Update README.md
#2
by datdang77 - opened
README.md
CHANGED
|
@@ -1,3 +1,48 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
| 4 |
+
# MatInvent — Retrieval-Retro / rr_thermo assets
|
| 5 |
+
|
| 6 |
+
Model weights and reference data for the `retrieval_retro` and `rr_thermo` reward
|
| 7 |
+
calculators in MatInvent. These are too large for git (~1.6 GB) and are fetched at
|
| 8 |
+
runtime with `hf_hub_download`.
|
| 9 |
+
|
| 10 |
+
Files are byte-identical between the two calculators, so one copy serves both, except
|
| 11 |
+
`MP_Energetics.json`, which only `rr_thermo` uses.
|
| 12 |
+
|
| 13 |
+
`random_mpc_pool.pt` and `embed/mpc_embeddings_pool.pt` concatenate the original
|
| 14 |
+
train/valid/test splits **in that order** (28434 rows: 0–22746 / 22747–25589 /
|
| 15 |
+
25590–28433). The order is load-bearing — `train_dataset[i]` must stay aligned with
|
| 16 |
+
row `i` of the embedding matrix.
|
| 17 |
+
|
| 18 |
+
## Source
|
| 19 |
+
|
| 20 |
+
### Retrieval-Retro — https://github.com/HeewoongNoh/Retrieval-Retro
|
| 21 |
+
|
| 22 |
+
Dataset downloaded from the link in that repository; the two `*_pool` files are our
|
| 23 |
+
merge of its train/valid/test splits.
|
| 24 |
+
|
| 25 |
+
- `random_mpc_pool.pt` (1.3G) — retrieval pool
|
| 26 |
+
- `mpc_embeddings_pool.pt` (3.5M) — retrieval embeddings
|
| 27 |
+
- `element_embeddings.json` (432K) — element vocabulary
|
| 28 |
+
- `random_template.json` (7.4K) — index → formula map
|
| 29 |
+
- `random_precursor_formation_energy.pt` (4.5K) — precursor formation energies
|
| 30 |
+
|
| 31 |
+
> Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge. NeurIPS 2024.
|
| 32 |
+
|
| 33 |
+
### Trained here
|
| 34 |
+
|
| 35 |
+
Trained on the dataset above using the Retrieval-Retro code. Hyperparameters are encoded
|
| 36 |
+
in the filenames (split, batch size, learning rate, hidden size, seed).
|
| 37 |
+
|
| 38 |
+
- `Retrieval_Retro_128_ours_Retrieval_Retro_random_3_best.pt` (46M) — main model
|
| 39 |
+
- `TL_pretrain(formation_exp)_embedder(graphnetwork)_lr(0.0005)_batch_size(256)_hidden(256)_seed(0)_.pt` (17M) — NRE formation-energy model
|
| 40 |
+
- `mpc_best.pt` (373K) — MPC retriever
|
| 41 |
+
|
| 42 |
+
### ARROWS — https://github.com/njszym/ARROWS
|
| 43 |
+
|
| 44 |
+
Redistributed under MIT (© Nathan J. Szymanski). Contents are Materials Project
|
| 45 |
+
formation energies (Bartel-corrected), CC-BY 4.0.
|
| 46 |
+
- `ARROWS/arrows/energetics/MP_Energetics.json` (216M) — read at import time
|
| 47 |
+
|
| 48 |
+
> N. J. Szymanski et al. *Nature Communications* (2023). https://doi.org/10.1038/s41467-023-42329-9
|