Model CoverageVision Language ModelsMeta ModelsMuse-Glimmer-30B

Muse-Glimmer-30B

View as Markdown

Muse-Glimmer-30B is a dense vision-language model with a 52-layer language backbone, a 50-layer vision tower, and a multimodal projector.

Set up NeMo AutoModel with the latest container or follow the installation instructions.

Fine-Tune Muse-Glimmer-30B

The recipes use the Hugging Face checkpoint meta-models/Muse-Glimmer-30B by default:

uv run automodel --nproc-per-node=8 \
  examples/vlm_finetune/muse_glimmer/muse_glimmer_30b_medpix.yaml

The implementation supports FSDP2, activation checkpointing, TP1/TP2, context parallelism, Transformer Engine packed THD inputs, pipeline parallelism, and LoRA. Multi-axis mRoPE with packed THD context parallelism is not supported.

Choose a Workflow

GoalStart Here
Full-parameter VLM SFTUse muse_glimmer_30b_medpix.yaml. Dataset: MedPix-VQA.
Single-node LoRA SFTUse muse_glimmer_30b_medpix_lora.yaml. Dataset: MedPix-VQA.
Single-node 16K packed text SFT with TP2 and CP4Use muse_glimmer_30b_tulu3_te_tp2_cp4_packed_16k.yaml. Dataset: Tulu 3.

Model Reference

Model Architecture

PropertyValue
TaskImage-Text-to-Text
ArchitectureMuseGlimmerForConditionalGeneration
Parameters30B
Model typesmuse_glimmer, muse_glimmer_text, muse_glimmer_vision
HF IDmeta-models/Muse-Glimmer-30B
  • Dense grouped-query attention language model with 32 query heads and 2 key-value heads
  • Alternating sliding and full attention across a 52-layer text backbone
  • Vision transformer with window and full attention across 50 layers
  • One-dimensional position IDs for packed Transformer Engine context parallelism