Instructions to use meta-models/Muse-Glimmer-30B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use meta-models/Muse-Glimmer-30B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="meta-models/Muse-Glimmer-30B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("meta-models/Muse-Glimmer-30B")
model = AutoModelForMultimodalLM.from_pretrained("meta-models/Muse-Glimmer-30B", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use meta-models/Muse-Glimmer-30B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "meta-models/Muse-Glimmer-30B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-models/Muse-Glimmer-30B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/meta-models/Muse-Glimmer-30B

SGLang

How to use meta-models/Muse-Glimmer-30B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "meta-models/Muse-Glimmer-30B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-models/Muse-Glimmer-30B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "meta-models/Muse-Glimmer-30B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-models/Muse-Glimmer-30B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use meta-models/Muse-Glimmer-30B with Docker Model Runner:
```
docker model run hf.co/meta-models/Muse-Glimmer-30B
```

Works well, comparable to Qwen 3.6 27B and then some

#36

by Omnomynous - opened about 21 hours ago

Discussion

Omnomynous

about 21 hours ago

Ran the BF16 variant from unsloth on my M2 Ultra 192GB. Gets about 14tok/s, but more importantly, maintains that speed throughout the entirety of context. Very usable for coding tasks. Shorter reasoning chains than Qwen 3.6. Also seems to prefer to make smaller, more surgical edits. I'm seeing more tool calls with shorter edits. Gets more done in 128k context than Qwen, but the limitation to 128k is a disappointment when the competition all now use 256k windows.

Has one HUGE advantage over qwen 3.6 27B, at least on my setup, and that is that it doesn't seem to need to prefill after a long searies of tool calls and thinking. On Qwen, if it finishes a task at 50% context and I want it to continue on to something, I'm in for a prefil wait up to a few minutes. With Glimmer, it responds and springs back into action with basically no wait. Impressive model, hamstrung by a non compteitive context limit.

Will still be using Qwen 3.6 35B MOE for my "daily driver" because the speed is so much better, but this will probably replace the 27B variant for the hard to scratch bugs and technical problems.

Nice job Meta team! Glad to see you guys back in the game. Was getting a little one sided around here with everything good made in China lately! You've still got the chops. Now keep it going! Give us a full fat 256k context, and an MOE speedy boy when "good" is good enough. :)

pawarshardul

about 21 hours ago

https://huggingface.co/meta-models/Muse-Glimmer-30B/discussions/22
yarn works -- i get 262k context on my 2x rtx3090 on unsloth q8kxl
may be you can try on your side to confirm!!

artden111

about 19 hours ago

I think this model is better than Qwen 3.6 27B. It analyzes the source code architecture more deeply and doesn't speak inarticulately like Qwen.

omaryshchenko

about 13 hours ago

I noticed that Glimmer keeps researching(command, read, grep etc) without results while Qwen found answer in five steps(used internal tool and then just read needed files and based on own knowledge proposed solution). It is good that Glimmer knows that there is no answer in context but it needs smarter research logic and usage of available tools.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment