Dedicated Endpoints
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Container
Run this model inference with full control and performance in your environment.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
License: other模型
- QLoRA(4-bit NF4 冻结基座 + LoRA on LLM proj + 视觉塔 blocks,~47.6M 可训)。
- 无自定义头:读 answer 位 6 个数字 token("0"…"5")softmax,期望值 = 分数。
- 训练:60k 样本,DeQA 软标签(高斯 σ=0.75)CE。
- 输入:
[通道图, 渲染图, base_color]+ 每通道评分准则 prompt。
结果(old-test 4917, SRCC)
| mean | base | normal | rough | metallic |
|---|---|---|---|---|
| 0.792 | 0.832 | 0.801 | 0.902 | 0.631 |
1/16 数据暴露追平 DINOv2-0.31B。读出可换温度(T≈0.25)以恢复两端、提升"守护精品"。
用法
python
from transformers import AutoModelForImageTextToTextfrom peft import PeftModelbase = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", load_in_4bit=True)model = PeftModel.from_pretrained(base, "<this-repo>")# 推理:构造 [channel, render, base_color] + prompt,读 6 数字 token softmax 取期望# 完整逻辑见代码仓库 vlm_scorer_eval.py
局限
同 DINOv2 卡:聚合 SRCC 受单人标注噪声封顶 ~0.79;metallic 最弱;训练数据私有不分发。
License
继承 Qwen2.5-VL 基座许可;本 adapter 为衍生物,研究用途。
Model provider
Color2333
Model tree
Base
Qwen/Qwen2.5-VL-7B-Instruct
Adapter
this model
Modalities
Input
Text, Image
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information