thedeepakreddy/pluto-llm

How thedeepakreddy/pluto-llm works

Pluto is a from-scratch PyTorch language-model training system, extended with a synthetic-speech dataset and speech-capable model. The primary workflow prepares and cleans text, trains a byte-level tokenizer, builds token shards, trains or resumes a model, then generates and reports results. The speech workflow synthesizes and encodes aligned audio data, trains SpeechGPT, and evaluates or generates speech. README statements establish the documented stages; source-index references support several internal calls, but unsampled implementations are represented only at their known paths.

Main entry points: cli.py CLI commands · build.py Build speech set · shards.py Token shards · generate.py Speech generation

Components 25

Text data pipeline

Prepare text data src/llm/data/prepare.py
BPE tokenizer src/llm/tokenizer/bpe.py
Token shards src/llm/shards.py

Text model lifecycle

GPT model src/llm/model/gpt.py
Training loop src/llm/train/loop.py
Text generation src/llm/generate.py
Evaluation reports src/llm/report.py

Speech data pipeline

Build speech set src/llm/speech/build.py
Chunk and align src/llm/speech/chunk.py
Speech split store src/llm/speech/store.py
Speech dataset src/llm/speech/dataset.py
Speech interfaces src/llm/speech/interfaces.py
Speech adapters src/llm/speech/adapters.py

Speech model lifecycle

SpeechGPT src/llm/model/voice.py
Speech training src/llm/train/speech_loop.py
Speech generation src/llm/speech/generate.py
Speech evaluation src/llm/speech/evaluate.py
Speech demo src/llm/speech/app.py

Runs and provenance

CLI commands src/llm/cli.py
Run config src/llm/config.py
Checkpoints src/llm/train/checkpoint.py
Artifact manifests src/llm/manifest.py
Run hub sync src/llm/hub.py

Other components

Engineer
Model services

Connections 36

Engineer → CLI commands: runs commands README.md
CLI commands → Run config: loads config src/llm/cli.py
CLI commands → Prepare text data: invokes preparation src/llm/cli.py
Prepare text data → Artifact manifests: records dataset src/llm/data/prepare.py
CLI commands → BPE tokenizer: trains tokenizer src/llm/cli.py
CLI commands → Token shards: builds shards src/llm/cli.py
Token shards → BPE tokenizer: encodes text src/llm/shards.py
Token shards → Artifact manifests: records provenance src/llm/shards.py
CLI commands → Training loop: starts training src/llm/cli.py
Training loop → Token shards: reads batches src/llm/train/loop.py
Training loop → GPT model: trains model src/llm/train/loop.py
Training loop → Checkpoints: saves state src/llm/train/loop.py
CLI commands → Checkpoints: resumes run src/llm/cli.py
CLI commands → Text generation: generates text src/llm/cli.py
CLI commands → Evaluation reports: creates reports src/llm/cli.py
Evaluation reports → Token shards: reads artifacts src/llm/report.py
Evaluation reports → BPE tokenizer: uses tokenizer src/llm/report.py
CLI commands → Run hub sync: pulls run README.md
Build speech set → Prepare text data: reads cleaned text src/llm/speech/build.py
Build speech set → Chunk and align: chunks stories src/llm/speech/build.py
Build speech set → Speech interfaces: creates speech tools src/llm/speech/build.py
Speech interfaces → Speech adapters: loads integrations src/llm/speech/interfaces.py
Speech adapters → Model services: uses TTS codec ASR README.md
Build speech set → Speech split store: writes splits src/llm/speech/build.py
Speech dataset → Speech split store: reads records src/llm/speech/dataset.py
CLI commands → Build speech set: builds speech data src/llm/cli.py
Speech training → Speech dataset: loads training data src/llm/train/speech_loop.py
Speech training → SpeechGPT: trains model src/llm/train/speech_loop.py
Speech training → Speech generation: evaluates samples src/llm/train/speech_loop.py
CLI commands → Speech training: trains speech model README.md
CLI commands → Speech generation: speaks or continues src/llm/cli.py
Speech generation → SpeechGPT: samples frames src/llm/speech/generate.py
CLI commands → Speech evaluation: evaluates speech src/llm/cli.py
Speech evaluation → Speech generation: generates test audio src/llm/speech/evaluate.py
Speech demo → Speech generation: serves generation src/llm/speech/app.py
CLI commands → Speech demo: launches demo README.md

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·