MiniMax-M2.5-REAP-139B-A10B-NVFP4-GB10

MiniMax-M2.5-REAP-139B-A10B-NVFP4-GB10

NVFP4 quantization of cerebras/MiniMax-M2.5-REAP-139B-A10B for NVIDIA DGX Spark (GB10).

The base model is a Cerebras REAP (Router-weighted Expert Activation Pruning) variant of MiniMaxAI/MiniMax-M2.5. REAP uniformly prunes experts from 256 → 154 (40% pruning), reducing total parameters from 230B to 139B while maintaining near-identical performance. This is the more aggressively pruned sibling of the 172B (25%) variant.

Model Details

Why 139B over 172B?

The 139B variant trades minimal quality for more memory headroom on a single DGX Spark. With 75 GB model weight vs 93 GB, there's an additional ~18 GB available for KV cache — which translates to longer achievable context or more concurrent sessions depending on how it's deployed.

Performance (Single NVIDIA DGX Spark — GB10, 128 GB)

TODO: Benchmark pending — model just quantized. Will update with llama-benchy results.

Expected: similar or slightly faster than 172B NVFP4 (27–29 tok/s) due to smaller model footprint.

Quantization Details

• Method: Post-training quantization via LLM Compressor (llmcompressor 0.10.0)
• Scheme: NVFP4 (compressed-tensors format)
• Calibration Dataset: HuggingFaceH4/ultrachat_200k (train_sft split)
• Calibration Samples: 64
• Max Sequence Length: 2048 tokens
• Ignore List: lm_head, model.embed_tokens, re:.*block_sparse_moe\.gate$
• Environment: LLMCOMPRESSOR_MOE_CALIBRATE_ALL_EXPERTS=1
• Hardware Used: NVIDIA DGX Spark (CPU offloading + 300GB swap)
• Total Quantization Time: 4.7 hours (281 minutes)
  • Quant pipeline model load: 50 seconds (27 BF16 shards into CPU RAM — this is llmcompressor load, NOT vLLM inference)
  • Calibration forward passes + weight calibration (28,892 weights): ~2+ hours (swap-dominated)
  • Model compression: 28,892 iterations in ~60 minutes (highly variable 1–16 it/s due to swap I/O)
  • Model save: 17 shards to disk
  • Bottleneck: swap I/O throughout (260GB model on 128GB RAM + 300GB swap)

Quantization Pipeline

The source model on HuggingFace is labeled BF16 but actually contains float8_e4m3fn weights with weight_scale_inv blocks of shape [128, 128]. A dequantization step was required before NVFP4 quantization:

1. Download: cerebras/MiniMax-M2.5-REAP-139B-A10B (131GB, 27 shards — FP8)
2. Dequant FP8 → BF16: Block-wise dequantization (multiply by scale_inv), output 260GB / 27 shards
3. Quantize BF16 → NVFP4: LLM Compressor oneshot with GB10-optimized ignore list
4. Output: 75GB / 17 shards (compressed-tensors format)

Ignore-list choice — attention stays quantized

This quant keeps all Linear layers quantized including self_attn (q/k/v projections). Other public NVFP4 variants (e.g. lukealonso/MiniMax-M2.7-NVFP4) keep attention in BF16, which leaves ~47% of per-token decode bandwidth on the table. I picked the aggressive path because NVFP4 calibration appeared to handle attention weight distributions cleanly on this architecture — but it's a different design choice than lukealonso's, and lukealonso's may be the right pick for deployers who prefer the safety margin of BF16 attention.

Container Setup for Quantization

# Image: avarok/dgx-vllm-nvfp4-kernel:v23 (has llmcompressor + deps)
# Override entrypoint since default launches vLLM server

docker run -d --name minimax-139b-quant \
  --gpus all --ipc=host \
  -v /opt/huggingface/models/MiniMax-M2.5-REAP-139B-A10B-BF16-real:/workspace/input_model \
  -v /opt/huggingface/models/MiniMax-M2.5-REAP-139B-A10B-NVFP4-GB10:/workspace/output_model \
  -v /opt/huggingface/models/quantize-minimax-139b.py:/workspace/quantize.py \
  -e LLMCOMPRESSOR_MOE_CALIBRATE_ALL_EXPERTS=1 \
  --entrypoint bash \
  avarok/dgx-vllm-nvfp4-kernel:v23 \
  -c "pip install --upgrade transformers && python /workspace/quantize.py"

Important: The --entrypoint bash override is required because the default entrypoint launches vLLM. The pip install --upgrade transformers is needed because the image ships an older transformers that doesn't support MiniMax M2 architecture.

