DB

dbrx instruct iMat

DbrxGGUF by dranger003Updated 2024-04-14
Download on Hugging Face
Parameters132B
Context32K
Layers40
Architecturedbrx
Smallest quant26.9 GB
Quick answer

dbrx instruct iMat needs about 55.7 GB of memory at IQ3_XS (a 53.8 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 9 quants, from 26.9 GB (IQ1_S) to 263.2 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 5.4 GB.

DBRX Instruct iMat – 132B parameter MoE LLM

DBRX Instruct iMat is a 132B parameter transformer with a fine‑grained mixture‑of‑experts architecture. It supports a 32768 token context window and is available in multiple GGUF quantizations. The model is released under an other license.

Good at

  • Large context handling
  • Fine‑grained MoE inference
  • High token throughput

Limits

  • No vision support
  • Requires recent llama.cpp commit
  • Large memory for higher quants

Best for

  • Long‑form text generation
  • Complex reasoning tasks
  • Fine‑grained prompt engineering

Who it’s for. Developers needing a large‑scale, context‑rich language model for text‑only workloads. Suitable for research and production environments that can accommodate the memory demands of MoE models.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S26.9 GB 28.7 GB
Too big Get file →
IQ1_M29.8 GB 31.6 GB
Too big Get file →
IQ2_XXS34.6 GB 36.5 GB
Too big Get file →
IQ2_XS38.5 GB 40.4 GB
Too big Get file →
IQ3_XS53.8 GB (2 parts) 55.7 GB
Too big Get file →
IQ4_XS70.1 GB (2 parts) 71.9 GB
Too big Get file →
Q6_K108.0 GB (3 parts) 109.8 GB
Too big Get file →
Q8_0139.8 GB (3 parts) 141.7 GB
Too big Get file →
F16263.2 GB (6 parts) 265.0 GB
Too big Get file →

For 8 GB VRAM, use IQ1_M (29.79 GB file, ~31.6 GB memory at 8 k context). For 16 GB VRAM, IQ2_XS (38.53 GB file, ~40.4 GB memory) is suitable. For 32 GB VRAM, IQ3_XS (53.83 GB file, ~55.7 GB memory) or IQ4_XS (70.1 GB file, ~71.9 GB memory) can be used, noting they are split into shards.

What GPU do you need for dbrx instruct iMat?

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: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: no quant fits at an 8K context.
  • 32 GB VRAM: the best quant that fits is IQ1_M (29.8 GB file, about 31.6 GB in use).
  • 48 GB VRAM: the best quant that fits is IQ2_XS (38.5 GB file, about 40.4 GB in use).

Which dbrx instruct iMat quant should you choose?

IQ3_XS (53.8 GB) is the middle option in this repository. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

If memory is tight, the smallest file is IQ1_S at 26.9 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

Architecturedbrx
Parameters132B (approximate, mixture of experts)
Layers40
Trained context32,768 tokens
KV cache (fp16)160 KB per token

Source

Quantized bydranger003
FamilyDbrx
Licenseother
Downloads (HF)779
Last updated2024-04-14

Frequently asked questions about dbrx instruct iMat

How much VRAM does dbrx instruct iMat need?

About 55.7 GB at IQ3_XS with an 8K context window. That is the 53.8 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 dbrx instruct iMat on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 28.7 GB, so you would need a larger GPU or CPU offloading.

What is the file size of dbrx instruct iMat GGUF?

It depends on the quant: IQ1_S is 26.9 GB, IQ1_M is 29.8 GB, IQ2_XXS is 34.6 GB, IQ2_XS is 38.5 GB, IQ3_XS is 53.8 GB, IQ4_XS is 70.1 GB and 3 more.

What context length does dbrx instruct iMat support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.3 GB at 8K and 5.4 GB at 32K.

Where can I download dbrx instruct iMat GGUF?

The files are on Hugging Face in the dranger003/dbrx-instruct-iMat.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 dbrx instruct iMat use?

Hugging Face lists the license as other. 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 IQ1_S on 8 GB VRAM?

IQ1_S is unstable; use IQ1_M instead for 8 GB VRAM.

Do I need the llama.cpp commit 4bd0f93e?

Yes, the latest commit is required to run this model.

Is the model vision capable?

No, the model has no vision support.

Other GGUFs of dbrx-instruct

Similar models

Browse all →

More from dranger003

All by dranger003 →

More Dbrx models

All Dbrx →

Other large models

All large models →