Added the neccessary info

read_me.md

@@ -1,98 +1,271 @@
-============================================================================================================================================================
+# Diploma Thesis Repository — SFT + DPO Training, Translation, and Safety Evaluation (SK/EN)
-1. LLM_test.py – Generovanie odpovedí modelov																      
 
-	Tento skript je prvý krok celého procesu.
+This repository contains the scripts I used in my thesis experiments to:
-	Spúšťaš ho, keď chceš nechať model (Gemma, LLaMA alebo Qwen) odpovedať na rôzne “nepríjemné” datasety (harmful prompts). Podľa toho sa potom hodnotí jeho bezpečnosť.
 
-1.1 Ako to funguje
+- prepare **PKU-SafeRLHF-30K** for **SFT** and **DPO**,
+- translate datasets and model outputs with **NLLB (SK ↔ EN)**,
+- train QLoRA adapters (SFT and DPO) for selected base models,
+- generate model responses on safety datasets,
+- evaluate responses with **Llama Guard 3**.
 
-	Pri spustení si script od teba vypýta:
+Most scripts assume a local folder layout under `/home/hyrenko/Diploma/...` (models, datasets, outputs). If your paths are different, update the constants at the top of each script.
 
-		aký model chceš použiť,
+---
 
-		aké GPU (ak máš),
+## Repository structure
 
-		aký dataset chceš otestovať,
+The scripts are grouped by purpose:
 
-		koľko promptov chceš spracovať.
+```text
+preparation/
+  prepar_dat_pku_dpo.py
 
-		Dataset si script načíta automaticky.
+program/
+  copymaster.py
+  Llama_test_trained.py
+  LLM_test.py
+  response_evaluate.py
 
-		Ak je gated, vypýta si HF token.
+Training/
+  convert_dpo_sft.py
+  training_dpo_sft_llama.py
+  training_dpo_sft_mistral_sk.py
 
-	Model každému promptu vygeneruje odpoveď a script kontroluje, či to náhodou nebola odpoveď v štýle “nemôžem odpovedať, som AI”. Toto sa počíta ako refusal.
+translate/
+  translate_do-not_answer.py
+  translate_PKF.py
+  Translate_sk_to_eng.py
+```
 
-	Výsledky idú do priečinka outputs/<timestamp>-model-dataset/.
+---
 
-Vo vnútri nájdeš:
+## Requirements
 
-	responses.json – odpovede v strojovom formáte,
+- Python **3.8+**
+- CUDA-capable GPU(s) for translation/training
+- Typical packages:
+  - `torch`, `transformers`, `datasets`
+  - `peft`, `trl`, `accelerate`, `bitsandbytes`
+  - `tqdm`
 
-	responses.txt – všetky prompty a odpovede pre ľudí,
+Install (example):
 
-	summary.txt – súhrn odmietnutí podľa kategórií.
+```bash
-============================================================================================================================================================
+pip install -U torch transformers datasets peft trl accelerate bitsandbytes tqdm
+```
 
-2. copymaster.py – Triedenie výstupov
+Notes:
+- `translate/translate_PKF.py` is written for a **2‑GPU** setup.
+- Some scripts expect local model folders (e.g. NLLB, Llama Guard). Update paths in the script headers.
 
-	Keď už máš hromadu priečinkov v outputs/, potrebuješ to nejako zoradiť, aby každý model mal svoje miesto. O to sa stará copymaster.py.
+---
 
-2.1 Čo robí:
+## Suggested workflow (high level)
 
-	Opýta sa, ktorý model chceš spracovať.
+A typical run looks like this:
 
-	Prejde všetky priečinky v outputs.
+1) **Translate PKU** to Slovak (optional, if you need SK training data)  
+2) **Prepare** SFT + DPO datasets on disk  
+3) **Train** adapters (SFT/DPO)  
+4) **Generate** responses (base vs adapters)  
+5) **Translate** responses SK → EN (for Llama Guard)  
+6) **Evaluate** outputs with Llama Guard 3  
 
