TI

TinyMistral 248M

TinymistralGGUF by mradermacherUpdated 2024-12-11
Download on Hugging Face
Parameters248M
Context32K
Layers12
Architecturellama
Smallest quant0.1 GB
Quick answer

TinyMistral 248M needs about 0.8 GB of memory at Q4_K_M (a 0.2 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 25 quants, from 0.1 GB (IQ1_S) to 0.2 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 0.4 GB.

TinyMistral 248M – a 248 M parameter llama‑style model

TinyMistral 248M is a 248 M parameter language model based on Mistral 7B. It supports a context length of 32,768 tokens and is intended for fine‑tuning on downstream tasks. The model is released under the Apache‑2.0 license.

Good at

  • Small footprint inference
  • Fine‑tuning on custom data
  • Fast token generation

Limits

  • No vision support
  • Requires 8 k context for memory estimates
  • Not intended for direct use

Best for

  • Rapid prototyping
  • Edge device inference
  • Low‑resource research

Who it’s for. Developers and researchers who need a lightweight llama‑style model for fine‑tuning or quick experimentation. The model is suitable for projects that require a small parameter count and a large context window.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S0.1 GB 0.7 GB
Fits Get file →
IQ1_M0.1 GB 0.7 GB
Fits Get file →
IQ2_XXS0.1 GB 0.7 GB
Fits Get file →
IQ2_XS0.1 GB 0.7 GB
Fits Get file →
IQ2_S0.1 GB 0.7 GB
Fits Get file →
IQ2_M0.1 GB 0.7 GB
Fits Get file →
Q2_K_S0.1 GB 0.7 GB
Fits Get file →
Q2_K0.1 GB 0.7 GB
Fits Get file →
IQ3_XXS0.1 GB 0.7 GB
Fits Get file →
IQ3_XS0.1 GB 0.7 GB
Fits Get file →
Q3_K_S0.1 GB 0.7 GB
Fits Get file →
IQ3_S0.1 GB 0.7 GB
Fits Get file →
IQ3_M0.1 GB 0.7 GB
Fits Get file →
Q3_K_M0.1 GB 0.7 GB
Fits Get file →
Q3_K_L0.1 GB 0.7 GB
Fits Get file →
IQ4_XS0.1 GB 0.7 GB
Fits Get file →
Q4_00.1 GB 0.7 GB
Fits Get file →
Q4_00.1 GB 0.7 GB
Fits Get file →
Q4_00.1 GB 0.7 GB
Fits Get file →
Q4_00.1 GB 0.7 GB
Fits Get file →
Q4_K_S0.1 GB 0.8 GB
Fits Get file →
Q4_K_M0.2 GB 0.8 GB
Fits Get file →
Q5_K_S0.2 GB 0.8 GB
Fits Get file →
Q5_K_M0.2 GB 0.8 GB
Fits Get file →
Q6_K0.2 GB 0.8 GB
Fits Get file →

For 8 GB VRAM, use IQ1_S or IQ1_M (file size 0.07 GB or 0.08 GB, memory ~0.7 GB). For 4 GB VRAM, use IQ2_XXS or IQ2_XS (0.08 GB or 0.09 GB). For 2 GB VRAM, use IQ2_S (0.09 GB). Memory estimates are approximate.

What GPU do you need for TinyMistral 248M?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: the best quant that fits is Q6_K (0.2 GB file, about 0.8 GB in use).

Which TinyMistral 248M quant should you choose?

Q4_K_M is the usual starting point for TinyMistral 248M: it balances file size and quality well. Here it is a 0.2 GB file.

If memory is tight, the smallest file is IQ1_S at 0.1 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters248M
Layers12
Trained context32,768 tokens
KV cache (fp16)12 KB per token

Source

Quantized bymradermacher
Licenseapache-2.0
Downloads (HF)686
Last updated2024-12-11

Frequently asked questions about TinyMistral 248M

How much VRAM does TinyMistral 248M need?

About 0.8 GB at Q4_K_M with an 8K context window. That is the 0.2 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run TinyMistral 248M on an 8 GB GPU?

Yes, with Q6_K: the 0.2 GB file needs about 0.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of TinyMistral 248M GGUF?

It depends on the quant: IQ1_S is 0.1 GB, IQ1_M is 0.1 GB, IQ2_XXS is 0.1 GB, IQ2_XS is 0.1 GB, IQ2_S is 0.1 GB, IQ2_M is 0.1 GB and 19 more.

What context length does TinyMistral 248M support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 0.1 GB at 8K and 0.4 GB at 32K.

Where can I download TinyMistral 248M GGUF?

The files are on Hugging Face in the mradermacher/TinyMistral-248M-i1-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does TinyMistral 248M use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run TinyMistral 248M on 8 GB VRAM?

Yes, use IQ1_S or IQ1_M quantizations; they require about 0.7 GB of VRAM at 8 k context.

Is TinyMistral 248M suitable for fine‑tuning?

Yes, the model is designed for fine‑tuning on downstream tasks and is not intended for direct use.

Other GGUFs of TinyMistral-248M

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Tinymistral models

All Tinymistral →

Other tiny models

All tiny models →