LL

llava v1.6 34B

LLaVAGGUF by cjpaisUpdated 2024-03-07
Download on Hugging Face
Parameters34.4B
Context4K
Layers60
Architecturellama
Smallest quant14.2 GB
Quick answer

llava v1.6 34B needs about 23.2 GB of memory at Q4_K_M (a 20.7 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 11 quants, from 14.2 GB (Q3_K_XS) to 36.5 GB (Q8_0).

LLaVA v1.6 34B – multimodal instruction‑following model

LLaVA v1.6 34B is a 34.4 billion‑parameter multimodal model based on NousResearch/Nous‑Hermes‑2‑Yi‑34B. It supports vision input and text generation with a 4096‑token context window. The model is released under the apache‑2.0 license.

Good at

  • Vision‑text instruction following
  • Multimodal question answering
  • Image captioning

Limits

  • No vision support in older versions
  • Requires large VRAM for full precision
  • License restricts commercial use

Best for

  • Research on multimodal LLMs
  • Prototyping vision‑enabled chatbots
  • Academic VQA studies

Who it’s for. Researchers and hobbyists working in computer vision, natural language processing, and machine learning. The model is suitable for those who need a multimodal LLM for experimentation and academic projects.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q3_K_XS14.2 GB 15.7 GB
Too big Get file →
Q3_K_XS14.2 GB 15.7 GB
Too big Get file →
Q3_K16.7 GB 18.2 GB
Too big Get file →
Q3_K_M16.7 GB 18.2 GB
Too big Get file →
Q4_K_M20.7 GB 22.2 GB
Too big Get file →
Q5_K_S23.7 GB 25.2 GB
Too big Get file →
Q5_K_S23.7 GB 25.2 GB
Too big Get file →
Q5_K24.3 GB 25.8 GB
Too big Get file →
Q5_K_M24.3 GB 25.8 GB
Too big Get file →
Q6_K28.2 GB 29.7 GB
Too big Get file →
Q8_036.5 GB 38.0 GB
Too big Get file →

For 8 GB VRAM use Q3_K_XS (14.16 GB file, 16.7 GB memory at 8k context). For 16 GB VRAM Q4_K_M (20.66 GB file, 23.2 GB memory). For 24 GB VRAM Q5_K_M (24.32 GB file, 26.8 GB memory).

What GPU do you need for llava v1.6 34B?

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 Q4_K_M (20.7 GB file, about 23.2 GB in use).
  • 32 GB VRAM: the best quant that fits is Q6_K (28.2 GB file, about 30.7 GB in use).
  • 48 GB VRAM: the best quant that fits is Q8_0 (36.5 GB file, about 39.1 GB in use).

Which llava v1.6 34B quant should you choose?

Q4_K_M is the usual starting point for llava v1.6 34B: it balances file size and quality well. Here it is a 20.7 GB file.

The Q8_0 file is 36.5 GB, about 1.8 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

If memory is tight, the smallest file is Q3_K_XS at 14.2 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters34.4B
Layers60
Trained context4,096 tokens
KV cache (fp16)240 KB per token

Source

Quantized bycjpais
FamilyLLaVA
Licenseapache-2.0
Downloads (HF)1,573
Last updated2024-03-07

Frequently asked questions about llava v1.6 34B

How much VRAM does llava v1.6 34B need?

About 23.2 GB at Q4_K_M with an 8K context window. That is the 20.7 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 llava v1.6 34B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 16.7 GB, so you would need a larger GPU or CPU offloading.

What is the file size of llava v1.6 34B GGUF?

It depends on the quant: Q3_K_XS is 14.2 GB, Q3_K_XS is 14.2 GB, Q3_K is 16.7 GB, Q3_K_M is 16.7 GB, Q4_K_M is 20.7 GB, Q5_K_S is 23.7 GB and 5 more.

What context length does llava v1.6 34B support?

The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 2.0 GB at 8K and 1.0 GB at 4K.

Where can I download llava v1.6 34B GGUF?

The files are on Hugging Face in the cjpais/llava-v1.6-34B-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 llava v1.6 34B 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 LLaVA v1.6 34B on 8 GB VRAM?

Use the Q3_K_XS quant, which requires about 16.7 GB of VRAM at 8k context. It is the smallest available quant for this model.

Does this model support image input?

Yes, LLaVA v1.6 34B has vision capabilities and can process images alongside text.

What license applies to this model?

The model is released under the apache‑2.0 license, allowing free use for research and non‑commercial purposes.

Similar models

Browse all →

More from cjpais

All by cjpais →

More LLaVA models

All LLaVA →

Other large models

All large models →