CA

calme-3.2-instruct-78b

CalmeGGUF by mradermacherUpdated 2024-11-30
Download on Hugging Face
Parameters78B
Context32K
Layers86
Architectureqwen2
Smallest quant24.3 GB
Quick answer

calme-3.2-instruct-78b needs about 38.5 GB of memory at IQ3_XS (a 35.1 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 17 quants, from 24.3 GB (IQ1_S) to 46.9 GB (Q4_K_S). Longer context needs more: at 32K tokens the KV cache alone adds about 11.5 GB.

Calme 3.2 Instruct 78B – a large Qwen2-based instruction model

Calme 3.2 Instruct 78B is a 78‑billion parameter model fine‑tuned from Qwen2.5‑72B. It supports a 32768 token context window and is available in multiple GGUF quantizations. The model is experimental and may be sensitive to hyper‑parameters.

Good at

  • Large context handling
  • Instruction following
  • Generic domain tasks

Limits

  • Experimental model, performance may vary
  • No vision support
  • License type is other

Best for

  • Chatbot development
  • Content generation
  • Research prototyping

Who it’s for. Developers and researchers who need a large instruction‑tuned model and are comfortable experimenting with experimental releases. The model is suitable for those who can adjust hyper‑parameters to optimize performance.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S24.3 GB 27.7 GB
Too big Get file →
IQ1_M25.4 GB 28.8 GB
Too big Get file →
IQ2_XXS27.3 GB 30.7 GB
Too big Get file →
IQ2_XS29.0 GB 32.4 GB
Too big Get file →
IQ2_S29.9 GB 33.3 GB
Too big Get file →
IQ2_M31.4 GB 34.8 GB
Too big Get file →
Q2_K31.8 GB 35.2 GB
Too big Get file →
IQ3_XXS34.0 GB 37.4 GB
Too big Get file →
IQ3_XS35.1 GB 38.5 GB
Too big Get file →
Q3_K_S36.8 GB 40.2 GB
Too big Get file →
IQ3_S36.9 GB 40.2 GB
Too big Get file →
IQ3_M37.9 GB 41.3 GB
Too big Get file →
Q3_K_M40.3 GB 43.7 GB
Too big Get file →
Q3_K_L42.3 GB 45.7 GB
Too big Get file →
IQ4_XS42.6 GB 45.9 GB
Too big Get file →
Q4_044.3 GB 47.7 GB
Too big Get file →
Q4_K_S46.9 GB 50.3 GB
Too big Get file →

For 8‑GB VRAM, use IQ1_S (24.29 GB file, ~27.7 GB memory at 8k context). For 16‑GB VRAM, IQ2_M (31.43 GB file, ~34.8 GB memory) or Q2_K (31.85 GB file, ~35.2 GB memory). Larger VRAM allows higher‑quality quants like Q4_K_S (46.95 GB file, ~50.3 GB memory).

What GPU do you need for calme-3.2-instruct-78b?

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 IQ2_XXS (27.3 GB file, about 30.7 GB in use).
  • 48 GB VRAM: the best quant that fits is Q4_0 (44.3 GB file, about 47.7 GB in use).

Which calme-3.2-instruct-78b quant should you choose?

IQ3_XS (35.1 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 24.3 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

Architectureqwen2
Parameters78B
Layers86
Trained context32,768 tokens
KV cache (fp16)344 KB per token

Source

Quantized bymradermacher
FamilyCalme
Licenseother
Downloads (HF)558
Last updated2024-11-30

Frequently asked questions about calme-3.2-instruct-78b

How much VRAM does calme-3.2-instruct-78b need?

About 38.5 GB at IQ3_XS with an 8K context window. That is the 35.1 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 calme-3.2-instruct-78b on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 27.7 GB, so you would need a larger GPU or CPU offloading.

What is the file size of calme-3.2-instruct-78b GGUF?

It depends on the quant: IQ1_S is 24.3 GB, IQ1_M is 25.4 GB, IQ2_XXS is 27.3 GB, IQ2_XS is 29.0 GB, IQ2_S is 29.9 GB, IQ2_M is 31.4 GB and 11 more.

What context length does calme-3.2-instruct-78b support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 2.9 GB at 8K and 11.5 GB at 32K.

Where can I download calme-3.2-instruct-78b GGUF?

The files are on Hugging Face in the mradermacher/calme-3.2-instruct-78b-i1-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 calme-3.2-instruct-78b 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 this model on 8 GB VRAM?

Use the IQ1_S quant; it requires about 27.7 GB memory at 8k context, which exceeds 8 GB VRAM. The model cannot run on 8 GB VRAM without further optimization.

Does this model support vision tasks?

No, the model has no vision capabilities.

What prompt template does it use?

It uses the ChatML prompt template.

Other GGUFs of calme-3.2-instruct-78b

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Calme models

All Calme →

Other large models

All large models →