Model reference · open weights

llama3-CLA-2

LLMs answerdotai Text gen 1 build Licence not stated 9 dl/mo

llama3-CLA-2 is an open-weight language model from answerdotai. llama3-8b-instruct-CLA-2 (BF16) weighs 16.1 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.

Request Rent a GPU

For self-hosting, compare this build's requirements below with the GPU you plan to rent.

llama3-CLA-2 is a text-generation model developed by answerdotai and published as answerdotai/llama3-8b-instruct-CLA-2.
It supports a context length of 8192 tokens and uses a mechanism where it shares KV activations every two layers.

Summary of the answerdotai/llama3-8b-instruct-CLA-2 model card, 2026-10-01

Hugging Face ↗

What it is

Released by	answerdotai
Released	2024-07-18
VRAM	16.1 GB for the weights

What it runs on

Memory and cards for llama3-8b-instruct-CLA-2 (BF16)

16.1 GBweights, file size

131 MBcache per 1K tokens

745 MBruntime overhead, at least

8,192 tokenscontext max

Card	Requests at once	Context max	Memory
Card	8K each	Context max	Memory
RTX 3060 12 GB … RTX 4060 Ti 16 GB 2 smaller cards
RTX 3090 24 GB	6	all 8K	23.4 GB
RTX 4090 24 GB	6	all 8K	23.4 GB
RTX 5090 32 GB	13	all 8K	31.0 GB
L40S 48 GB	25	all 8K	44.0 GB
A100 80 GB	57	all 8K	78.2 GB
H100 80 GB	53	all 8K	78.1 GB
RTX PRO 6000 Blackwell 96 GB	67	all 8K	93.8 GB
DGX Spark (GB10) 128 GB unified	80	all 8K	107 GB
H200 141 GB	109	all 8K	138 GB
B200 180 GB	143	all 8K	176 GB
2× RTX 3060 12 GB tensor parallel	5	all 8K	11.6 GB a card
2× RTX 4060 Ti 16 GB tensor parallel	12	all 8K	15.4 GB a card
2× RTX 4090 24 GB tensor parallel	27	all 8K	23.4 GB a card
2× RTX 3090 24 GB tensor parallel	27	all 8K	23.4 GB a card

Memory needed at each load

Requests at once	8K tokens each
1	17.9 GB
5	22.2 GB
8	25.4 GB
16	34.0 GB
32	51.2 GB
64	85.5 GB

One card, with vLLM's small-card settings.

From the model card

What answerdotai says about llama3-CLA-2

Read the model card

Transformers : https://github.com/huggingface/transformers/blob/47fedc1665f14a60260b1b8357e682669300093a/src/transformers/models/llama/modeling_llama.py

Training Script: https://github.com/AnswerDotAI/fsdp_qlora/blob/3f7c583e985ff35e37a7b7497a7d4fedb77df695/experiments/cla/train.sh

This model shares KV activations every 2 layers. For example, layer 1 uses layer 0 kv activations, layer 3 uses layer 2 kv activations, etc..

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

Details

Type	Language models
Task	Text gen
Context	8,192 tokens
Runs with	transformers
Popularity	9 downloads / month
Weights	16.1 GB, llama3-8b-instruct-CLA-2 (BF16, file size)