Skip to content

YanoljaNEXT EEVE Instruct 10.8B VRAM requirements

How much VRAM YanoljaNEXT EEVE Instruct 10.8B needs
4-bit weights: 6.1 GBWith 4K context: 7.5 GB8-bit, 4K context: 12.1 GB
Parameters
10.8B
Longest context
4K
Released
Feb 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM YanoljaNEXT EEVE Instruct 10.8B needs

Memory YanoljaNEXT EEVE Instruct 10.8B needs at each precision and context length
PrecisionWeightsWith 4K
16-bit20.1 GB21.5 GB
8-bit10.7 GB12.1 GB
6-bit8.3 GB9.7 GB
5-bit7.2 GB8.6 GB
4-bit6.1 GB7.5 GB
3-bit4.9 GB6.3 GB
2-bit3.8 GB5.2 GB
1.6-bit2.1 GB3.5 GB

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

GPUs and Macs that run YanoljaNEXT EEVE Instruct 10.8B

The best way YanoljaNEXT EEVE Instruct 10.8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40604-bit4K28 tok/s1.6 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4060 report
RTX 50604-bit4K45 tok/s1.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5060 report
RTX 30704-bit4K45 tok/s1.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 3070 report
RTX 3080 10 GB5-bit4K64 tok/s0.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 3080 10 GB report
RTX 3060 12 GB6-bit4K28 tok/s1.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 3060 12 GB report
Arc B5806-bit4K24 tok/s1.3 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Arc B580 report
RTX 40706-bit4K38 tok/s0.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4070 report
RTX 4070 Super6-bit4K38 tok/s0.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4070 Super report
RTX 50706-bit4K50 tok/s0.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit4K17 tok/s1.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit4K27 tok/s1.0 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit4K23 tok/s1.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit4K31 tok/s0.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit4K39 tok/s0.5 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit4K43 tok/s0.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4080 Super report
RTX 5070 Ti8-bit4K51 tok/s0.5 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5070 Ti report
RTX 50808-bit4K55 tok/s0.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5080 report
Radeon RX 7900 XTX8-bit4K45 tok/s1.0 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Radeon RX 7900 XTX report
RTX 30908-bit4K54 tok/s0.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 3090 report
RTX 40908-bit4K57 tok/s0.3 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 4090 report
Radeon AI PRO R97008-bit4K31 tok/s0.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Radeon AI PRO R9700 report
RTX 50908-bit4K96 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 5090 report
2× RTX 30908-bit4K54 tok/s0.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 2× RTX 3090 report
2× RTX 40908-bit4K57 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K74 tok/s0.3 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K55 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX 6000 Ada report
L40S8-bit4K50 tok/s0.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on L40S report
2× RTX 50908-bit4K96 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 2× RTX 5090 report
4× RTX 30908-bit4K54 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K96 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit4K11 tok/s15 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit4K8.4 tok/s15 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K8.4 tok/s15 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit4K18 tok/s7.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K18 tok/s7.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K18 tok/s7.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K32 tok/s3.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M4 Max 128 GB report
M5 Max 128 GB8-bit4K35 tok/s1.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K40 tok/s2.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K40 tok/s2.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K40 tok/s2.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K55 tok/s0.7 sPDF: YanoljaNEXT EEVE Instruct 10.8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K13 tok/s2.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K13 tok/s2.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K18 tok/s0.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on RTX Spark report
DGX Spark8-bit4K17 tok/s0.8 sPDF: YanoljaNEXT EEVE Instruct 10.8B on DGX Spark report
Dual DGX Spark8-bit4K17 tok/s0.4 sPDF: YanoljaNEXT EEVE Instruct 10.8B on Dual DGX Spark report
DGX Station GB3008-bit4K269 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K96 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K96 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K96 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K107 tok/s0.5 sPDF: YanoljaNEXT EEVE Instruct 10.8B on A100 80 GB SXM report
H100 SXM8-bit4K160 tok/s0.2 sPDF: YanoljaNEXT EEVE Instruct 10.8B on H100 SXM report
4× H200 NVL8-bit4K208 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B on 4× H200 NVL report
8× H200 NVL8-bit4K208 tok/s0.1 sPDF: YanoljaNEXT EEVE Instruct 10.8B 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 tensorblock/EEVE-Korean-Instruct-10.8B-v1.0-GGUF
vLLM
vllm serve yanolja/YanoljaNEXT-EEVE-Instruct-10.8B

Questions

How much VRAM does YanoljaNEXT EEVE Instruct 10.8B need?

YanoljaNEXT EEVE Instruct 10.8B needs 6.1 GB for its weights at 4-bit, 7.5 GB with an 4K context and 7.5 GB with 4K. At 8-bit, 12.1 GB.

What is the minimum VRAM for YanoljaNEXT EEVE Instruct 10.8B?

About 6.3 GB at 3-bit with an 4K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is YanoljaNEXT EEVE Instruct 10.8B?

10.8B parameters: 20.1 GB of weights at 16-bit, 6.1 GB at 4-bit.

Can YanoljaNEXT EEVE Instruct 10.8B run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 57 tokens a second.

More models