What are the hardware requirements for Azure Dusk V0.2?

Azure Dusk V0.2 requires approximately 24.5 GB of VRAM and supports a context window of 1000K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Azure Dusk V0.2 and how large is it?

Azure Dusk V0.2 is developed by Epiculous, a model with 12.2b 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 Azure Dusk V0.2?

Azure Dusk V0.2 is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Azure Dusk V0.2 by Epiculous — VRAM 24.5GB, 1000K context

Name: Azure Dusk V0.2
Author: Epiculous

Azure Dusk V0.2 is an open-source language model by Epiculous. Features: 12.2b LLM, VRAM: 24.5GB, Context: 1000K, License: apache-2.0, Instruction-Based, LLM Explorer Score: 0.18.

Conversational Dataset:anthracite-org/kalo-op... Dataset:anthracite-org/kalo op... Dataset:anthracite-org/nopm cl... Dataset:anthracite-org/stheno-... Dataset:epiculous/synthrp-gens... Dataset:epiculous/synthstruct-... Dataset:gryphe/sonnet3.5-charc... Dataset:pjmixers/hieunguyenmin... De En Es Fr Instruct It Ja Mistral Model-index Pt Region:us Ru Safetensors Sharded Tensorflow Zh

Model Card on HF 🤗: https://huggingface.co/Epiculous/Azure_Dusk-v0.2

Azure Dusk V0.2 Parameters and Internals

Model Type

text-generation

Additional Notes

This version follows Azure_Dusk-v0.2 and includes expanded data and unique training methodology with RSLoRA and ChatML.

Supported Languages

en (yes), fr (yes), de (yes), es (yes), it (yes), pt (yes), ru (yes), zh (yes), ja (yes)

Training Details

Data Sources:

Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned, anthracite-org/stheno-filtered-v1.1, PJMixers/hieunguyenminh_roleplay-deduped-ShareGPT, Gryphe/Sonnet3.5-Charcard-Roleplay, Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned, anthracite-org/kalo-opus-instruct-22k-no-refusal, anthracite-org/nopm_claude_writing_fixed, anthracite-org/kalo_opus_misc_240827

Methodology:

Training was done using RSLoRA instead of regular LoRA. A two-phased approach was utilized; first training on RP data followed by instruct data.

Training Time:

Twice over 2 epochs each on two 2x NVIDIA A6000 GPUs

Hardware Used:

2x NVIDIA A6000 GPUs

Model Architecture:

ChatML formatting with Changed prompting structure.

Input Output

Input Format:

Trained on ChatML formatting.

Accepted Modalities:

text

Output Format:

text

LLM Name	Azure Dusk V0.2
Repository 🤗	https://huggingface.co/Epiculous/Azure_Dusk-v0.2
Model Size	12.2b
Required VRAM	24.5 GB
Updated	2026-08-02
Maintainer	Epiculous
Model Type	mistral
Instruction-Based	Yes
Model Files	4.9 GB: 1-of-5 4.9 GB: 2-of-5 4.9 GB: 3-of-5 4.9 GB: 4-of-5 4.9 GB: 5-of-5
Supported Languages	en fr de es it pt ru zh ja
Model Architecture	MistralForCausalLM
License	apache-2.0
Context Length	1024000
Model Max Length	1024000
Transformers Version	4.44.0
Tokenizer Class	PreTrainedTokenizerFast
Padding Token	<pad>
Vocabulary Size	131072
Torch Data Type	bfloat16

Rank the Azure Dusk V0.2 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

Best Alternatives to Azure Dusk V0.2

Best Alternatives	Context / RAM	Downloads	Likes
...istral Nemo Instruct 2407 MPOA	1000K / 24.5 GB	37	3
...ish Mistral Nemo Instruct 2407	1000K / 24.5 GB	77	2
Mistral Nemo Kurdish	1000K / 24.5 GB	37	3
Violet Twilight V0.2	1000K / 24.5 GB	66	49
Kalypso	1000K / 24.5 GB	45	24
Crimson Dawn V0.2	1000K / 24.5 GB	17	13
...al Nemo Japanese Instruct 2408	1000K / 24.5 GB	2358	49
Mistral Nemo Instruct 2407 FP8	1000K / 13.6 GB	4691	19
Educa Ai Nemo Sft	1000K / 49.3 GB	13	4
...l Nemo Abliterated Nemo Pro V2	1000K / 24.5 GB	19	1

Note: green Score (e.g. "73.2") means that the model is better than Epiculous/Azure_Dusk-v0.2.