BI

bitnet_b1_58-3B

BitnetGGUF by RichardErkhovUpdated 2024-07-13
Download on Hugging Face
Parameters3.32B
Context2K
Layers26
Architecturebitnet
Smallest quant1.9 GB
Quick answer

bitnet_b1_58-3B needs about 5.7 GB of memory at Q4_K_M (a 2.5 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 22 quants, from 1.9 GB (IQ3_S) to 3.5 GB (Q8_0).

bitnet_b1_58-3B is a 3.32B-parameter Bitnet model in GGUF format, quantized by RichardErkhov. Its trained context length is 2K tokens. The repository has 22 quant files, from 1.9 GB (IQ3_S) to 3.5 GB (Q8_0).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ3_S1.9 GB 3.1 GB
Fits Get file →
IQ3_XS1.9 GB 3.1 GB
Fits Get file →
Q2_K1.9 GB 3.1 GB
Fits Get file →
Q3_K_S1.9 GB 3.1 GB
Fits Get file →
Q4_01.9 GB 3.1 GB
Fits Get file →
IQ4_NL1.9 GB 3.1 GB
Fits Get file →
IQ4_XS1.9 GB 3.1 GB
Fits Get file →
IQ3_M2.0 GB 3.2 GB
Fits Get file →
Q3_K2.1 GB 3.3 GB
Fits Get file →
Q3_K_M2.1 GB 3.3 GB
Fits Get file →
Q4_12.1 GB 3.3 GB
Fits Get file →
Q3_K_L2.2 GB 3.3 GB
Fits Get file →
Q5_02.3 GB 3.5 GB
Fits Get file →
Q4_K_S2.3 GB 3.5 GB
Fits Get file →
Q4_K2.5 GB 3.7 GB
Fits Get file →
Q4_K_M2.5 GB 3.7 GB
Fits Get file →
Q5_12.5 GB 3.7 GB
Fits Get file →
Q5_K_S2.5 GB 3.7 GB
Fits Get file →
Q5_K2.7 GB 3.9 GB
Fits Get file →
Q5_K_M2.7 GB 3.9 GB
Fits Get file →
Q6_K3.5 GB 4.7 GB
Fits Get file →
Q8_03.5 GB 4.7 GB
Fits Get file →

What GPU do you need for bitnet_b1_58-3B?

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: the best quant that fits is Q5_K (2.7 GB file, about 5.9 GB in use).
  • 8 GB VRAM: the best quant that fits is Q6_K (3.5 GB file, about 6.8 GB in use).

Which bitnet_b1_58-3B quant should you choose?

Q4_K_M is the usual starting point for bitnet_b1_58-3B: it balances file size and quality well. Here it is a 2.5 GB file.

The Q8_0 file is 3.5 GB, about 1.4 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 IQ3_S at 1.9 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

Architecturebitnet
Parameters3.32B
Layers26
Trained context2,048 tokens
KV cache (fp16)325 KB per token

Source

Quantized byRichardErkhov
FamilyBitnet
Downloads (HF)347
Last updated2024-07-13

Frequently asked questions about bitnet_b1_58-3B

How much VRAM does bitnet_b1_58-3B need?

About 5.7 GB at Q4_K_M with an 8K context window. That is the 2.5 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 bitnet_b1_58-3B on an 8 GB GPU?

Yes, with Q6_K: the 3.5 GB file needs about 6.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of bitnet_b1_58-3B GGUF?

It depends on the quant: IQ3_S is 1.9 GB, IQ3_XS is 1.9 GB, Q2_K is 1.9 GB, Q3_K_S is 1.9 GB, Q4_0 is 1.9 GB, IQ4_NL is 1.9 GB and 16 more.

What context length does bitnet_b1_58-3B support?

The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 2.7 GB at 8K and 0.7 GB at 2K.

Where can I download bitnet_b1_58-3B GGUF?

The files are on Hugging Face in the RichardErkhov/1bitLLM_-_bitnet_b1_58-3B-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Similar models

Browse all →

More from RichardErkhov

All by RichardErkhov →

More Bitnet models

All Bitnet →

Other tiny models

All tiny models →