-	Všetko, čo obsahuje v názve “gemma”, “llama” alebo “qwen”, podľa toho čo si vybral, skopíruje do:
+You can also skip the Slovak translation and work directly with the original English PKU dataset via `Training/convert_dpo_sft.py`.
-	/response/<model>/
 
-	Každý JSON dostane svoje číslo: 1.json, 2.json, 3.json…
+---
 
-============================================================================================================================================================
+## Scripts (what they do + how to run)
 
-3. response_evaluate.py – Hodnotenie bezpečnosti
+Run commands below from the **repository root**.
 
-	Toto je hlavný a najväčší skript.
+---
-	Robí reálne hodnotenie, či sú prompty a odpovede modelov bezpečné alebo nie.
 
-	Používaš ho na porovnávanie modelov medzi sebou.
+### `preparation/prepar_dat_pku_dpo.py`
+**Purpose:** Takes a translated PKU dataset saved on disk and produces two outputs:
+- an **SFT** dataset (plain text prompts/completions),
+- a **DPO** dataset (prompt/chosen/rejected).
 
-3.1 Čo robí:
+**Run:**
+```bash
+python3 preparation/prepar_dat_pku_dpo.py
+```
 
-	Pýta si od teba, ktorú sadu z response/ chceš hodnotiť (llama, gemma, qwen).
+**What to edit first:**
+- `SRC_DIR` (input dataset saved via `datasets.save_to_disk`)
+- output paths (`SFT_OUT`, `DPO_OUT`)
 
-	Nechá ťa vybrať GPU alebo CPU.
+---
 
-	Načíta si Llama Guard 3–8B.
+### `Training/convert_dpo_sft.py`
+**Purpose:** Downloads **PKU-Alignment/PKU-SafeRLHF-30K** from Hugging Face and converts it into:
+- SFT: `./data/pku_sft.jsonl`
+- DPO: `./data/pku_dpo/` (HF `save_to_disk` format)
 
-	Každý prompt aj odpoveď vyhodnotí zvlášť:
+It shows a small interactive menu (SFT / DPO / BOTH).
 
-	prompt → je bezpečný / nebezpečný
+**Run:**
+```bash
+python3 Training/convert_dpo_sft.py
+```
 
-	odpoveď → model odpovedal bezpečne / nebezpečne
+---
 
-	Okrem Guardu používa aj tvoje vlastné heuristiky:
+### `translate/translate_PKF.py`
+**Purpose:** Translates PKU-SafeRLHF-30K to Slovak using a **local NLLB** model and a **2‑GPU** multiprocessing setup.
 
-	ak prompt obsahuje “sex”, “dirty joke” atď. → označí ho rovno ako unsafe,
+**Run:**
+```bash
+python3 translate/translate_PKF.py
+```
 
-	ak odpoveď obsahuje odmietnutie → automaticky safe.
+**Resume / merge only:**
+```bash
+python3 translate/translate_PKF.py --resume
+```
 
-	Každý hodnotený záznam uloží do samostatného JSON.
+**What to edit first:**
+- `NLLB_PATH` (local NLLB model directory)
+- output directory constants inside the script
 
-	Po spracovaní celého priečinka vytvorí:
+---
 
-	summary.json pre každý vstupný súbor,
+### `translate/translate_do-not_answer.py`
+**Purpose:** Translates **LibrAI/do-not-answer** (by default the `question` field) using NLLB and saves the translated dataset to disk.
 
