Skip to content

CodeLlama 13B Instruct HF VRAM requirements

How much VRAM CodeLlama 13B Instruct HF needs
4-bit weights: 7.3 GBWith 8K context: 14.3 GBWith 16K context: 20.5 GB8-bit, 8K context: 19.8 GB
Parameters
13B
Longest context
16K
Released
Aug 2023
License
Llama 2 Community
Strongest at
Not rated

Its licence (Llama 2 Community) is not a standard open-source one: check the terms before commercial use.

How much VRAM CodeLlama 13B Instruct HF needs

Memory CodeLlama 13B Instruct HF needs at each precision and context length
PrecisionWeightsWith 8KWith 16K
16-bit24.2 GB31.2 GB37.4 GB
8-bit12.9 GB19.8 GB26.0 GB
6-bit9.9 GB16.9 GB23.1 GB
5-bit8.6 GB15.5 GB21.8 GB
4-bit7.3 GB14.3 GB20.5 GB
3-bit5.9 GB12.8 GB19.1 GB
2-bit4.5 GB11.5 GB17.7 GB
1.6-bit2.5 GB9.4 GB15.7 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run CodeLlama 13B Instruct HF

The best way CodeLlama 13B Instruct HF runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: CodeLlama 13B Instruct HF on RTX 4060 report
RTX 5060Too bigPDF: CodeLlama 13B Instruct HF on RTX 5060 report
RTX 3070Too bigPDF: CodeLlama 13B Instruct HF on RTX 3070 report
RTX 3080 10 GBToo bigPDF: CodeLlama 13B Instruct HF on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: CodeLlama 13B Instruct HF on RTX 3060 12 GB report
Arc B580Too bigPDF: CodeLlama 13B Instruct HF on Arc B580 report
RTX 4070Too bigPDF: CodeLlama 13B Instruct HF on RTX 4070 report
RTX 4070 SuperToo bigPDF: CodeLlama 13B Instruct HF on RTX 4070 Super report
RTX 5070Too bigPDF: CodeLlama 13B Instruct HF on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: CodeLlama 13B Instruct HF on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: CodeLlama 13B Instruct HF on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: CodeLlama 13B Instruct HF on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: CodeLlama 13B Instruct HF on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: CodeLlama 13B Instruct HF on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: CodeLlama 13B Instruct HF on RTX 4080 Super report
RTX 5070 TiToo bigPDF: CodeLlama 13B Instruct HF on RTX 5070 Ti report
RTX 5080Too bigPDF: CodeLlama 13B Instruct HF on RTX 5080 report
Radeon RX 7900 XTX5-bit16K56 tok/s1.2 sPDF: CodeLlama 13B Instruct HF on Radeon RX 7900 XTX report
RTX 30905-bit16K66 tok/s0.8 sPDF: CodeLlama 13B Instruct HF on RTX 3090 report
RTX 40905-bit16K71 tok/s0.3 sPDF: CodeLlama 13B Instruct HF on RTX 4090 report
Radeon AI PRO R97008-bit16K27 tok/s0.8 sPDF: CodeLlama 13B Instruct HF on Radeon AI PRO R9700 report
RTX 50908-bit16K83 tok/s0.3 sPDF: CodeLlama 13B Instruct HF on RTX 5090 report
2× RTX 30908-bit16K45 tok/s0.5 sPDF: CodeLlama 13B Instruct HF on 2× RTX 3090 report
2× RTX 40908-bit16K49 tok/s0.2 sPDF: CodeLlama 13B Instruct HF on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit16K64 tok/s0.4 sPDF: CodeLlama 13B Instruct HF on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit16K46 tok/s0.3 sPDF: CodeLlama 13B Instruct HF on RTX 6000 Ada report
L40S8-bit16K42 tok/s0.5 sPDF: CodeLlama 13B Instruct HF on L40S report
2× RTX 50908-bit16K83 tok/s0.2 sPDF: CodeLlama 13B Instruct HF on 2× RTX 5090 report
4× RTX 30908-bit16K45 tok/s0.2 sPDF: CodeLlama 13B Instruct HF on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit16K83 tok/s0.2 sPDF: CodeLlama 13B Instruct HF on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: CodeLlama 13B Instruct HF on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: CodeLlama 13B Instruct HF on Mac mini M4 24 GB report
Mac mini M4 32 GB3-bit16K15 tok/s18 sPDF: CodeLlama 13B Instruct HF on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: CodeLlama 13B Instruct HF on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit16K16 tok/s9.0 sPDF: CodeLlama 13B Instruct HF on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit16K16 tok/s9.0 sPDF: CodeLlama 13B Instruct HF on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit16K27 tok/s4.5 sPDF: CodeLlama 13B Instruct HF on M4 Max 128 GB report
M5 Max 128 GB8-bit16K30 tok/s1.3 sPDF: CodeLlama 13B Instruct HF on M5 Max 128 GB report
M3 Ultra 96 GB8-bit16K35 tok/s2.5 sPDF: CodeLlama 13B Instruct HF on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit16K35 tok/s2.5 sPDF: CodeLlama 13B Instruct HF on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit16K35 tok/s2.5 sPDF: CodeLlama 13B Instruct HF on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit16K48 tok/s0.8 sPDF: CodeLlama 13B Instruct HF on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit16K11 tok/s2.6 sPDF: CodeLlama 13B Instruct HF on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit16K11 tok/s2.6 sPDF: CodeLlama 13B Instruct HF on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K15 tok/s0.9 sPDF: CodeLlama 13B Instruct HF on RTX Spark report
DGX Spark8-bit16K14 tok/s0.9 sPDF: CodeLlama 13B Instruct HF on DGX Spark report
Dual DGX Spark8-bit16K14 tok/s0.5 sPDF: CodeLlama 13B Instruct HF on Dual DGX Spark report
DGX Station GB3008-bit16K253 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit16K83 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit16K83 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit16K83 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit16K93 tok/s0.6 sPDF: CodeLlama 13B Instruct HF on A100 80 GB SXM report
H100 SXM8-bit16K142 tok/s0.2 sPDF: CodeLlama 13B Instruct HF on H100 SXM report
4× H200 NVL8-bit16K190 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on 4× H200 NVL report
8× H200 NVL8-bit16K190 tok/s0.1 sPDF: CodeLlama 13B Instruct HF on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf TheBloke/CodeLlama-13B-Instruct-GGUF
vLLM
vllm serve codellama/CodeLlama-13b-Instruct-hf
MLX (Mac)
mlx_lm.chat --model mlx-community/CodeLlama-13b-Instruct-hf-8bit-mlx

Questions

How much VRAM does CodeLlama 13B Instruct HF need?

CodeLlama 13B Instruct HF needs 7.3 GB for its weights at 4-bit, 14.3 GB with an 8K context and 20.5 GB with 16K. At 8-bit, 19.8 GB.

What is the minimum VRAM for CodeLlama 13B Instruct HF?

About 12.8 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the Radeon RX 7900 XTX.

How big is CodeLlama 13B Instruct HF?

13B parameters: 24.2 GB of weights at 16-bit, 7.3 GB at 4-bit.

Can CodeLlama 13B Instruct HF run on an RTX 4090?

Yes. At 5-bit it holds 16K of context and writes about 71 tokens a second.

More models