#transformers
12pages tagged with "transformers"
programming
- attention interface— unified attentioninterface for centralized attention method management with runtime switching
- modular transformers— modular transformers architecture for reduced code duplication and easier model contributions
- quantization— 18+ quantization methods as first-class citizens in transformers v5.0
- serving and continuous batching— transformers serve command and continuous batching with paged attention for efficient inference
- small encoder training throughput on consumer gpus— measured training throughput for 5m and 21m parameter transformer encoders across rtx 5070 ti, 4070 ti super, 4060 ti, and zen 4 cpus
- tokenizer backend changes— unified tokenizer backend system eliminating the fast/slow distinction in transformers v5.0
- TokenizersBackend does not exist— how to fix the "Tokenizer class TokenizersBackend does not exist or is not currently imported" error when loading models
- tokenizing raw executables for malware analysis with bbpe— pretrained huggingface-compatible bpe tokenizers for raw binaries — use them for malware classification, binary similarity, and reverse engineering with transformers
- transformers v5.0— hugging face transformers v5.0 release overview, features, and migration guide
- transformers v5.0 migration guide— breaking changes and migration guide for upgrading from transformers v4.x to v5.0
- vllm and sglang integration— transformers as backend for vllm and sglang inference engines with near-native performance