Goliath-longLORA-120b-rope8-32k-fp16 needs about 75.7 GB of memory at Q4_K_M (a 70.6 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 13 quants, from 27.1 GB (IQ1_M) to 235.5 GB (F16).
Goliath-longLORA-120b-rope8-32k-fp16 is a 118B-parameter Goliath model in GGUF format, quantized by bartowski. Its trained context length is 4K tokens. The repository has 13 quant files, from 27.1 GB (IQ1_M) to 235.5 GB (F16).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_M | 27.1 GB | 29.9 GB | Too big | Get file → | |
| IQ2_XXS | 31.1 GB | 33.9 GB | Too big | Get file → | |
| IQ2_M | 39.6 GB | 42.4 GB | Too big | Get file → | |
| Q2_K | 43.2 GB | 46.0 GB | Too big | Get file → | |
| IQ3_XXS | 45.2 GB | 48.0 GB | Too big | Get file → | |
| Q3_K_S | 50.7 GB (2 parts) | 53.5 GB | Too big | Get file → | |
| IQ3_M | 52.6 GB (2 parts) | 55.4 GB | Too big | Get file → | |
| Q3_K_M | 56.6 GB (2 parts) | 59.4 GB | Too big | Get file → | |
| IQ4_XS | 62.8 GB (2 parts) | 65.6 GB | Too big | Get file → | |
| Q4_K_M | 70.6 GB (2 parts) | 73.4 GB | Too big | Get file → | |
| Q5_K_M | 83.2 GB (2 parts) | 86.0 GB | Too big | Get file → | |
| Q8_0 | 125.1 GB (3 parts) | 127.9 GB | Too big | Get file → | |
| F16 | 235.5 GB (6 parts) | 238.3 GB | Too big | Get file → |
What GPU do you need for Goliath-longLORA-120b-rope8-32k-fp16?
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: no quant fits at an 8K context.
- 48 GB VRAM: the best quant that fits is IQ2_M (39.6 GB file, about 44.7 GB in use).
Which Goliath-longLORA-120b-rope8-32k-fp16 quant should you choose?
Q4_K_M is the usual starting point for Goliath-longLORA-120b-rope8-32k-fp16: it balances file size and quality well. Here it is a 70.6 GB file.
The Q8_0 file is 125.1 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 IQ1_M at 27.1 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
Frequently asked questions about Goliath-longLORA-120b-rope8-32k-fp16
How much VRAM does Goliath-longLORA-120b-rope8-32k-fp16 need?
About 75.7 GB at Q4_K_M with an 8K context window. That is the 70.6 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 Goliath-longLORA-120b-rope8-32k-fp16 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 32.2 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Goliath-longLORA-120b-rope8-32k-fp16 GGUF?
It depends on the quant: IQ1_M is 27.1 GB, IQ2_XXS is 31.1 GB, IQ2_M is 39.6 GB, Q2_K is 43.2 GB, IQ3_XXS is 45.2 GB, Q3_K_S is 50.7 GB and 7 more.
What context length does Goliath-longLORA-120b-rope8-32k-fp16 support?
The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 4.6 GB at 8K and 2.3 GB at 4K.
Where can I download Goliath-longLORA-120b-rope8-32k-fp16 GGUF?
The files are on Hugging Face in the bartowski/Goliath-longLORA-120b-rope8-32k-fp16-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 Goliath-longLORA-120b-rope8-32k-fp16 use?
Hugging Face lists the license as llama2. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.