README
š
Embedl - efficient AI for the edge
Jonna Matthiesen
JonnaMat
AI & ML interests
None yet
Recent Activity
new activity about 23 hours ago
embedl/Qwen3-0.6B-FlashHead:vllm serve updated a Space 1 day ago
embedl/README published an article 1 day ago
How to Build a vLLM Plugin: A Guide to the general_plugins Entry PointOrganizations
vllm serve
4
#2 opened 1 day ago
by 
ParisXu
updated a
Space 1 day ago
published an article 1 day ago
Article
How to Build a vLLM Plugin: A Guide to the general_plugins Entry Point
posted an update 1 day ago
Post
59
ā” FlashHead: Fast LM Head Inference - Now a Simple vLLM Plugin
flash-head replaces the dense LM head with a two-stage retrieval pipeline - up to 2x inference speedup, training-free. Previously required custom Docker images; now it's just:
āØ The plugin activates automatically via vLLM's
š§© Supported models (full collection):
Qwen Qwen3, 
meta-llama Llama3, 
google Gemma3, 
nvidia Cosmos-Reason2 - BF16 and W4A16 variants.
https://huggingface.co/collections/embedl/flashhead
š embedl/Edge-Inference-Benchmarks
š§ Benchmark it yourself:
FlashHead shines at low batch sizes; the typical real-time / on-device use case. š
flash-head replaces the dense LM head with a two-stage retrieval pipeline - up to 2x inference speedup, training-free. Previously required custom Docker images; now it's just:
pip install flash-head
vllm serve embedl/Qwen3-1.7B-FlashHead-W4A16āØ The plugin activates automatically via vLLM's
vllm.general_plugins entry point. No source patches, no custom imports.š§© Supported models (full collection):
https://huggingface.co/collections/embedl/flashhead
š embedl/Edge-Inference-Benchmarks
š§ Benchmark it yourself:
vllm bench latency --model embedl/Qwen3-1.7B-FlashHead-W4A16 --batch-size 1
# Baseline comparison
FLASHHEAD_ENABLED=0 vllm bench latency --model embedl/Qwen3-1.7B-FlashHead-W4A16 --batch-size 1FlashHead shines at low batch sizes; the typical real-time / on-device use case. š
updated a
collection 1 day ago
updated 4
models 1 day ago