PetJev-600M
PetJev-600M is PawChrona Team's multimodal decision model, fine-tuned at the omni level from d1-omni-600M using PawChrona-MMQA. It accepts text, images, video frames or audio and returns decisions with candidate probabilities in one forward pass.
PetJev-600M 是 PawChrona Team 基于 d1-omni-600M，使用 PawChrona-MMQA 进行 omni 层级微调的宠物多模态决策模型。支持文本、图像、视频帧或音频输入，单次前向输出决策及候选概率。
Usage / 使用
import torch
from PIL import Image
from transformers import AutoModel
model = AutoModel.from_pretrained(
"PawChrona/PetJev-600M", trust_remote_code=True, dtype=torch.float32
).eval()
image = Image.open("pet.jpg").convert("RGB")
question = {
"type": "choice",
"instructions": "Which activity is visible?", # 图中显示哪种活动？
"criteria": {"rest": "Resting", "eat": "Eating", "play": "Playing", "walk": "Walking"},
}
result = model.system_one(None, {"activity": question}, images=[image])
print(result["answers"]["activity"])
The repository contains complete model weights. Choice probabilities include PawChrona's global temperature calibration. Request access to download; requests are manually reviewed. Use is subject to the included LFM Open License v1.0.
仓库提供完整模型权重；选择题概率包含 PawChrona 全局温度校准。下载需提交访问申请并经人工审核；使用遵循仓库内 LFM Open License v1.0。
Citation / 引用
@misc{pawchrona2026petjev600m,
author = {{PawChrona Team}},
title = {PetJev-600M},
year = {2026},
url = {https://huggingface.co/PawChrona/PetJev-600M}
}
- Downloads last month
- -
Model tree for PawChrona/PetJev-600M
Base model
LiquidAI/LFM2.5-350M-Base Finetuned
LiquidAI/LFM2.5-Encoder-350M Finetuned
LiquidAI/d1-omni-600M