transformers

Name: transformers
Author: itsmostafa

16installs0forks

How to install transformers?

itsmostafa/llm-engineering-skills/transformers

$npx -y skills add itsmostafa/llm-engineering-skills --skill transformers

Installs into the current project.

Summary

Loading and using pretrained models with Hugging Face Transformers. Use when working with pretrained models from the Hub, running inference with Pipeline API, fine-tuning models with Trainer, or handling text, vision, audio, and multimodal tasks.

Files · 1

View on GitHub

SKILL.md

1	# Using Hugging Face Transformers
2
3	Transformers is the model-definition framework for state-of-the-art machine learning across text, vision, audio, and multimodal domains. It provides unified APIs for loading pretrained models, running inference, and fine-tuning.
4
5	## Table of Contents
6
7	- [Core Concepts](#core-concepts)
8	- [Pipeline API](#pipeline-api)
9	- [Model Loading](#model-loading)
10	- [Inference Patterns](#inference-patterns)
11	- [Fine-tuning with Trainer](#fine-tuning-with-trainer)
12	- [Working with Modalities](#working-with-modalities)
13	- [Memory and Performance](#memory-and-performance)
14	- [Best Practices](#best-practices)
15	- [References](#references)
16
17	## Core Concepts
18
19	### The Three Core Classes
20
21	Every model in Transformers has three core components:
22
23	```python
24	from transformers import AutoConfig, AutoModel, AutoTokenizer, AutoProcessor
25
26	# Configuration: hyperparameters and architecture settings
27	config = AutoConfig.from_pretrained("bert-base-uncased")
28
29	# Model: the neural network weights
30	model = AutoModel.from_pretrained("bert-base-uncased")
31
32	# Tokenizer: converts text inputs to tensors
33	tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
34
35	# Processor: unified preprocessing for vision, audio, and multimodal models
36	processor = AutoProcessor.from_pretrained("openai/whisper-large-v3")
37	```
38
39	### The `from_pretrained` Pattern
40
41	All loading uses `from_pretrained()` which handles downloading, caching, and device placement:
42
43	```python
44	from transformers import AutoModelForCausalLM, AutoTokenizer
45	import torch
46
47	model_name = "meta-llama/Llama-3.2-1B"
48
49	tokenizer = AutoTokenizer.from_pretrained(model_name)
50	model = AutoModelForCausalLM.from_pretrained(
51	model_name,
52	dtype=torch.bfloat16,
53	device_map="auto", # Automatic device placement
54	)
55	```
56
57	Transformers v5 examples use `dtype`. On Transformers v4, the equivalent argument is `torch_dtype`.
58
59	### Auto Classes
60
61	Use task-specific Auto classes for the correct model head:
62
63	```python
64	from transformers import (
65	AutoModelForCausalLM, # Text generation (GPT, Llama)
66	AutoModelForSeq2SeqLM, # Encoder-decoder (T5, BART)
67	AutoModelForSequenceClassification, # Classification
68	AutoModelForTokenClassification, # NER, POS tagging
69	AutoModelForQuestionAnswering, # Extractive QA
70	AutoModelForMaskedLM, # BERT-style masked LM
71	AutoModelForImageClassification, # Vision models
72	AutoModelForSpeechSeq2Seq, # Speech recognition
73	)
74	```
75
76	## Pipeline API
77
78	The `pipeline()` function provides high-level inference with minimal code:
79
80	### Text Tasks
81
82	```python
83	from transformers import pipeline
84
85	# Text generation
86	generator = pipeline("text-generation", model="Qwen/Qwen2.5-1.5B")
87	output = generator("The secret to success is", max_new_tokens=50)
88
89	# Text classification
90	classifier = pipeline("sentiment-analysis")
91	result = classifier("I love this product!")
92	# [{'label': 'POSITIVE', 'score': 0.9998}]
93
94	# Named entity recognition
95	ner = pipeline("ner", aggregation_strategy="simple")
96	entities = ner("Hugging Face is based in New York City.")
97
98	# Question answering
99	qa = pipeline("question-answering")
100	answer = qa(question="What is the capital?", context="Paris is the capital of France.")
101
102	# Summarization
103	summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
104	summary = summarizer(long_text, max_length=130, min_length=30)
105
106	# Translation
107	translator = pipeline("translation_en_to_fr", model="Helsinki-NLP/opus-mt-en-fr")
108	result = translator("Hello, how are you?")
109	```
110
111	### Chat/Conversational
112
113	```python
114	from transformers import pipeline
115	import torch
116
117	pipe = pipeline(
118	"text-generation",
119	model="meta-llama/Llama-3.2-3B-Instruct",
120	dtype=torch.bfloat16,
121	device_map="auto",
122	)
123
124	messages = [
125	{"role": "system", "content": "You are a helpful assistant."},
126	{"role": "user", "content": "Explain quantum computing in simple terms."},
127	]
128
129	response = pipe(messages, max_new_tokens=256)
130	print(response[0]["generated_text"][-1]["content"])
131	```
132
133	### Vision Tasks
134
135	```python
136	classifier = pipeline("image-classification", model="google/vit-base-patch16-224")
137	detector = pipeline("object-detection", model="facebook/detr-resnet-50")
138	```
139
140	### Audio Tasks
141
142	```python
143	transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3")
144	text = transcriber("path/to/audio.mp3")
145	```
146
147	### Multimodal Tasks
148
149	```python
150	vqa = pipeline("visual-question-answering", model="Salesforce/blip-vqa-base")
151	captioner = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")
152	```
153
154	## Model Loading
155
156	### Device Placement
157
158	```python
159	from transformers import AutoModelForCausalLM
160	import torch
161
162	# Automatic placement across available de

