GPT-oss-MOFinder-enriched: chemical info expansion
This repository gives the attempts expanding our default assembly, 8-field chemical reasoning, to more wet-lab procedural details feeding a GPT base model. We reports the performance of finetuned GPT opensource models with 9 or even longer prompting on MOF synthesizability task.
Result
| Train fields | Inference | Test size | Accuracy | Precision |
|---|---|---|---|---|
| add "stirring" field | Holdout | 1320P+ 1275N | 0.8347 | 0.8113 |
| 22 questions | 11P + 11N | 18/22 | 0.7333 | |
| add "vessel_type, vessel_volume_mL, and agitation" fields | Holdout | 1320P+ 1275N | 0.8324 | 0.8310 |
| 22 questions | 11P + 11N | 18/22 | 0.7333 | |
| default 8-field MOFinder | Holdout | 1320P+ 1275N | 0.8301 | 0.8273 |
| 22 questions | 11P + 11N | 20/22 | 0.8462 |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "GPT-oss-MOFinder-enriched"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
To go throught the quick start details for reprduction, see the official released MOFinder main checkpoint which is well-trained on full-field PN data.
Citation and license
When using these weights, cite MOFinder, identify StarLiu714/GPT-oss-MOF and the Hub revision used, and reference the GPT-oss-20B base model.
This release is licensed under the GNU General Public License v3.0; see LICENSE.
The underlying GPT-oss-20B model retains its upstream Apache-2.0 license, reproduced in LICENSE.base-model. This release includes MOF fine-tuning changes merged into the base model and a BF16 weight export.
MOFinder codebase is separately licensed under MIT.