YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
ME-LLM — Reproducibility Release
Versioned, manuscript-specific release for "ME-LLM: Multimodal-Enhanced Pretrained
Language Models for Semantic-Aware Time Series Forecasting" (IEEE Transactions on
Big Data). Tag: paper-v1.0.
Every reported value in Section IV is linked to its configuration, seed, and raw result
file through the experiments/ tree:
experiments/ME-LLM/<dataset>/<pred_len>/seed2021/
├── config.template.json # per-value training config
├── results/expected_seed_metrics.json # {MSE, MAE}
├── summary.json # status / best_checkpoint
└── checkpoints/finetune/ # best.ckpt (4 of 24; see Checkpoints)
experiments/— per-value experiment tree (24 = 6 datasets x 4 horizons).expected/— aggregated reported values (CSV + nested JSON).manifests/release_manifest_public.json— top-level release index.configs/— the 24 per-value training configs.scripts/repro/— regeneration scripts.
See RELEASE.md for the full description. The model source code lives on
the main branch of this repository.
Checkpoints
This HuggingFace repository ships the trained checkpoints that are omitted from the
GitHub release. Available checkpoints (pred_len 96 only):
| Dataset | pred_len | File |
|---|---|---|
| ETTh1 | 96 | experiments/ME-LLM/ETTh1/96/seed2021/checkpoints/finetune/best.ckpt |
| ETTh2 | 96 | experiments/ME-LLM/ETTh2/96/seed2021/checkpoints/finetune/best.ckpt |
| ETTm1 | 96 | experiments/ME-LLM/ETTm1/96/seed2021/checkpoints/finetune/best.ckpt |
| Weather | 96 | experiments/ME-LLM/Weather/96/seed2021/checkpoints/finetune/best.ckpt |
The remaining 20 experiment directories (ETTm2, Traffic, and all pred_len
192/336/720) contain only a .gitkeep placeholder. Their checkpoints can be
regenerated via scripts/repro/.
Citation
Please cite the ME-LLM paper when using this repository.