Downloads · 30 days
30
5% of all-time downloads
qwen2.5-vl-7b-eagle3-sgl-en-zh
Rayzl/qwen2.5-vl-7b-eagle3-sgl-en-zh
qwen2.5-vl-7b-eagle3-sgl-en-zh is a machine learning model from Rayzl. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
We trained a qwen2.5-vl-7b eagle3 draft model on 200k data random select from FreedomIntelligence/ALLaVA-4V and FreedomIntelligence/ALLaVA-4V-Chinese with specforge
All-time downloads
575
Public
Parameters
403M
2.7 GB on disk
Likes
1
Public
Repo makeup
Click a slice to open those files.
.safetensors806 MB · 100%
Parameter types
How the weights are stored.
BF16403M · 100%
Model card
From the Hugging Face model README
Introduce
We trained a qwen2.5-vl-7b eagle3 draft model on 200k data random select from FreedomIntelligence/ALLaVA-4V and FreedomIntelligence/ALLaVA-4V-Chinese with specforge
Usage
infer with sglang benchmark with mmstar
start server:
python -m sglang.launch_server --model-path Qwen/Qwen2.5-VL-7B-Instruct --speculative-draft Rayzl/qwen2.5-vl-7b-eagle3-sgl --trust-remote-code --chat-template qwen2-vl --chunked-prefill-size -1 --cuda-graph-max-bs 1 --speculative-algo EAGLE3 --speculative-num-steps 4 --speculative-eagle-topk 6 --speculative-num-draft-tokens 24 --tp 1 --mem-fraction-static 0.7 --host 0.0.0.0 --port 8080
benchmark:
python run_mmstar.py --host http://0.0.0.0 --port 8080 --parallel 1 --num-questions 100
- Type
- llama
- License
- mit
- Languages
- en, zh
- Created
- Aug 13, 2025
- Updated
- Sep 8, 2025