AI21 Jamba2 Mini needs about 36.0 GB of memory at Q4_K_M (a 31.2 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 11 quants, from 18.7 GB (Q2_K) to 54.8 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 17.2 GB.
AI21 Jamba2 Mini is a 51.6B-parameter Jamba model in GGUF format, quantized by mradermacher. Its trained context length is 256K tokens. The repository has 11 quant files, from 18.7 GB (Q2_K) to 54.8 GB (Q8_0).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 18.7 GB | 23.5 GB | Too big | Get file → | |
| Q3_K_S | 22.3 GB | 27.1 GB | Too big | Get file → | |
| Q3_K_M | 24.5 GB | 29.3 GB | Too big | Get file → | |
| Q3_K_L | 26.4 GB | 31.2 GB | Too big | Get file → | |
| IQ4_XS | 27.8 GB | 32.6 GB | Too big | Get file → | |
| Q4_K_S | 29.4 GB | 34.1 GB | Too big | Get file → | |
| Q4_K_M | 31.2 GB | 36.0 GB | Too big | Get file → | |
| Q5_K_S | 35.5 GB | 40.3 GB | Too big | Get file → | |
| Q5_K_M | 36.6 GB | 41.4 GB | Too big | Get file → | |
| Q6_K | 42.3 GB | 47.1 GB | Too big | Get file → | |
| Q8_0 | 54.8 GB | 59.6 GB | Too big | Get file → |
What GPU do you need for AI21 Jamba2 Mini?
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 Q2_K (18.7 GB file, about 23.5 GB in use).
- 32 GB VRAM: the best quant that fits is Q3_K_L (26.4 GB file, about 31.2 GB in use).
- 48 GB VRAM: the best quant that fits is Q6_K (42.3 GB file, about 47.1 GB in use).
Which AI21 Jamba2 Mini quant should you choose?
Q4_K_M is the usual starting point for AI21 Jamba2 Mini: it balances file size and quality well. Here it is a 31.2 GB file.
The Q8_0 file is 54.8 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 18.7 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
Source
Frequently asked questions about AI21 Jamba2 Mini
How much VRAM does AI21 Jamba2 Mini need?
About 36.0 GB at Q4_K_M with an 8K context window. That is the 31.2 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 AI21 Jamba2 Mini on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 23.5 GB, so you would need a larger GPU or CPU offloading.
What is the file size of AI21 Jamba2 Mini GGUF?
It depends on the quant: Q2_K is 18.7 GB, Q3_K_S is 22.3 GB, Q3_K_M is 24.5 GB, Q3_K_L is 26.4 GB, IQ4_XS is 27.8 GB, Q4_K_S is 29.4 GB and 5 more.
What context length does AI21 Jamba2 Mini support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 4.3 GB at 8K and 17.2 GB at 32K.
Where can I download AI21 Jamba2 Mini GGUF?
The files are on Hugging Face in the mradermacher/AI21-Jamba2-Mini-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 AI21 Jamba2 Mini 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.