BA

jondurbin_bagel-dpo-34b-v0.2

BagelGGUF by TheBlokeUpdated 2024-01-02
Download on Hugging Face
Parameters34.4B
Context195K
Layers60
Architecturellama
Smallest quant14.6 GB
Quick answer

jondurbin_bagel-dpo-34b-v0.2 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 12 quants, from 14.6 GB (Q2_K) to 36.5 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 8.1 GB.

Bagel DPO 34B v0.2 – a 34.4B parameter Llama model fine‑tuned with DPO

Bagel DPO 34B v0.2 is a 34.4B parameter Llama model fine‑tuned on the Bagel dataset with a toxic DPO dataset. It is available in GGUF format and supports coding and roleplay tasks. The model was quantised by TheBloke using Massed Compute hardware.

Good at

  • Coding tasks
  • Roleplay scenarios
  • Instruction following

Limits

  • No vision support
  • Large context requires high VRAM
  • License is other

Best for

  • Local code generation
  • Interactive storytelling
  • Custom instruction experiments

Who it’s for. Developers and researchers who need a large Llama model with reduced censorship and strong coding or roleplay abilities. The model is suitable for local deployment via llama.cpp or text‑generation‑webui.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K14.6 GB 17.1 GB
Too big Get file →
Q3_K_S15.0 GB 17.5 GB
Too big Get file →
Q3_K_M16.6 GB 19.2 GB
Too big Get file →
Q3_K_L18.1 GB 20.7 GB
Too big Get file →
Q4_019.5 GB 22.0 GB
Too big Get file →
Q4_K_S19.5 GB 22.1 GB
Too big Get file →
Q4_K_M20.7 GB 23.2 GB
Too big Get file →
Q5_023.7 GB 26.2 GB
Too big Get file →
Q5_K_S23.7 GB 26.2 GB
Too big Get file →
Q5_K_M24.3 GB 26.8 GB
Too big Get file →
Q6_K28.2 GB 30.7 GB
Too big Get file →
Q8_036.5 GB 39.1 GB
Too big Get file →

For 8 GB VRAM use Q2_K (file size 14.56 GB, estimated 17.1 GB memory at 8 k context). For 12 GB VRAM use Q3_K_S (14.96 GB, 17.5 GB memory). For 24 GB VRAM use Q4_K_M (20.66 GB, 23.2 GB memory).

What GPU do you need for jondurbin_bagel-dpo-34b-v0.2?

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 jondurbin_bagel-dpo-34b-v0.2 quant should you choose?

Q4_K_M is the usual starting point for jondurbin_bagel-dpo-34b-v0.2: 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 Q2_K at 14.6 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
Parameters34.4B
Layers60
Trained context200,000 tokens
KV cache (fp16)240 KB per token

Source

Quantized byTheBloke
FamilyBagel
Licenseother
Downloads (HF)546
Last updated2024-01-02

Frequently asked questions about jondurbin_bagel-dpo-34b-v0.2

How much VRAM does jondurbin_bagel-dpo-34b-v0.2 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 jondurbin_bagel-dpo-34b-v0.2 on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 17.1 GB, so you would need a larger GPU or CPU offloading.

What is the file size of jondurbin_bagel-dpo-34b-v0.2 GGUF?

It depends on the quant: Q2_K is 14.6 GB, Q3_K_S is 15.0 GB, Q3_K_M is 16.6 GB, Q3_K_L is 18.1 GB, Q4_0 is 19.5 GB, Q4_K_S is 19.5 GB and 6 more.

What context length does jondurbin_bagel-dpo-34b-v0.2 support?

The GGUF header lists a trained context length of 200,000 tokens (195K). The KV cache grows with the context you use: about 2.0 GB at 8K and 8.1 GB at 32K.

Where can I download jondurbin_bagel-dpo-34b-v0.2 GGUF?

The files are on Hugging Face in the TheBloke/bagel-dpo-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 jondurbin_bagel-dpo-34b-v0.2 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 Bagel DPO 34B v0.2 on 8 GB VRAM?

Use the Q2_K quantised file; it requires about 17.1 GB memory at 8 k context, which fits in 8 GB VRAM with swap.

Does this model support image input?

No, Bagel DPO 34B v0.2 does not support vision or image input.

What is the context length of this model?

The model was trained with a context length of 200 000 tokens.

Other GGUFs of bagel-dpo-34b-v0.2

Similar models

Browse all →

More from TheBloke

All by TheBloke →

More Bagel models

All Bagel →

Other large models

All large models →