What are the hardware requirements for Haiku?

Haiku requires approximately 1.3 GB of VRAM and supports a context window of 32K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Haiku and how large is it?

Haiku is developed by kerzgrr, a model with 655.8m parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Haiku?

Haiku is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Haiku by kerzgrr — VRAM 1.3GB, 32K context

Name: Haiku
Author: kerzgrr

Haiku is an open-source language model by kerzgrr. Features: 655.8m LLM, VRAM: 1.3GB, Context: 32K, License: apache-2.0, Instruction-Based, LLM Explorer Score: 0.28.

Apo Chat Conversational Dataset:agent-ark/toucan-1.5m Dataset:allenai/tulu-3-sft-per... Dataset:allenai/wildchat-4.8m Dataset:huggingfacefw/fineweb-... Dataset:huggingfaceh4/no robot... Dataset:huggingfaceh4/ultracha... Dataset:huggingfacetb/smoltalk Dataset:huggingfacetb/smoltalk... Dataset:interstellarninja/herm... Dataset:nousresearch/hermes-3-... Dataset:nvidia/nemotron-agenti... Dataset:nvidia/nemotron-cascad... Dataset:nvidia/nemotron-instru... Dataset:nvidia/nemotron-post-t... Dataset:nvidia/nemotron-sft-ag... Dataset:nvidia/nemotron-sft-in... En Gated-mla Haiku Hybrid Instruct Instruction-tuned Kimi-delta-attention Preference-optimization Pytorch Reasoning Region:us Safetensors Sft Tiny gdn Tool-use

Model Card on HF 🤗: https://huggingface.co/kerzgrr/Haiku

Haiku Parameters and Internals

LLM Name	Haiku
Repository 🤗	https://huggingface.co/kerzgrr/Haiku
Model Name	Haiku
Model Size	655.8m
Required VRAM	1.3 GB
Updated	2026-10-12
Maintainer	kerzgrr
Model Type	haiku
Instruction-Based	Yes
Model Files	1.3 GB
Supported Languages	en
Model Architecture	HaikuForCausalLM
License	apache-2.0
Context Length	32768
Model Max Length	32768
Transformers Version	4.45.0
Tokenizer Class	PreTrainedTokenizerFast
Padding Token	<\|padding\|>
Vocabulary Size	65536
Torch Data Type	bfloat16

Rank the Haiku capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation