Jonna Matthiesen

JonnaMat's activity

• All
• Models
• Datasets
• Spaces
• Buckets
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

updated 2 collections 2 days ago

Qwen3.5

Collection

Qwen/Qwen3.5 variants optimized by embedl. • 5 items • Updated 1 day ago

FlashHead

Collection

Efficient Drop-In Replacement for the Classification Head in Language Model Inference. https://github.com/embedl/flash-head • 23 items • Updated 2 days ago • 2

New activity in embedl/Edge-Inference-Benchmarks 6 days ago

Qwen3.5-FlashHead

#6 opened 6 days ago by

JonnaMat

updated a Space 8 days ago

Running

6

Edge Inference Benchmarks

🚀

6

On-Device benchmarks across devices and models.

liked a model 13 days ago

embedl/Cosmos-Reason2-2B-W4A16-Edge2-FlashHead

Image-Text-to-Text • 2B • Updated 5 days ago • 1.71k • 7

posted an update 14 days ago

view post

Post

113

⚡ FlashHead benchmarks for Llama 3.2, Gemma 3, and Qwen3 are now on embedl/Edge-Inference-Benchmarks !
These are some of the models used in the FlashHead paper - now easier to explore and compare interactively.

🚀 Jetson AGX Thor (tok/s, batch=1):
- Llama-3.2-1B: 77 → 285 (FlashHead+W4A16, 3.7x)
- Llama-3.2-3B: 34 → 112 (3.3x)
- Gemma-3-1B: 79 → 153 (1.9x)
- Qwen3-1.7B: 49 → 189 (3.8x)
- Qwen3-0.6B: 140 → 177 (1.3x)

✅ Accuracy matches baseline on MMLU-Pro, IFEval, BBH, TruthfulQA, GSM8K.

See translation

• 

🚀

1

1

+

Reply

updated 6 models 14 days ago

embedl/Qwen3-1.7B-FlashHead-W4A16

2B • Updated 5 days ago • 151 • 3

embedl/gemma-3-1b-it-FlashHead-W4A16

1B • Updated 5 days ago • 148 • 3

embedl/Llama-3.2-3B-Instruct-FlashHead-W4A16

4B • Updated 5 days ago • 113 • 4

embedl/Llama-3.2-1B-Instruct-FlashHead-W4A16

2B • Updated 5 days ago • 60 • 6

embedl/Cosmos-Reason2-2B-W4A16-Edge2-FlashHead

Image-Text-to-Text • 2B • Updated 5 days ago • 1.71k • 7

embedl/Qwen3-1.7B-FlashHead

2B • Updated 5 days ago • 123 • 3