paddlepaddle/paddlenlp

How paddlepaddle/paddlenlp works

PaddleNLP is a Paddle-based NLP and large-language-model development suite spanning data preparation, model training and alignment, compression, inference, and application serving. A representative workflow is to prepare data, train or fine-tune a supported model, optionally compress it, then run inference or deploy a service. The README documents distributed training, hardware diversity, PEFT, and high-performance inference; sampled sources confirm selected pipeline, engine, and serving relationships. Other subsystem links are documented capability-level connections because their implementation was not sampled.

Main entry points: run_pretrain.py LLM training entry · engine.py LLM runtime · transformers Model families · trainer.py Trainer

Components 22

Data and models

Dataset loaders paddlenlp/datasets/dataset.py
Data utilities paddlenlp/data
Model families paddlenlp/transformers

Training and alignment

LLM training entry llm/run_pretrain.py
Trainer paddlenlp/trainer/trainer.py
Parallel engine slm/model_zoo/gpt-3/ppfleetx/core/engine/auto_engine.py
Eager engine slm/model_zoo/gpt-3/ppfleetx/core/engine/eager_engine.py
Unified checkpoint paddlenlp/trainer/unified_checkpoint/async_handler.py
PEFT methods paddlenlp/peft
Alignment workflows llm/alignment

Compression and kernels

Quantization llm/run_quantization.py
Optimized operators ops/src/paddlenlp_kernel

Inference and serving

Generation APIs paddlenlp/generation/utils.py
LLM runtime llm/server/server/server/engine/engine.py
Resource manager llm/server/server/server/engine/resource_manager.py
Task queue llm/server/server/server/engine/task_queue_manager.py
Token processor llm/server/server/server/engine/token_processor.py
HTTP server llm/server/server/server/http_server/app.py

Applications and tooling

NLP pipelines slm/pipelines/pipelines/pipelines/standard_pipelines.py
SimpleServer paddlenlp/server/server.py
PaddleNLP CLI paddlenlp/cli/main.py

Other components

Developer

Connections 22

Developer → LLM training entry: starts training README.md
Dataset loaders → Data utilities: provides data README.md
Data utilities → LLM training entry: feeds training README.md
Model families → LLM training entry: trains models README.md
LLM training entry → Trainer: runs training README.md
LLM training entry → Parallel engine: supports parallelism README.md
Parallel engine → Eager engine: offers engine README.md
Trainer → Unified checkpoint: saves checkpoints README.md
Trainer → PEFT methods: supports fine-tuning README.md
LLM training entry → Alignment workflows: supports alignment README.md
Model families → Quantization: compresses models README.md
Quantization → LLM runtime: enables inference README.md
Optimized operators → LLM runtime: accelerates inference README.md
Model families → Generation APIs: generates outputs README.md
HTTP server → LLM runtime: routes requests llm/server/server/server/http_server/app.py
LLM runtime → Resource manager: uses resources llm/server/server/server/engine/engine.py
LLM runtime → Task queue: uses queue llm/server/server/server/engine/engine.py
LLM runtime → Token processor: uses processor llm/server/server/server/engine/engine.py
Developer → NLP pipelines: builds applications README.md
NLP pipelines → Model families: retrieves answers README.md
Developer → SimpleServer: deploys services README.md
Developer → PaddleNLP CLI: uses commands paddlenlp/cli/main.py

More Python diagrams: coleam00/context-engineering-intro · startrail-org/leann · jiayi-pan/tinyzero · pypa/pipx · pyinstaller/pyinstaller

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·