Swap Configuration

The 260GB BF16 model exceeds 128GB physical RAM. A 300GB swap file was created:

sudo fallocate -l 300G /opt/huggingface/swapfile
sudo chmod 600 /opt/huggingface/swapfile
sudo mkswap /opt/huggingface/swapfile
sudo swapon /opt/huggingface/swapfile

This causes significant I/O stalls during compression (speed drops from 16 it/s to 1 it/s when paging), but the process completes successfully.

Running on a Single DGX Spark

Docker image: avarok/dgx-vllm-nvfp4-kernel:v23 (vLLM 0.16.0-rc2, CUDA 13.0, SM 12.1)

Download the model:

huggingface-cli download saricles/MiniMax-M2.5-REAP-139B-A10B-NVFP4-GB10 \
  --local-dir /opt/huggingface/models/MiniMax-M2.5-REAP-139B-NVFP4

Launch:

docker run -d --name minimax-139b --gpus all --ipc=host \
  -v /opt/huggingface/models/MiniMax-M2.5-REAP-139B-NVFP4:/models/MiniMax-M2.5-REAP-139B-NVFP4 \
  -p 8000:8000 \
  -e VLLM_NVFP4_GEMM_BACKEND=marlin \
  -e VLLM_TEST_FORCE_FP8_MARLIN=1 \
  -e VLLM_USE_FLASHINFER_MOE_FP4=0 \
  -e VLLM_MARLIN_USE_ATOMIC_ADD=1 \
  -e MODEL=/models/MiniMax-M2.5-REAP-139B-NVFP4 \
  -e PORT=8000 \
  -e MAX_MODEL_LEN=131072 \
  -e GPU_MEMORY_UTIL=0.93 \
  -e "VLLM_EXTRA_ARGS=--trust-remote-code --kv-cache-dtype fp8 --attention-backend flashinfer --enable-auto-tool-choice --tool-call-parser minimax_m2 --reasoning-parser minimax_m2_append_think" \
  avarok/dgx-vllm-nvfp4-kernel:v23

Note: With 75 GB model weight (vs 93 GB for 172B), MAX_MODEL_LEN can likely go higher — 131072 should be achievable. Benchmarks will confirm exact limits.

Test it:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M2.5-REAP-139B-NVFP4",
    "messages": [{"role": "user", "content": "Hello!"}],
    "temperature": 1.0,
    "top_p": 0.95,
    "top_k": 40,
    "min_p": 0.01,
    "max_tokens": 512
  }'

Environment Variables

Recommended Sampling Parameters

Per MiniMax documentation:

{
  "temperature": 1.0,
  "top_p": 0.95,
  "top_k": 40,
  "min_p": 0.01
}

Comparison with other public M2.5 REAP NVFP4 quants

Different ignore-list philosophies — lukealonso keeps attention in BF16 (safer / more conservative), the GB10-targeted variants here quantize it (smaller, faster on this hardware, slightly more aggressive). Use whichever matches the tradeoff you prefer.

Related Models

• saricles/MiniMax-M2.5-REAP-172B-A10B-NVFP4-GB10 — sibling 172B REAP NVFP4 (93 GB, 28 tok/s)
• saricles/Qwen3-Next-80B-A3B-Coder-NVFP4-GB10 — Qwen3 Coder NVFP4 (62 tok/s)
• cerebras/MiniMax-M2.5-REAP-139B-A10B — Source FP8 model
• cerebras/MiniMax-M2.5-REAP-172B-A10B — 172B FP8 variant

Acknowledgments

• Base model by MiniMax
• REAP sparse-inference pruning by Cerebras (paper)
• Quantization tooling by vLLM / LLM Compressor
• Quantized by saricles on NVIDIA DGX Spark