Instructions to use mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 with Transformers:

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

pipe = pipeline("text-generation", model="mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128")
model = AutoModel.from_pretrained("mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128

SGLang

How to use mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 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 "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128" \
    --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": "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128" \
        --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": "mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 with Docker Model Runner:
```
docker model run hf.co/mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128
```

Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128

File size: 2,074 Bytes

8a61e51
 
8c59d0d
8a61e51
8c59d0d
 
 
 
 
 
8a61e51
 
8c59d0d
8a61e51
8c59d0d
8a61e51
8c59d0d
8a61e51
8c59d0d
 
 
8a61e51
d3296b5
 
 
 
 
 
 
 
 
 
 
 
593f04b
d3296b5
 
 
8c59d0d

---
license: apache-2.0
base_model: meta-models/Muse-Glimmer-30B
library_name: transformers
pipeline_tag: text-generation
tags:
  - gptq
  - vllm
  - dflash
  - muse-glimmer
---

# Muse Glimmer DFlash assistant GPTQ Int4 G128

GPTQ 4-bit / group-128 / symmetric of Muse Glimmer's **5-layer DFlash drafter**. Apache 2.0, derived from Meta's DFlash assistant (not a standalone chat model).

Use as the speculative draft for [vLLM-XPU on one Arc Pro B70](https://github.com/mgaruccio/muse-glimmer-b70) with [`Muse-Glimmer-30B-GPTQ-Int4-sym-G128`](https://huggingface.co/mgaruccio/Muse-Glimmer-30B-GPTQ-Int4-sym-G128). Embeddings and lm_head are shared with the target at runtime.

- Quantizer: GPTQModel 7.3.2
- 35 decoder linears quantized; encoder / norms BF16
- vLLM fused module names in `quantize_config.json`: `qkv_proj`, `o_proj`, `gate_up_proj`, `down_proj`

## Experimental B70 concurrency sweep artifact

The C8–C128 short-burst sweep used this assistant with an experimental frozen
32,768-token vocabulary shortlist and DFlash K3. Download the public-safe
runtime input from [`artifacts/glimmer-b70-k3-shortlist-32768.json`](artifacts/glimmer-b70-k3-shortlist-32768.json).
It contains only the immutable original-vocabulary IDs and contract metadata—no
calibration prompts, votes, raw generations, or model weights.

This is **not** the default recipe or a production capacity claim. Its highest
observed short-burst median was 840.8 aggregate tok/s at C96; C48 was the
workload-specific latency/throughput knee. The measured workload used repeated
83-token prompts and capped 256-token reasoning-only outputs. See the
[presentation report](https://github.com/mgaruccio/muse-glimmer-b70/blob/38ea47b202135c22effa9d844ce73eb5603ab2a1/docs/concurrency-sweep.md)
for methodology, caveats, and the exact source patch at
[`b70-inference@faf4ba9`](https://github.com/mgaruccio/b70-inference/tree/faf4ba9889254719c878728fdd1a48806e7fd88b/scripts/experimental).

```bash
hf download mgaruccio/Muse-Glimmer-30B-assistant-GPTQ-Int4-sym-G128 --local-dir ./models/draft
```