AY

aya-23-35B

AyaGGUF by legraphistaUpdated 2024-05-26
Download on Hugging Face
Parameters35B
Context8K
Layers40
Architecturecommand-r
Smallest quant8.5 GB
Quick answer

aya-23-35B needs about 27.1 GB of memory at Q3_K_S (a 15.9 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 24 quants, from 8.5 GB (IQ1_S) to 70.0 GB.

Aya‑23‑35B quantized for efficient local inference

Aya‑23‑35B is a 35 billion‑parameter command‑r model quantized with IMatrix. It supports a 8192‑token context window and is released under cc‑by‑nc‑4.0. The model is available in multiple GGUF quantizations ranging from IQ1_S to Q8_0.

Good at

  • Large‑context reasoning
  • Multilingual text generation
  • Fine‑tuned prompt handling

Limits

  • No vision support
  • License restricts commercial use
  • Large memory requirement for high‑quality quants

Best for

  • Chatbots on 8‑GB VRAM
  • Research prototyping with 16‑GB VRAM
  • Low‑latency inference on 32‑GB VRAM

Who it’s for. Developers and researchers who need a 35 billion‑parameter model for local deployment. The model is suitable for non‑commercial projects and academic use.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S8.5 GB 19.8 GB
Too big Get file →
IQ1_M9.1 GB 20.4 GB
Too big Get file →
IQ2_XXS10.2 GB 21.4 GB
Too big Get file →
IQ2_XS11.1 GB 22.3 GB
Too big Get file →
IQ2_S11.8 GB 23.1 GB
Too big Get file →
IQ2_M12.7 GB 23.9 GB
Too big Get file →
Q2_K_S12.7 GB 24.0 GB
Too big Get file →
Q2_K13.8 GB 25.1 GB
Too big Get file →
IQ3_XXS13.8 GB 25.1 GB
Too big Get file →
IQ3_XS15.1 GB 26.3 GB
Too big Get file →
IQ3_S15.9 GB 27.1 GB
Too big Get file →
Q3_K_S15.9 GB 27.1 GB
Too big Get file →
IQ3_M16.7 GB 27.9 GB
Too big Get file →
Q3_K17.6 GB 28.9 GB
Too big Get file →
Q3_K_L19.1 GB 30.4 GB
Too big Get file →
IQ4_XS19.2 GB 30.4 GB
Too big Get file →
IQ4_NL20.2 GB 31.5 GB
Too big Get file →
Q4_K_S20.4 GB 31.6 GB
Too big Get file →
Q4_K21.5 GB 32.8 GB
Too big Get file →
Q5_K_S24.3 GB 35.6 GB
Too big Get file →
Q5_K25.0 GB 36.2 GB
Too big Get file →
Q6_K28.7 GB 39.9 GB
Too big Get file →
Q8_037.2 GB (2 parts) 48.4 GB
Too big Get file →
—70.0 GB (3 parts) 81.2 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (8.52 GB file, ~19.8 GB memory). For 16 GB VRAM, IQ2_S (11.84 GB file, ~23.1 GB memory) or Q2_K_S (12.74 GB file, ~24 GB memory). For 32 GB VRAM, Q4_K (21.53 GB file, ~32.8 GB memory) or Q5_K_S (24.34 GB file, ~35.6 GB memory).

What GPU do you need for aya-23-35B?

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: the best quant that fits is Q2_K_S (12.7 GB file, about 24.0 GB in use).
  • 32 GB VRAM: the best quant that fits is Q4_K_S (20.4 GB file, about 31.6 GB in use).
  • 48 GB VRAM: the best quant that fits is Q6_K (28.7 GB file, about 39.9 GB in use).

Which aya-23-35B quant should you choose?

Q3_K_S (15.9 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 8.5 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

Architecturecommand-r
Parameters35B
Layers40
Trained context8,192 tokens
KV cache (fp16)1280 KB per token

Source

Quantized bylegraphista
FamilyAya
Downloads (HF)1,406
Last updated2024-05-26

Frequently asked questions about aya-23-35B

How much VRAM does aya-23-35B need?

About 27.1 GB at Q3_K_S with an 8K context window. That is the 15.9 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 aya-23-35B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 19.8 GB, so you would need a larger GPU or CPU offloading.

What is the file size of aya-23-35B GGUF?

It depends on the quant: IQ1_S is 8.5 GB, IQ1_M is 9.1 GB, IQ2_XXS is 10.2 GB, IQ2_XS is 11.1 GB, IQ2_S is 11.8 GB, IQ2_M is 12.7 GB and 18 more.

What context length does aya-23-35B support?

The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 10.7 GB at 8K and 10.7 GB at 8K.

Where can I download aya-23-35B GGUF?

The files are on Hugging Face in the legraphista/aya-23-35B-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 aya-23-35B use?

Hugging Face lists the license as cc-by-nc-4.0. 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 Aya‑23‑35B on 8 GB VRAM?

Use the IQ1_S quantization; it requires about 19.8 GB of VRAM at 8k context, so 8 GB VRAM is insufficient.

Is the model suitable for commercial applications?

The cc‑by‑nc‑4.0 license prohibits commercial use without permission.

How many files do I need to download for the Q8_0 quant?

The Q8_0 quant is split into 2 shards; download both to use the full model.

Other GGUFs of aya-23-35B

Similar models

Browse all →

More from legraphista

All by legraphista →

More Aya models

All Aya →

Other large models

All large models →