Instructions to use meta-models/Muse-Glimmer-30B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
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
Render a bare tool name as an exact recipient, not a namespace pattern
A client that offers unqualified tool names cannot address any of them.
The symptom
Every agent session opens with a rejected tool call. The model emits read.filePath, bash.command, glob.pattern -- the tool name with one of its own parameter names glued on. The harness has no such tool, rejects the call, and the model recovers on the retry, so the session works and every single task pays for a wasted round trip.
Where it comes from
render_system_meta derives a namespace from every offered tool and renders it as a wildcard:
{%- set tns = fn.name.split('.')[0] -%}
...
{%- set rns.recipients = rns.recipients + ['"' + tns + '.*"'] -%}
split('.')[0] on a name with no dot returns the whole name, so a client offering read, bash and glob gets:
# Valid recipients: "self", "read.*", "bash.*", "glob.*", "user".
Nothing called read is addressable any more -- only things under read. The model is being told the tool it wants does not exist and something inside it does, so it invents a second segment from whatever is at hand. The malformed call is the model following the prompt correctly.
render_tools has the same assumption: it emits a namespace metadata entry for tools that have no namespace.
The fix
Treat the dot as the signal it is. A name containing one is namespaced and keeps its "ns.*" pattern; a name without one is already the whole address and is rendered verbatim.
bare # Valid recipients: "self", "read", "bash", "glob", "user".
namespaced # Valid recipients: "self", "fs.*", "sh.*", "user".
Clients that offer namespaced names are unaffected -- byte-for-byte identical output. The change only reaches names the template was previously mangling.
Verification
Rendered both shapes with jinja2 and diffed the preamble, then ran the patched template end to end through OpenCode against a quantized ONNX export of this model: a file-read task and a multi-tool debugging task (read, edit, shell, todo) both completed with no invalid tool calls, where the unpatched template failed the first call of every task.