What are the hardware requirements for GRM 3.2 Sky MLX BF16?

GRM 3.2 Sky MLX BF16 requires approximately 69.9 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed GRM 3.2 Sky MLX BF16 and how large is it?

GRM 3.2 Sky MLX BF16 is developed by swdjojo, a model with 35.1b 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 GRM 3.2 Sky MLX BF16?

GRM 3.2 Sky MLX BF16 is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

GRM 3.2 Sky MLX BF16 by swdjojo — VRAM 69.9GB

Name: GRM 3.2 Sky MLX BF16
Author: swdjojo

GRM 3.2 Sky MLX BF16 is an open-source language model by swdjojo. Features: 35.1b LLM, VRAM: 69.9GB.

Base model:finetune:orionllm/g... Base model:orionllm/grm-3.2-sk... Conversational Eval-results Image-text-to-text Mlx Qwen3 5 moe Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/swdjojo/GRM-3.2-Sky-MLX-BF16

GRM 3.2 Sky MLX BF16 Parameters and Internals

LLM Name	GRM 3.2 Sky MLX BF16
Repository 🤗	https://huggingface.co/swdjojo/GRM-3.2-Sky-MLX-BF16
Base Model(s)	OrionLLM/GRM-3.2-Sky OrionLLM/GRM-3.2-Sky
Model Size	35.1b
Required VRAM	69.9 GB
Updated	2026-08-09
Maintainer	swdjojo
Model Type	qwen3_5_moe
Model Files	5.3 GB: 1-of-14 5.0 GB: 2-of-14 5.0 GB: 3-of-14 5.0 GB: 4-of-14 5.1 GB: 5-of-14 5.0 GB: 6-of-14 5.0 GB: 7-of-14 5.0 GB: 8-of-14 5.1 GB: 9-of-14 5.0 GB: 10-of-14 5.0 GB: 11-of-14 5.0 GB: 12-of-14 5.1 GB: 13-of-14 4.3 GB: 14-of-14
Model Architecture	Qwen3_5MoeForConditionalGeneration
Model Max Length	262144
Transformers Version	5.8.1
Tokenizer Class	Qwen2Tokenizer
Padding Token	<\|endoftext\|>
Errors	replace

Rank the GRM 3.2 Sky MLX BF16 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 GRM 3.2 Sky MLX BF16

Best Alternatives	Context / RAM	Downloads	Likes
Agents A1 Bf16	0K / 69.9 GB	596	3

Note: green Score (e.g. "73.2") means that the model is better than swdjojo/GRM-3.2-Sky-MLX-BF16.