Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen angezeigt.

--- transformer [2025/11/12 22:16]
jango angelegt
+++ transformer [2026/03/13 23:45] (aktuell)
jango ↷ Seitename wurde von transformers auf transformer geändert
@@ Zeile 1: / Zeile 1: @@
+Transformers ist eine spezielle Architektur von [[LLM]] Modellen und auch eine [[coding:python|Python]] Bibliothek für [[machine_learning|Machine learning]]. Siehe auch [[gpt|Generative pretrained transformers (GPT)]].
+=====Text generation=====
+<code python>
+from transformers import GPT2Tokenizer, GPT2LMHeadModel, pipeline
+tokenizer = GPT2Tokenizer.from_pretrained("gpt2-large", cache_dir="cache")
+model = GPT2LMHeadModel.from_pretrained("gpt2-large", cache_dir="cache")
+generator = pipeline(model=model, tokenizer=tokenizer, task="text-generation", max_new_tokens=10, early_stopping=True)
+result = generator("algebra is the science of")
+print(result)
+</code>
+<code python>
+#pip install einops
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import transformers
+import torch
+model = "tiiuae/falcon-40b"
+tokenizer = AutoTokenizer.from_pretrained(model, cache_dir="falcon40b")
+pipeline = transformers.pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True,
+    device_map="auto",
+)
+sequences = pipeline(
+   "AI is going to replace humans because",
+    max_length=100,
+    do_sample=True,
+    top_k=10,
+    num_return_sequences=1,
+    eos_token_id=tokenizer.eos_token_id,
+)
+for seq in sequences:
+    print(f"Result: {seq['generated_text']}")
+</code>
+=====Object detection=====
+Siehe auch [[yolov5|Yolov5]].
+<code python>
+from transformers import DetrImageProcessor, DetrForObjectDetection
+import torch
+from PIL import Image
+import requests
+url = "https://bilder.ottoversand.at/i/empiriecom/c6d7b41b9382cccd4b131c48a1306ac4"
+image = Image.open(requests.get(url, stream=True).raw)
+processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-101")
+model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-101")
+inputs = processor(images=image, return_tensors="pt")
+outputs = model(**inputs)
+# convert outputs (bounding boxes and class logits) to COCO API
+# let's only keep detections with score > 0.9
+target_sizes = torch.tensor([image.size[::-1]])
+results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]
+for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
+    box = [round(i, 2) for i in box.tolist()]
+    print(f"Detected {model.config.id2label[label.item()]} with confidence {round(score.item(), 3)} at location {box}")
+</code>
+<code python>
+from transformers import pipeline
+from PIL import Image, ImageDraw, ImageFont
+from transformers import DetrFeatureExtractor, DetrForObjectDetection
+#classifier = pipeline("sentiment-analysis")
+#classifier = pipeline("feature-extraction")
+#classifier = pipeline("question-answering")
+#classifier = pipeline("text-generation", model="gpt2")
+feature_extractor = DetrFeatureExtractor.from_pretrained('facebook/detr-resnet-101')
+model = DetrForObjectDetection.from_pretrained('facebook/detr-resnet-101')
+object_detector = pipeline("object-detection", model = model, feature_extractor = feature_extractor)
+def draw_bounding_box(im, score, label, xmin, ymin, xmax, ymax, index, num_boxes):
+    im_with_rectangle = ImageDraw.Draw(im)
+    im_with_rectangle.rounded_rectangle((xmin, ymin, xmax, ymax), outline = "red", width = 5, radius = 10)
+    im_with_rectangle.text((xmin+10, ymin+10), label, fill="white", stroke_fill = "red")
+    return im, label
+with Image.open("test.jpg") as im:
+    bounding_boxes = object_detector(im)
+    num_boxes = len(bounding_boxes)
+    index = 0
+    labels = []
+    for bounding_box in bounding_boxes:
+        box = bounding_box["box"]
+        im, label = draw_bounding_box(im, bounding_box["score"], bounding_box["label"], box["xmin"], box["ymin"], box["xmax"], box["ymax"], index, num_boxes)
+        #if label not in labels:
+        labels.append(label)
+        index += 1
+    #im.save("boxes.jpg")
+    im.show()
+    print(labels)
+</code>
+<code python>
+from transformers import ViTImageProcessor, ViTForImageClassification
+from PIL import Image
+import requests
+url = 'https://rothys.com/cdn/shop/files/05.31_HP_3_Desktop_9nsINbZq8H.jpg?v=1685080183&width=1024'
+image = Image.open(requests.get(url, stream=True).raw)
+processor = ViTImageProcessor.from_pretrained('google/vit-base-patch16-224')
+model = ViTForImageClassification.from_pretrained('google/vit-base-patch16-224')
+inputs = processor(images=image, return_tensors="pt")
+outputs = model(**inputs)
+logits = outputs.logits
+# model predicts one of the 1000 ImageNet classes
+predicted_class_idx = logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
+</code>
+=====Links=====
+  * [[https://www.youtube.com/watch?v=wjZofJX0v4M|Overview]]
+  * [[https://www.youtube.com/watch?v=eMlx5fFNoYc|Attention]]

MBCDN

Benutzer-Werkzeuge

Webseiten-Werkzeuge

Unterschiede

Seiten-Werkzeuge