoragent@cursor.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants