lIlBrother llama2 merge v0.4 needs about 15.2 GB of memory at Q4_K_M (a 8.0 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 19 quants, from 4.8 GB (Q3_K) to 14.0 GB (Q8_0).
lIlBrother llama2 merge v0.4 is a 13.2B-parameter Llama model in GGUF format, quantized by RichardErkhov. Its trained context length is 2K tokens. The repository has 19 quant files, from 4.8 GB (Q3_K) to 14.0 GB (Q8_0).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q3_K | 4.8 GB | 6.9 GB | Fits | Get file → | |
| Q2_K | 4.9 GB | 7.1 GB | Fits | Get file → | |
| Q3_K_S | 5.8 GB | 7.9 GB | Tight | Get file → | |
| Q3_K_M | 6.4 GB | 8.6 GB | Too big | Get file → | |
| Q3_K_L | 7.0 GB | 9.2 GB | Too big | Get file → | |
| IQ4_XS | 7.1 GB | 9.3 GB | Too big | Get file → | |
| Q4_0 | 7.5 GB | 9.6 GB | Too big | Get file → | |
| IQ4_NL | 7.5 GB | 9.7 GB | Too big | Get file → | |
| Q4_K_S | 7.5 GB | 9.7 GB | Too big | Get file → | |
| Q4_K | 8.0 GB | 10.1 GB | Too big | Get file → | |
| Q4_K_M | 8.0 GB | 10.1 GB | Too big | Get file → | |
| Q4_1 | 8.3 GB | 10.5 GB | Too big | Get file → | |
| Q5_0 | 9.1 GB | 11.3 GB | Too big | Get file → | |
| Q5_K_S | 9.1 GB | 11.3 GB | Too big | Get file → | |
| Q5_K | 9.3 GB | 11.5 GB | Too big | Get file → | |
| Q5_K_M | 9.3 GB | 11.5 GB | Too big | Get file → | |
| Q5_1 | 9.9 GB | 12.1 GB | Too big | Get file → | |
| Q6_K | 10.8 GB | 13.0 GB | Too big | Get file → | |
| Q8_0 | 14.0 GB | 16.2 GB | Too big | Get file → |
What GPU do you need for lIlBrother llama2 merge v0.4?
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 Q3_K (4.8 GB file, about 12.0 GB in use).
- 16 GB VRAM: the best quant that fits is Q4_1 (8.3 GB file, about 15.5 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (14.0 GB file, about 21.2 GB in use).
Which lIlBrother llama2 merge v0.4 quant should you choose?
Q4_K_M is the usual starting point for lIlBrother llama2 merge v0.4: it balances file size and quality well. Here it is a 8.0 GB file.
The Q8_0 file is 14.0 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 Q3_K at 4.8 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about lIlBrother llama2 merge v0.4
How much VRAM does lIlBrother llama2 merge v0.4 need?
About 15.2 GB at Q4_K_M with an 8K context window. That is the 8.0 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 lIlBrother llama2 merge v0.4 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 12.0 GB, so you would need a larger GPU or CPU offloading.
What is the file size of lIlBrother llama2 merge v0.4 GGUF?
It depends on the quant: Q3_K is 4.8 GB, Q2_K is 4.9 GB, Q3_K_S is 5.8 GB, Q3_K_M is 6.4 GB, Q3_K_L is 7.0 GB, IQ4_XS is 7.1 GB and 13 more.
What context length does lIlBrother llama2 merge v0.4 support?
The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 6.7 GB at 8K and 1.7 GB at 2K.
Where can I download lIlBrother llama2 merge v0.4 GGUF?
The files are on Hugging Face in the RichardErkhov/lIlBrother_-_llama2-merge-v0.4-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.