CO

Cosmos 37b BASE i1

CosmosGGUF by mradermacherUpdated 2024-12-15
Download on Hugging Face
Parameters37.5B
Context128K
Layers95
Architecturellama
Smallest quant8.0 GB
Quick answer

Cosmos 37b BASE i1 needs about 26.1 GB of memory at Q4_K_M (a 22.4 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 22 quants, from 8.0 GB (IQ1_S) to 30.7 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 12.8 GB.

Cosmos 37b BASE i1 – 37.5B llama model with tool‑usage support

Cosmos 37b BASE i1 is a 37.5B parameter llama model. It is trained on a 131072 token context and supports tool usage. The model is available in multiple GGUF quantizations.

Good at

  • Tool integration
  • Creative writing
  • Large context handling

Limits

  • No instruction tuning – use -INSTRUCT for chat
  • No vision support
  • License not specified

Best for

  • Developers building tool‑enabled assistants
  • Researchers testing large context models
  • Creative writers needing long‑form generation

Who it’s for. This model is suitable for developers and researchers who need a large llama model with tool‑usage capability and a long context window. It is best used when instruction tuning is not required and the user can provide the necessary context.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S8.0 GB 11.7 GB
Too big Get file →
IQ1_M8.8 GB 12.5 GB
Too big Get file →
IQ2_XXS10.0 GB 13.7 GB
Too big Get file →
IQ2_XS11.1 GB 14.8 GB
Too big Get file →
IQ2_S11.8 GB 15.5 GB
Too big Get file →
IQ2_M12.8 GB 16.4 GB
Too big Get file →
Q2_K_S12.8 GB 16.5 GB
Too big Get file →
Q2_K13.9 GB 17.6 GB
Too big Get file →
IQ3_XXS14.4 GB 18.1 GB
Too big Get file →
IQ3_XS15.4 GB 19.1 GB
Too big Get file →
Q3_K_S16.2 GB 19.9 GB
Too big Get file →
IQ3_S16.3 GB 19.9 GB
Too big Get file →
IQ3_M16.9 GB 20.6 GB
Too big Get file →
Q3_K_M18.1 GB 21.8 GB
Too big Get file →
Q3_K_L19.7 GB 23.4 GB
Too big Get file →
IQ4_XS20.1 GB 23.7 GB
Too big Get file →
Q4_021.2 GB 24.9 GB
Too big Get file →
Q4_K_S21.3 GB 25.0 GB
Too big Get file →
Q4_K_M22.4 GB 26.1 GB
Too big Get file →
Q5_K_S25.8 GB 29.5 GB
Too big Get file →
Q5_K_M26.5 GB 30.1 GB
Too big Get file →
Q6_K30.7 GB 34.4 GB
Too big Get file →

For 8‑GB VRAM use IQ1_S (8.04 GB file, 11.7 GB memory). For 12‑GB VRAM use IQ2_XS (11.13 GB file, 14.8 GB memory). For 16‑GB VRAM use Q2_K (13.87 GB file, 17.6 GB memory).

What GPU do you need for Cosmos 37b BASE i1?

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: the best quant that fits is IQ1_S (8.0 GB file, about 11.7 GB in use).
  • 16 GB VRAM: the best quant that fits is IQ2_S (11.8 GB file, about 15.5 GB in use).
  • 24 GB VRAM: the best quant that fits is IQ4_XS (20.1 GB file, about 23.7 GB in use).
  • 32 GB VRAM: the best quant that fits is Q5_K_M (26.5 GB file, about 30.1 GB in use).
  • 48 GB VRAM: the best quant that fits is Q6_K (30.7 GB file, about 34.4 GB in use).

Which Cosmos 37b BASE i1 quant should you choose?

Q4_K_M is the usual starting point for Cosmos 37b BASE i1: it balances file size and quality well. Here it is a 22.4 GB file.

If memory is tight, the smallest file is IQ1_S at 8.0 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
Parameters37.5B
Layers95
Trained context131,072 tokens
KV cache (fp16)380 KB per token

Source

Quantized bymradermacher
FamilyCosmos
Downloads (HF)877
Last updated2024-12-15

Frequently asked questions about Cosmos 37b BASE i1

How much VRAM does Cosmos 37b BASE i1 need?

About 26.1 GB at Q4_K_M with an 8K context window. That is the 22.4 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 Cosmos 37b BASE i1 on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 11.7 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Cosmos 37b BASE i1 GGUF?

It depends on the quant: IQ1_S is 8.0 GB, IQ1_M is 8.8 GB, IQ2_XXS is 10.0 GB, IQ2_XS is 11.1 GB, IQ2_S is 11.8 GB, IQ2_M is 12.8 GB and 16 more.

What context length does Cosmos 37b BASE i1 support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 3.2 GB at 8K and 12.8 GB at 32K.

Where can I download Cosmos 37b BASE i1 GGUF?

The files are on Hugging Face in the mradermacher/Cosmos-37b-BASE-i1-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Can I run Cosmos 37b BASE i1 on 8 GB VRAM?

Use the IQ1_S quantization; it requires 11.7 GB memory at 8k context, which is approximate for 8 GB VRAM with swapping.

Does this model support vision?

No, the model does not have vision capabilities.

Is there instruction tuning available?

The base model has not had instruction tuning; use -INSTRUCT for chat or instruct behavior.

Other GGUFs of Cosmos-37b-BASE

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Cosmos models

All Cosmos →

Other large models

All large models →