BigCodeLlama 92b needs about 39.4 GB of memory at IQ3_XXS (a 35.3 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 15 quants, from 19.3 GB (IQ1_S) to 49.1 GB (IQ4_XS).
BigCodeLlama 92b is a 92.1B-parameter Bigcodellama model in GGUF format, quantized by mradermacher. Its trained context length is 2K tokens. The repository has 15 quant files, from 19.3 GB (IQ1_S) to 49.1 GB (IQ4_XS).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 19.3 GB | 20.7 GB | Too big | Get file → | |
| IQ1_M | 21.2 GB | 22.6 GB | Too big | Get file → | |
| IQ2_XXS | 24.4 GB | 25.8 GB | Too big | Get file → | |
| IQ2_XS | 27.1 GB | 28.5 GB | Too big | Get file → | |
| IQ2_S | 28.5 GB | 29.8 GB | Too big | Get file → | |
| IQ2_M | 31.0 GB | 32.4 GB | Too big | Get file → | |
| Q2_K | 33.8 GB | 35.2 GB | Too big | Get file → | |
| IQ3_XXS | 35.3 GB | 36.7 GB | Too big | Get file → | |
| IQ3_XS | 37.7 GB | 39.1 GB | Too big | Get file → | |
| Q3_K_S | 39.7 GB | 41.1 GB | Too big | Get file → | |
| IQ3_S | 39.8 GB | 41.2 GB | Too big | Get file → | |
| IQ3_M | 41.2 GB | 42.6 GB | Too big | Get file → | |
| Q3_K_M | 44.3 GB | 45.7 GB | Too big | Get file → | |
| Q3_K_L | 48.2 GB | 49.6 GB | Too big | Get file → | |
| IQ4_XS | 49.1 GB | 50.5 GB | Too big | Get file → |
What GPU do you need for BigCodeLlama 92b?
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 IQ1_S (19.3 GB file, about 23.4 GB in use).
- 32 GB VRAM: the best quant that fits is IQ2_XS (27.1 GB file, about 31.2 GB in use).
- 48 GB VRAM: the best quant that fits is IQ3_M (41.2 GB file, about 45.2 GB in use).
Which BigCodeLlama 92b quant should you choose?
IQ3_XXS (35.3 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 19.3 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
Source
Frequently asked questions about BigCodeLlama 92b
How much VRAM does BigCodeLlama 92b need?
About 39.4 GB at IQ3_XXS with an 8K context window. That is the 35.3 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 BigCodeLlama 92b on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 23.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of BigCodeLlama 92b GGUF?
It depends on the quant: IQ1_S is 19.3 GB, IQ1_M is 21.2 GB, IQ2_XXS is 24.4 GB, IQ2_XS is 27.1 GB, IQ2_S is 28.5 GB, IQ2_M is 31.0 GB and 9 more.
What context length does BigCodeLlama 92b support?
The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 3.6 GB at 8K and 0.9 GB at 2K.
Where can I download BigCodeLlama 92b GGUF?
The files are on Hugging Face in the mradermacher/BigCodeLlama-92b-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 BigCodeLlama 92b use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.