People who liked this also viewed

6 picks

Type

transformers

16installs0forks

View on GitHub

How to install transformers?

itsmostafa/llm-engineering-skills/transformers

$npx -y skills add itsmostafa/llm-engineering-skills --skill transformers

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Using Hugging Face Transformers
2
3	Transformers is the model-definition framework for state-of-the-art machine learning across text, vision, audio, and multimodal domains. It provides unified APIs for loading pretrained models, running inference, and fine-tuning.
4
5	## Table of Contents
6
7	- [Core Concepts](#core-concepts)
8	- [Pipeline API](#pipeline-api)
9	- [Model Loading](#model-loading)
10	- [Inference Patterns](#inference-patterns)
11	- [Fine-tuning with Trainer](#fine-tuning-with-trainer)
12	- [Working with Modalities](#working-with-modalities)
13	- [Memory and Performance](#memory-and-performance)
14	- [Best Practices](#best-practices)
15	- [References](#references)
16
17	## Core Concepts
18
19	### The Three Core Classes
20
21	Every model in Transformers has three core components:
22
23	```python
24	from transformers import AutoConfig, AutoModel, AutoTokenizer, AutoProcessor
25
26	# Configuration: hyperparameters and architecture settings
27	config = AutoConfig.from_pretrained("bert-base-uncased")
28
29	# Model: the neural network weights
30	model = AutoModel.from_pretrained("bert-base-uncased")
31
32	# Tokenizer: converts text inputs to tensors
33	tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
34
35	# Processor: unified preprocessing for vision, audio, and multimodal models
36	processor = AutoProcessor.from_pretrained("openai/whisper-large-v3")
37	```
38
39	### The `from_pretrained` Pattern
40
41	All loading uses `from_pretrained()` which handles downloading, caching, and device placement:
42
43	```python
44	from transformers import AutoModelForCausalLM, AutoTokenizer
45	import torch
46
47	model_name = "meta-llama/Llama-3.2-1B"
48
49	tokenizer = AutoTokenizer.from_pretrained(model_name)
50	model = AutoModelForCausalLM.from_pretrained(
51	model_name,
52	dtype=torch.bfloat16,
53	device_map="auto", # Automatic device placement
54	)
55	```
56
57	Transformers v5 examples use `dtype`. On Transformers v4, the equivalent argument is `torch_dtype`.
58
59	### Auto Classes
60
61	Use task-specific Auto classes for the correct model head:
62
63	```python
64	from transformers import (
65	AutoModelForCausalLM, # Text generation (GPT, Llama)
66	AutoModelForSeq2SeqLM, # Encoder-decoder (T5, BART)
67	AutoModelForSequenceClassification, # Classification
68	AutoModelForTokenClassification, # NER, POS tagging
69	AutoModelForQuestionAnswering, # Extractive QA
70	AutoModelForMaskedLM, # BERT-style masked LM
71	AutoModelForImageClassification, # Vision models
72	AutoModelForSpeechSeq2Seq, # Speech recognition
73	)
74	```
75
76	## Pipeline API
77
78	The `pipeline()` function provides high-level inference with minimal code:
79
80	### Text Tasks
81
82	```python
83	from transformers import pipeline
84
85	# Text generation
86	generator = pipeline("text-generation", model="Qwen/Qwen2.5-1.5B")
87	output = generator("The secret to success is", max_new_tokens=50)
88
89	# Text classification
90	classifier = pipeline("sentiment-analysis")
91	result = classifier("I love this product!")
92	# [{'label': 'POSITIVE', 'score': 0.9998}]
93
94	# Named entity recognition
95	ner = pipeline("ner", aggregation_strategy="simple")
96	entities = ner("Hugging Face is based in New York City.")
97
98	# Question answering
99	qa = pipeline("question-answering")
100	answer = qa(question="What is the capital?", context="Paris is the capital of France.")
101
102	# Summarization
103	summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
104	summary = summarizer(long_text, max_length=130, min_length=30)
105
106	# Translation
107	translator = pipeline("translation_en_to_fr", model="Helsinki-NLP/opus-mt-en-fr")
108	result = translator("Hello, how are you?")
109	```
110
111	### Chat/Conversational
112
113	```python
114	from transformers import pipeline
115	import torch
116
117	pipe = pipeline(
118	"text-generation",
119	model="meta-llama/Llama-3.2-3B-Instruct",
120	dtype=torch.bfloat16,
121	device_map="auto",
122	)
123
124	messages = [
125	{"role": "system", "content": "You are a helpful assistant."},
126	{"role": "user", "content": "Explain quantum computing in simple terms."},
127	]
128
129	response = pipe(messages, max_new_tokens=256)
130	print(response[0]["generated_text"][-1]["content"])
131	```
132
133	### Vision Tasks
134
135	```python
136	classifier = pipeline("image-classification", model="google/vit-base-patch16-224")
137	detector = pipeline("object-detection", model="facebook/detr-resnet-50")
138	```
139
140	### Audio Tasks
141
142	```python
143	transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3")
144	text = transcriber("path/to/audio.mp3")
145	```
146
147	### Multimodal Tasks
148
149	```python
150	vqa = pipeline("visual-question-answering", model="Salesforce/blip-vqa-base")
151	captioner = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")
152	```
153
154	## Model Loading
155
156	### Device Placement
157
158	```python
159	from transformers import AutoModelForCausalLM
160	import torch
161
162	# Automatic placement across available de

People who liked this also viewed

6 picks

Type