inference-optimization/MiniMax-M2.5.w4a16

MiniMax-M2.5-quantized.w4a16

Model Overview

• Model Architecture: MiniMaxM2ForCausalLM
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT4
• Intended Use Cases:
  • Reasoning.
  • Function calling.
  • Subject matter experts via fine-tuning.
  • Multilingual instruction following.
  • Translation.
• Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws).
• Release Date: 02/12/2026
• Version: 1.0
• Model Developers: RedHat (Neural Magic)

Model Optimizations

This model was obtained by quantizing the weights of MiniMax-M2.5 to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%.

Only the weights of the linear operators within transformers blocks are quantized. Weights are quantized using a asymmetric per-group scheme, with group size 64. The GPTQ algorithm is applied for quantization, as implemented in the llm-compressor library.

Deployment

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "RedHatAI/MiniMax-M2.5-quantized.w4a16"
number_gpus = 1
sampling_params = SamplingParams(temperature=1.0, top_p=0.95, top_k=40, min_p=0, max_tokens=256)

messages = [
    {"role": "user", "content": prompt}
]

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [{"role": "user", "content": "Give me a short introduction to large language model."}]

prompts = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)

llm = LLM(model=model_id, tensor_parallel_size=number_gpus)

outputs = llm.generate(prompts, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Creation

Creation details

This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below.

from datasets import load_dataset
from transformers import AutoModelForCausalLM, AutoTokenizer, AutoProcessor
from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import GPTQModifier

MODEL_ID = "inference-optimization/MiniMax-M2.5-BF16"

# Load model.
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(MODEL_ID)


NUM_CALIBRATION_SAMPLES=512
MAX_SEQUENCE_LENGTH=2048

# Load dataset.
ds = load_dataset("HuggingFaceH4/ultrachat_200k", split=f"train_sft[:{NUM_CALIBRATION_SAMPLES}]", trust_remote_code=True)
ds = ds.shuffle(seed=42)

# Preprocess the data into the format the model is trained with.
def preprocess(example):
    return {"text": tokenizer.apply_chat_template(example["messages"], tokenize=False, )}

ds = ds.map(preprocess)

# Tokenize the data (be careful with bos tokens - we need add_special_tokens=False since the chat_template already added it).
def tokenize(sample):
    return tokenizer(sample["text"], padding=False, max_length=MAX_SEQUENCE_LENGTH, truncation=True, add_special_tokens=False)
ds = ds.map(tokenize, remove_columns=ds.column_names)

# Configure the quantization algorithm to run.
recipe = GPTQModifier( scheme="W4A16", weight_observer="mse", targets= [r"re:.*block_sparse_moe\.experts\.\d+\.w[1-3]$", r"re:.*mlp\.experts\.\d+\.(gate|up|gate_up|down)_proj$" ], ignore=["re:.*self_attn.*", "lm_head"])


# Apply quantization.
oneshot(
    model=model, dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    processor=processor,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES
)

# Save to disk compressed.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + ".w4a16"
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)

</details>
 



## Evaluation

The model was evaluated on the ifeval, mmlu_pro and gsm8k_platinum  using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness), on reasoning tasks using [lighteval](https://github.com/neuralmagic/lighteval/tree/reasoning).
[vLLM](https://docs.vllm.ai/en/stable/) was used for all evaluations.


<details>
  <summary>Evaluation details</summary>

  Deploy using vllm to create an OpenAI-compatible API endpoint:

- vLLM:
    ```shell
    vllm serve RedHatAI/MiniMax-M2.5.w4a16 --max-model-len 262144 --reasoning-parser deepseek_r1
    ```

  **lm-evaluation-harness**

lm_eval --model local-chat-completions
--tasks mmlu_pro_chat
--model_args "model=RedHatAI/MiniMax-M2.5.w4a16,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=64,max_retries=3,tokenized_requests=False,tokenizer_backend=None,timeout=1200"
--num_fewshot 0
--apply_chat_template
--gen_kwargs "do_sample=True,temperature=1.0,top_p=0.95,top_k=40,min_p=0.0,max_gen_toks=64000

lm_eval --model local-chat-completions
--tasks ifeval
--model_args "model=RedHatAI/MiniMax-M2.5.w4a16,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=64,max_retries=3,tokenized_requests=False,tokenizer_backend=None,timeout=1200"
--num_fewshot 0
--apply_chat_template
--gen_kwargs "do_sample=True,temperature=1.0,top_p=0.95,top_k=40,min_p=0.0,max_gen_toks=64000

lm_eval --model local-chat-completions
--tasks gsm8k_platinum_cot_llama
--model_args "model=RedHatAI/MiniMax-M2.5.w4a16,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=64,max_retries=3,tokenized_requests=False,tokenizer_backend=None,timeout=1200"
--num_fewshot 0
--apply_chat_template
--gen_kwargs "do_sample=True,temperature=1.0,top_p=0.95,top_k=40,min_p=0.0,max_gen_toks=64000

**lighteval**

lighteval_model_arguments.yaml
```yaml 
model_parameters:
  model_name: RedHatAI/MiniMax-M2.5.w4a16
  dtype: auto
  gpu_memory_utilization: 0.9
  max_model_length: 40960
  generation_parameters:
    temperature: 1.0
    top_k: 40
    min_p: 0.0
    top_p: 0.95
    max_new_tokens: 64000

lighteval endpoint litellm lighteval_model_arguments.yaml  \
  "aime25|0,math_500|0,gpqa:diamond|0"

Accuracy

Benchmark	inference-optimization/MiniMax-M2.5-BF16	inference-optimization/MiniMax-M2.5.w4a16	Recovery (%)
GSM8k Platinum (0-shot)	95.15	96.36	101.27
IfEval (0-shot)	88.17	85.58	97.06
AIME 2025	87.50	84.17	96.19
GPQA diamond	83.67	84.51	101.01
Math 500	87.33	87.60	100.31
Mmlu Pro Chat	80.83	81.25	100.51