Christian Medina 92999b7783 fix: load model to single GPU (MoE requires all layers on same device)

2026-07-02 17:50:01 -04:00

training

fix: disable gradient checkpointing (device mismatch with distributed MoE)

2026-07-02 17:29:19 -04:00

.gitignore

fix: remove large dataset from git tracking, add to .gitignore

2026-06-30 15:06:13 -04:00

deploy-agenx-lora.sh

fix: use $HOME/loras by default

2026-06-30 15:51:28 -04:00

deploy-and-train.sh

refactor: simplify deploy script, add train-on-this-server.sh

2026-06-30 15:22:54 -04:00

inference.py

refactor: restructure project - scripts at root level

2026-07-01 16:39:54 -04:00

inspect_model.py

docs: add model inspection script and comment failing tests

2026-07-02 13:53:45 -04:00

prepare_dataset.py

refactor: restructure project - scripts at root level

2026-07-01 16:39:54 -04:00

test_model_loading.py

feat: add Test 11 - PEFT prepare + manual 4-bit quantization

2026-07-02 14:30:19 -04:00

train-on-this-server.sh

fix: single-process training, remove FSDP (model pre-distributed via device_map)

2026-07-02 15:53:51 -04:00

train.py

fix: load model to single GPU (MoE requires all layers on same device)

2026-07-02 17:50:01 -04:00

S

Description

LoRA training infrastructure for Cyron summary generation

2.6 MiB

0 Stars 1 Watchers 0 Forks

Languages

Python 94.9%

Shell 5.1%