Mistral-Small-Instruct-2409 GGUF size and VRAM requirements

General Purpose LLM License: other ⬇ 4,909 ❤ 53
Parameters22.25B
Context131,072

Mistral-Small-Instruct-2409 is a 22.25 billion parameter AI model designed for instruction-following tasks, part of the Mistral family developed by Mistral AI. It is optimized for general-purpose language understanding and generation, with a focus on structured interactions and user-directed queries. The model is available via Hugging Face and compatible with tools like LM Studio for deployment.

Under the hood it uses 56 transformer layers, a hidden size of 6,144, 48 attention heads. It uses grouped-query attention (48 query heads sharing 8 key/value heads), which already trims KV-cache memory compared with full multi-head attention.

To run bartowski/Mistral-Small-Instruct-2409-GGUF locally at a 4,096-token context, its quantized versions need between 7.26 GB (IQ2_XXS, lowest quality) and 43.12 GB (F16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for Mistral-Small-Instruct-2409-GGUF is about 0.88 GB, rising to roughly 28.0 GB at its full 131,072-token context. Shorter prompts free up memory for a higher-quality quantization.

For most users the best balance is IQ2_XS, needing about 7.86 GB. That means bartowski/Mistral-Small-Instruct-2409-GGUF fits entirely in the VRAM of an 8 GB GPU or larger, running fully on the GPU.

Available GGUF quantizations for bartowski/Mistral-Small-Instruct-2409-GGUF include IQ2_XXS, IQ2_XS, IQ2_M, Q2_K, Q2_K_L, IQ3_XS, Q3_K_S, IQ3_M, Q3_K_M, Q3_K_L, Q3_K_XL, IQ4_XS, Q4_0_4_4, Q4_0_4_8, Q4_0_8_8, Q4_0, Q4_K_S, Q4_K_M, Q4_K_L, Q5_K_S, Q5_K_M, Q5_K_L, Q6_K, Q6_K_L, Q8_0, F16. The model supports a native context length of up to 131,072 tokens; a longer context grows the KV cache and the memory needed.

→ Guide: How much VRAM do you need?

General Purpose LLM

Recommended Use Cases & Local Setup

General Purpose & Instruction

Ideal for: General knowledge, document summarization, email drafting, language translation, and multi-turn conversational chat.

Ollama

Simple one-line CLI installation running silently in the background

LM Studio

Polished desktop client with one-click model downloads and GPU offloading

Jan.ai

Open-source privacy-focused desktop assistant

Prompting & Sampling Tip: Runs optimally with standard chat templates (ChatML or Llama 3 format). A context window of 4,096 to 8,192 tokens balances memory usage and conversational memory.

GGUF file size and memory by quantization

Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.

Quant.Bits QualityWeights KVTotal Speed~Verdict
IQ2_XXS 2.16 Very low 5.58 GB 0.88 GB 7.26 GB 71.6 t/s Fits in VRAM
IQ2_XS 2.39 Very low 6.19 GB 0.88 GB 7.86 GB 64.6 t/s Fits in VRAM
IQ2_M 2.74 Low 7.1 GB 0.88 GB 8.77 GB 7.0 t/s Offload
Q2_K 2.97 Low 7.7 GB 0.88 GB 9.38 GB 6.5 t/s Offload
Q2_K_L 3.05 Low 7.89 GB 0.88 GB 9.56 GB 6.3 t/s Offload
IQ3_XS 3.3 Low 8.55 GB 0.88 GB 10.22 GB 5.9 t/s Offload
Q3_K_S 3.47 Fair 8.98 GB 0.88 GB 10.65 GB 5.6 t/s Offload
IQ3_M 3.62 Fair 9.37 GB 0.88 GB 11.05 GB 5.3 t/s Offload
Q3_K_M 3.87 Fair 10.02 GB 0.88 GB 11.69 GB 5.0 t/s Offload
Q3_K_L 4.22 Good 10.92 GB 0.88 GB 12.6 GB 4.6 t/s Offload
Q3_K_XL 4.28 Good 11.09 GB 0.88 GB 12.76 GB 4.5 t/s Offload
IQ4_XS 4.29 Good 11.12 GB 0.88 GB 12.79 GB 4.5 t/s Offload
Q4_0_4_4 4.52 Good 11.71 GB 0.88 GB 13.38 GB 4.3 t/s Offload
Q4_0_4_8 4.52 Good 11.71 GB 0.88 GB 13.38 GB 4.3 t/s Offload
Q4_0_8_8 4.52 Good 11.71 GB 0.88 GB 13.38 GB 4.3 t/s Offload
Q4_0 4.54 Good 11.75 GB 0.88 GB 13.42 GB 4.3 t/s Offload
Q4_K_S 4.55 Good 11.79 GB 0.88 GB 13.47 GB 4.2 t/s Offload
Q4_K_M 4.8 Good 12.43 GB 0.88 GB 14.1 GB 4.0 t/s Offload
Q4_K_L 4.85 Good 12.56 GB 0.88 GB 14.24 GB 4.0 t/s Offload
Q5_K_S 5.51 Very good 14.27 GB 0.88 GB 15.95 GB 3.5 t/s Offload
Q5_K_M 5.65 Very good 14.64 GB 0.88 GB 16.32 GB 3.4 t/s Offload
Q5_K_L 5.7 Very good 14.76 GB 0.88 GB 16.43 GB 3.4 t/s Offload
Q6_K 6.56 Excellent 17.0 GB 0.88 GB 18.67 GB 2.9 t/s Offload
Q6_K_L 6.6 Excellent 17.09 GB 0.88 GB 18.76 GB 2.9 t/s Offload
Q8_0 8.5 Excellent 22.02 GB 0.88 GB 23.69 GB 2.3 t/s Offload
F16 16.0 Excellent 41.44 GB 0.88 GB 43.12 GB Insufficient

KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.

Frequently asked questions

What kind of model is bartowski/Mistral-Small-Instruct-2409-GGUF?

bartowski/Mistral-Small-Instruct-2409-GGUF is an instruction-tuned chat model with 22.25 billion parameters, based on the llama architecture. It is released under the other license and distributed as GGUF files for local inference.

How much VRAM do you need to run bartowski/Mistral-Small-Instruct-2409-GGUF?

You need about 7.86 GB of VRAM to run bartowski/Mistral-Small-Instruct-2409-GGUF entirely on the GPU using the IQ2_XS quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run bartowski/Mistral-Small-Instruct-2409-GGUF on an 8 GB GPU?

Yes. With 8 GB of VRAM you can run bartowski/Mistral-Small-Instruct-2409-GGUF fully on the GPU using IQ2_XS (about 7.86 GB).

Can I run bartowski/Mistral-Small-Instruct-2409-GGUF on a 16 GB GPU?

Yes. With 16 GB of VRAM you can run bartowski/Mistral-Small-Instruct-2409-GGUF fully on the GPU using Q5_K_S (about 15.95 GB).

Can I run bartowski/Mistral-Small-Instruct-2409-GGUF on a 24 GB GPU?

Yes. With 24 GB of VRAM you can run bartowski/Mistral-Small-Instruct-2409-GGUF fully on the GPU using Q8_0 (about 23.69 GB).

What context length does bartowski/Mistral-Small-Instruct-2409-GGUF support?

bartowski/Mistral-Small-Instruct-2409-GGUF supports a native context length of up to 131,072 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.

What is the best quantization for bartowski/Mistral-Small-Instruct-2409-GGUF?

For bartowski/Mistral-Small-Instruct-2409-GGUF, a strong default is Q4_K_M, which needs about 14.1 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.