DO

2x_bagel 34b V0.2_Ultra_Fp32

DoublebagelGGUF by mradermacherUpdated 2024-12-16
Download on Hugging Face
Parameters56.7B
Context195K
Layers100
Architecturellama
Smallest quant21.0 GB
Quick answer

2x_bagel 34b V0.2_Ultra_Fp32 needs about 37.9 GB of memory at Q4_K_M (a 34.0 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 13 quants, from 21.0 GB (Q2_K) to 46.5 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 13.4 GB.

2x_bagel 34b V0.2_Ultra_Fp32 is a 56.7B-parameter Doublebagel model in GGUF format, quantized by mradermacher. Its trained context length is 195K tokens. The repository has 13 quant files, from 21.0 GB (Q2_K) to 46.5 GB (Q6_K).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K21.0 GB 24.9 GB
Too big Get file →
IQ3_XS23.3 GB 27.2 GB
Too big Get file →
Q3_K_S24.6 GB 28.4 GB
Too big Get file →
IQ3_S24.6 GB 28.5 GB
Too big Get file →
IQ3_M25.6 GB 29.4 GB
Too big Get file →
Q3_K_M27.4 GB 31.2 GB
Too big Get file →
Q3_K_L29.8 GB 33.7 GB
Too big Get file →
IQ4_XS30.7 GB 34.5 GB
Too big Get file →
Q4_K_S32.2 GB 36.1 GB
Too big Get file →
Q4_K_M34.0 GB 37.9 GB
Too big Get file →
Q5_K_S39.1 GB 42.9 GB
Too big Get file →
Q5_K_M40.1 GB 43.9 GB
Too big Get file →
Q6_K46.5 GB 50.4 GB
Too big Get file →

What GPU do you need for 2x_bagel 34b V0.2_Ultra_Fp32?

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 Q3_K_M (27.4 GB file, about 31.2 GB in use).
  • 48 GB VRAM: the best quant that fits is Q5_K_M (40.1 GB file, about 43.9 GB in use).

Which 2x_bagel 34b V0.2_Ultra_Fp32 quant should you choose?

Q4_K_M is the usual starting point for 2x_bagel 34b V0.2_Ultra_Fp32: it balances file size and quality well. Here it is a 34.0 GB file.

If memory is tight, the smallest file is Q2_K at 21.0 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
Parameters56.7B
Layers100
Trained context200,000 tokens
KV cache (fp16)400 KB per token

Source

Quantized bymradermacher
Licenseapache-2.0
Downloads (HF)108
Last updated2024-12-16

Frequently asked questions about 2x_bagel 34b V0.2_Ultra_Fp32

How much VRAM does 2x_bagel 34b V0.2_Ultra_Fp32 need?

About 37.9 GB at Q4_K_M with an 8K context window. That is the 34.0 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 2x_bagel 34b V0.2_Ultra_Fp32 on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 24.9 GB, so you would need a larger GPU or CPU offloading.

What is the file size of 2x_bagel 34b V0.2_Ultra_Fp32 GGUF?

It depends on the quant: Q2_K is 21.0 GB, IQ3_XS is 23.3 GB, Q3_K_S is 24.6 GB, IQ3_S is 24.6 GB, IQ3_M is 25.6 GB, Q3_K_M is 27.4 GB and 7 more.

What context length does 2x_bagel 34b V0.2_Ultra_Fp32 support?

The GGUF header lists a trained context length of 200,000 tokens (195K). The KV cache grows with the context you use: about 3.4 GB at 8K and 13.4 GB at 32K.

Where can I download 2x_bagel 34b V0.2_Ultra_Fp32 GGUF?

The files are on Hugging Face in the mradermacher/2xbagel-34b-v0.2-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 2x_bagel 34b V0.2_Ultra_Fp32 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.

Other GGUFs of DoubleBagel-57B-v1.0

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Doublebagel models

All Doublebagel →

Other large models

All large models →