CodeLlama 34B Python Fp16 needs about 22.3 GB of memory at Q4_K_M (a 20.2 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 18 quants, from 4.0 GB (Q4_K) to 23.8 GB (Q5_K).
CodeLlama 34B Python Fp16 is a 33.7B-parameter Llama model in GGUF format, quantized by RichardErkhov. Its trained context length is 16K tokens. The repository has 18 quant files, from 4.0 GB (Q4_K) to 23.8 GB (Q5_K).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K | 4.0 GB | 6.1 GB | Fits | Get file → | |
| Q3_K_S | 12.4 GB | 14.6 GB | Too big | Get file → | |
| Q2_K | 12.5 GB | 14.6 GB | Too big | Get file → | |
| IQ3_XS | 13.9 GB | 16.0 GB | Too big | Get file → | |
| Q4_0 | 14.3 GB | 16.4 GB | Too big | Get file → | |
| IQ3_S | 14.7 GB | 16.8 GB | Too big | Get file → | |
| IQ3_M | 15.2 GB | 17.3 GB | Too big | Get file → | |
| Q3_K | 16.3 GB | 18.4 GB | Too big | Get file → | |
| Q3_K_M | 16.3 GB | 18.4 GB | Too big | Get file → | |
| Q5_K_S | 17.2 GB | 19.3 GB | Too big | Get file → | |
| Q4_1 | 17.6 GB | 19.7 GB | Too big | Get file → | |
| Q3_K_L | 17.8 GB | 19.9 GB | Too big | Get file → | |
| IQ4_XS | 18.2 GB | 20.3 GB | Too big | Get file → | |
| Q4_K_S | 19.2 GB | 21.3 GB | Too big | Get file → | |
| IQ4_NL | 19.2 GB | 21.3 GB | Too big | Get file → | |
| Q4_K_M | 20.2 GB | 22.3 GB | Too big | Get file → | |
| Q5_0 | 22.0 GB | 24.1 GB | Too big | Get file → | |
| Q5_K | 23.8 GB | 25.9 GB | Too big | Get file → |
What GPU do you need for CodeLlama 34B Python 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: the best quant that fits is Q4_K (4.0 GB file, about 6.1 GB in use).
- 12 GB VRAM: the best quant that fits is Q4_K (4.0 GB file, about 6.1 GB in use).
- 16 GB VRAM: the best quant that fits is IQ3_XS (13.9 GB file, about 16.0 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_K_M (20.2 GB file, about 22.3 GB in use).
- 32 GB VRAM: the best quant that fits is Q5_K (23.8 GB file, about 25.9 GB in use).
Which CodeLlama 34B Python Fp16 quant should you choose?
Q4_K_M is the usual starting point for CodeLlama 34B Python Fp16: it balances file size and quality well. Here it is a 20.2 GB file.
If memory is tight, the smallest file is Q4_K at 4.0 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about CodeLlama 34B Python Fp16
How much VRAM does CodeLlama 34B Python Fp16 need?
About 22.3 GB at Q4_K_M with an 8K context window. That is the 20.2 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 CodeLlama 34B Python Fp16 on an 8 GB GPU?
Yes, with Q4_K: the 4.0 GB file needs about 6.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of CodeLlama 34B Python Fp16 GGUF?
It depends on the quant: Q4_K is 4.0 GB, Q3_K_S is 12.4 GB, Q2_K is 12.5 GB, IQ3_XS is 13.9 GB, Q4_0 is 14.3 GB, IQ3_S is 14.7 GB and 12 more.
What context length does CodeLlama 34B Python Fp16 support?
The GGUF header lists a trained context length of 16,384 tokens (16K). The KV cache grows with the context you use: about 1.6 GB at 8K and 3.2 GB at 16K.
Where can I download CodeLlama 34B Python Fp16 GGUF?
The files are on Hugging Face in the RichardErkhov/TheBloke_-_CodeLlama-34B-Python-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 CodeLlama 34B Python 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.