Spaces:
Sleeping
Sleeping
File size: 2,352 Bytes
d081bf3 d82a7f0 95595f1 d081bf3 d2ce059 afac99d 95595f1 afac99d 95595f1 afac99d eb6d527 95595f1 eb6d527 9fc69b3 afac99d eb6d527 95595f1 d081bf3 d82a7f0 d2ce059 eb6d527 95595f1 eb6d527 afac99d d081bf3 a966c99 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | import gradio as gr
import base64
from PIL import Image
import io
import json
import torch
from transformers import AutoModelForVision2Seq, AutoProcessor
# ------------------------------------------------------------
# 1. Load VLLM Model (Qwen3-VL-8B-Instruct)
# ------------------------------------------------------------
model_name = "Qwen/Qwen2-VL-7B-Instruct" # HF 官方推薦名稱（VL）
processor = AutoProcessor.from_pretrained(model_name)
model = AutoModelForVision2Seq.from_pretrained(
model_name,
torch_dtype=torch.float16,
low_cpu_mem_usage=True
).to("cuda")
# ------------------------------------------------------------
# 2. Main Process Function
# ------------------------------------------------------------
def process(payload):
try:
# 取得資料
data = payload
img_bytes = base64.b64decode(data["image_b64"])
img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
# ------------------------------------------------------------
# 3. Vision-Language model inference
# ------------------------------------------------------------
prompt = "Describe what you see in this image in detail."
inputs = processor(images=img, text=prompt, return_tensors="pt").to("cuda", torch.float16)
output_ids = model.generate(
**inputs,
max_new_tokens=200,
temperature=0.2
)
response_text = processor.batch_decode(output_ids, skip_special_tokens=True)[0]
# ------------------------------------------------------------
# 4. Return results to Jetson
# ------------------------------------------------------------
reply = {
"received": True,
"robot_id": data.get("robot_id"),
"size": img.size,
"vllm_analysis": response_text
}
return reply
except Exception as e:
return None, {"error": str(e)}
# ------------------------------------------------------------
# 5. Gradio UI
# ------------------------------------------------------------
demo = gr.Interface(
fn=process,
inputs=gr.JSON(label="Input Payload (Dict format)"),
outputs=[
gr.Image(type="pil", label="Image Preview"),
gr.JSON(label="Reply to Jetson")
],
api_name="predict"
)
demo.launch()
|