MiniMax M2.5 BF16 INT4 AWQ — Hardware Requirements & GPU Compatibility
ChatCodeFunctions
An AWQ INT4 quantization of MiniMax M2.5 prepared by mratsim, featuring 39.1 billion effective parameters in a Mixture-of-Experts architecture. This model supports chat, code generation, and function calling, making it a versatile general-purpose assistant. The AWQ quantization is optimized for GPU inference with minimal quality loss. As an MoE model, M2.5 activates only a subset of its parameters per token, offering strong performance relative to its total size while keeping inference costs manageable. The INT4 AWQ format requires GPU inference with compatible frameworks like vLLM or AutoAWQ. Expect to need 24 GB or more of VRAM. A solid choice for users with high-end GPUs who want a capable all-rounder with function calling support.
56.0K downloads
34 likesSpecifications
- Publisher
- mratsim
- Parameters
- 39.1B
- Architecture
- MiniMaxM2ForCausalLM
- Context Length
- 196,608 tokens
- Vocabulary Size
- 200,064
- Release Date
- 2026-02-17
- License
- Other
Get Started
HuggingFace
How Much VRAM Does MiniMax M2.5 BF16 INT4 AWQ Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| IQ2_XXS | 2.20 | 11.3 GB | 36.0 GB | 10.76 GB | Importance-weighted 2-bit, extreme compression — significant quality loss |
| IQ2_M | 2.70 | 13.8 GB | 38.5 GB | 13.20 GB | Importance-weighted 2-bit, medium |
| IQ3_XXS | 3.10 | 15.7 GB | 40.4 GB | 15.16 GB | Importance-weighted 3-bit |
| Q2_K | 3.40 | 17.2 GB | 41.9 GB | 16.62 GB | 2-bit quantization with K-quant improvements |
| Q3_K_S | 3.50 | 17.7 GB | 42.4 GB | 17.11 GB | 3-bit small quantization |
| Q3_K_M | 3.90 | 19.6 GB | 44.3 GB | 19.07 GB | 3-bit medium quantization |
| Q4_0 | 4.00 | 20.1 GB | 44.8 GB | 19.56 GB | 4-bit legacy quantization |
| IQ4_XS | 4.30 | 21.6 GB | 46.3 GB | 21.02 GB | Importance-weighted 4-bit, compact |
| Q4_1 | 4.50 | 22.6 GB | 47.3 GB | 22.00 GB | 4-bit legacy quantization with offset |
| Q4_K_S | 4.50 | 22.6 GB | 47.3 GB | 22.00 GB | 4-bit small quantization |
| IQ4_NL | 4.50 | 22.6 GB | 47.3 GB | 22.00 GB | Importance-weighted 4-bit, non-linear |
| Q4_K_M | 4.80 | 24.0 GB | 48.7 GB | 23.47 GB | 4-bit medium quantization — most popular sweet spot |
| Q5_K_S | 5.50 | 27.4 GB | 52.2 GB | 26.89 GB | 5-bit small quantization |
| Q5_K_M | 5.70 | 28.4 GB | 53.1 GB | 27.87 GB | 5-bit medium quantization — good quality/size tradeoff |
| Q6_K | 6.60 | 32.8 GB | 57.5 GB | 32.27 GB | 6-bit quantization, very good quality |
| Q8_0 | 8.00 | 39.7 GB | 64.4 GB | 39.12 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run MiniMax M2.5 BF16 INT4 AWQ?
Q4_K_M · 24.0 GBMiniMax M2.5 BF16 INT4 AWQ (Q4_K_M) requires 24.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 32+ GB is recommended. Using the full 197K context window can add up to 24.7 GB, bringing total usage to 48.7 GB. 1 GPU can run it, including NVIDIA GeForce RTX 5090.
All compatible consumer-level GPUs are running near their VRAM limit. You may also want to consider professional GPUs (e.g., NVIDIA A100, H100) which offer significantly more VRAM. For more headroom and better throughput, consider a multi-GPU configuration with tensor parallelism (supported by tools like vLLM, llama.cpp, or text-generation-inference).
Decent
— Enough VRAM, may be tightWhich Devices Can Run MiniMax M2.5 BF16 INT4 AWQ?
Q4_K_M · 24.0 GB15 devices with unified memory can run MiniMax M2.5 BF16 INT4 AWQ, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio M4 Max (36 GB).
Runs great
— Plenty of headroomDecent
— Enough memory, may be tightRelated Models
Frequently Asked Questions
- How much VRAM does MiniMax M2.5 BF16 INT4 AWQ need?
MiniMax M2.5 BF16 INT4 AWQ requires 24.0 GB of VRAM at Q4_K_M, or 39.7 GB at Q8_0. Full 197K context adds up to 24.7 GB (48.7 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 39.1B × 4.8 bits ÷ 8 = 23.5 GB
KV Cache + Overhead ≈ 0.5 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 25.2 GB (at full 197K context)
VRAM usage by quantization
Q4_K_M24.0 GBQ4_K_M + full context48.7 GB- Can NVIDIA GeForce RTX 4090 run MiniMax M2.5 BF16 INT4 AWQ?
Yes, at IQ4_NL (22.6 GB) or lower. Higher quantizations like Q4_K_M (24.0 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
- What's the best quantization for MiniMax M2.5 BF16 INT4 AWQ?
For MiniMax M2.5 BF16 INT4 AWQ, Q4_K_M (24.0 GB) offers the best balance of quality and VRAM usage. Q5_K_S (27.4 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 11.3 GB.
VRAM requirement by quantization
IQ2_XXS11.3 GB~53%Q3_K_S17.7 GB~77%Q4_122.6 GB~88%Q4_K_M ★24.0 GB~89%Q5_K_S27.4 GB~92%Q8_039.7 GB~99%★ Recommended — best balance of quality and VRAM usage.
- Can I run MiniMax M2.5 BF16 INT4 AWQ on a Mac?
MiniMax M2.5 BF16 INT4 AWQ requires at least 11.3 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run MiniMax M2.5 BF16 INT4 AWQ locally?
Yes — MiniMax M2.5 BF16 INT4 AWQ can run locally on consumer hardware. At Q4_K_M quantization it needs 24.0 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is MiniMax M2.5 BF16 INT4 AWQ?
At Q4_K_M, MiniMax M2.5 BF16 INT4 AWQ can reach ~121 tok/s on AMD Instinct MI300X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 24.0 × 0.55 = ~121 tok/s
Estimated speed at Q4_K_M (24.0 GB)
AMD Instinct MI300X~121 tok/sNVIDIA H100 SXM~91 tok/sAMD Instinct MI250X~75 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of MiniMax M2.5 BF16 INT4 AWQ?
At Q4_K_M, the download is about 23.47 GB. The full-precision Q8_0 version is 39.12 GB. The smallest option (IQ2_XXS) is 10.76 GB.