At a glance

Which variant should I pick?

40% expert-pruned MiniMax-M2.1 using REAP (Router-weighted Expert Activation Pruning)

Stress Test Results

Tested at 4 temperatures (0.0, 0.2, 0.7, 1.0) across 6 prompt types (24 total tests):

Result: 24/24 tests passed, 0 loops detected

Usage

python
Copy code
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "0xSero/MiniMax-M2.1-139B",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "0xSero/MiniMax-M2.1-139B",
    trust_remote_code=True,
)

messages = [{"role": "user", "content": "Write a Python function to calculate fibonacci numbers."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

DynamicCache Compatibility Fix (transformers 4.55+)

If you encounter TypeError: CacheLayerMixin.__init__() got an unexpected keyword argument, add this before importing the model:

python
Copy code
from transformers import cache_utils
_orig = cache_utils.DynamicCache.__init__
def _patched(self, *args, **kwargs):
    cfg = kwargs.get("config")
    if cfg and hasattr(cfg, "model_type") and "minimax" in str(getattr(cfg, "model_type", "")):
        kwargs.pop("config", None)
        kwargs.pop("max_cache_len", None)
        kwargs.pop("max_batch_size", None)
        return _orig(self, None)
    return _orig(self, *args, **kwargs)
cache_utils.DynamicCache.__init__ = _patched

Model Comparison

Quantized Versions

• MiniMax-M2.1-REAP-40-W4A16 (Coming Soon) - 4-bit weights, ~58GB VRAM

Why 40% Pruning?

The 40% pruning ratio offers the best balance of:

• Size reduction: 139B vs 456B original (70% smaller)
• VRAM savings: ~278GB vs ~912GB (fits on 4x H100 80GB)
• Stability: 0 loops in comprehensive stress testing
• Performance: Minimal quality degradation from strategic expert selection

REAP Methodology

REAP (Router-weighted Expert Activation Pruning) uses calibration data to identify which experts are most important based on router activation patterns. Unlike random or magnitude-based pruning, REAP preserves the experts that are actually used during inference.

Calibration Dataset: 2098 samples

• pile-10k: 498 samples (general text)
• evol-codealpaca: 800 samples (code generation)
• xlam-function-calling: 800 samples (function calling)

License & citation

License inherited from the base model.

bibtex
Copy code
@misc{lasby2025reap,
  title  = {REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression},
  author = {Mike Lasby and Ivan Lazarevich and Nish Sinnadurai and Sean Lie and Yani Ioannou and Vithursan Thangarasa},
  year   = {2025}, eprint = {2510.13999}, archivePrefix = {arXiv}
}

Sponsors

Made possible by NVIDIA · TNG Technology · Lambda · Prime Intellect · Hot Aisle.