marvinscham/masterthesis-playground
1
0
Fork 0
mirror of https://github.com/marvinscham/masterthesis-playground.git synced 2026-03-22 08:22:43 +01:00
Code Issues Packages Projects Releases Wiki Activity

Cleanup

Browse Source
This commit is contained in:
Marvin Scham
2026-02-24 23:39:14 +01:00
parent a2967767d3
commit 0d2807f59f
15 changed files with 648 additions and 10319 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
raft/README.md
View File

@@ -18,12 +18,6 @@ python prepare_corpus.py --input_csv ../data/intermediate/culture_reviews.csv --
python make_raft_data.py --out_dir out --n_examples 10
```
## Training der QLoRA-Adapter
```bash
python train_mistral_raft.py --train_jsonl out/raft_train.jsonl --out_dir out/mistral_balitwin_lora
```
## Inferenz
### Pre-Merged Modell + Adapter
@@ -31,11 +25,3 @@ python make_raft_data.py --out_dir out --n_examples 10
```bash
python rag_chat_merged.py --model_dir /path/to/model_folder --out_dir out
```
### Per Baseline Mistral 7B + PEFT-Adapter
Hinweis: das Skript wurde nach wenigen oberflächlichen Evaluationsrunden nicht weiter verwendet, da der beste Kandidat durch einen Merge des Basismodells und seiner PEFT-Adapter beschleunigt werden konnte und dieses Skript nicht länger relevant war.
```bash
python deprecated_rag_chat.py --lora_dir out/mistral_balitwin_lora
```