-	summary_all.json pre celý model.
+**Run (defaults are usable as-is):**
+```bash
+python3 translate/translate_do-not_answer.py
+```
 
-3.2 Výsledkom je úplná štatistika:
+**Useful options:**
+```bash
+python3 translate/translate_do-not_answer.py --help
+python3 translate/translate_do-not_answer.py   --base_dir /home/hyrenko/Diploma/datasets   --out_name do_not_answer_sk   --model /home/hyrenko/Diploma/models/nllb-200-1.3B   --translate_fields question,risk_area
+```
 
-	koľko promptov bolo unsafe,
+---
 
-	koľko odpovedí bolo unsafe,
+### `Training/training_dpo_sft_llama.py`
+**Purpose:** Unified training script for **Llama (e.g. llama3.1‑8b)**:
+- SFT (QLoRA + masked loss)
+- DPO (TRL DPOTrainer)
 
-	koľko párov bolo naraz unsafe,
+It opens a menu (SFT / DPO / BOTH) and then relaunches itself via **accelerate** for multi-process training.
 
-	porovnanie modelov podľa bezpečnosti.
+**Run:**
-============================================================================================================================================================
+```bash
+python3 Training/training_dpo_sft_llama.py
+```
+
+If `accelerate` is not configured on your machine yet:
+```bash
+accelerate config
+```
+
+---
+
+### `Training/training_dpo_sft_mistral_sk.py`
+**Purpose:** Combined QLoRA training for **mistral-sk-7b**:
+- `sft`
+- `dpo`
+- `both`
+
+This script uses subcommands and supports CLI overrides (see `--help`).
+
+**Help:**
+```bash
+python3 Training/training_dpo_sft_mistral_sk.py --help
+```
+
+**Multi-GPU examples:**
+```bash
+torchrun --nproc_per_node=2 Training/training_dpo_sft_mistral_sk.py sft
+torchrun --nproc_per_node=2 Training/training_dpo_sft_mistral_sk.py dpo
+torchrun --nproc_per_node=2 Training/training_dpo_sft_mistral_sk.py both
+```
+
+---
+
+### `program/LLM_test.py`
+**Purpose:** Interactive generator for model responses. Lets you pick:
+- model source (base / adapters),
+- GPU,
+- dataset,
+- generation limits.
+
+Writes a `responses.json` under your configured outputs directory.
+
+**Run:**
+```bash
+python3 program/LLM_test.py
+```
+
+---
+
+### `program/Llama_test_trained.py`
+**Purpose:** Targeted evaluator for Llama runs. Useful for running **base vs SFT vs DPO** on a chosen dataset with explicit CLI flags.
+
+**Run:**
+```bash
+python3 program/Llama_test_trained.py --help
+```
+
+**Example:**
+```bash
+python3 program/Llama_test_trained.py --dataset do-not-answer --mode dpo --limit 200
+```
+
+---
+
+### `translate/Translate_sk_to_eng.py`
+**Purpose:** Translates Slovak `responses.json` → English (so Llama Guard can score English outputs).  
+It scans your runs folder and writes translated outputs into `outputs_translated` (path is in the script).
+
+**Run:**
+```bash
+python3 translate/Translate_sk_to_eng.py
+```
+
+---
+
+### `program/copymaster.py`
+**Purpose:** Copies `responses.json` files from `outputs/` into structured folders under `response/{gemma|llama|qwen}/` (used by the evaluator).
+
+**Run:**
+```bash
+python3 program/copymaster.py
+```
+
+**What to edit first:**
+- `OUTPUTS_DIR` (source runs folder)
+- `DEST_DIR` (destination base folder)
+
+---
+
+### `program/response_evaluate.py`
+**Purpose:** Runs **Llama Guard 3** on prompts + generated responses and saves:
+- per-item evaluation JSON,
+- summary stats (under `OUTPUT_ROOT`, configured in the script).
+
+Inputs are expected in folders like:
+- `/home/hyrenko/Diploma/response/{llama|gemma|qwen}`
+- `/home/hyrenko/Diploma/outputs_translated` (translated mistral-sk runs)
+
+**Run:**
+```bash
+python3 program/response_evaluate.py
+```
+
+**What to edit first:**
+- `MODEL_PATH` (local Llama Guard 3 model folder)
+- input/output directories (`*_INPUT_DIR`, `OUTPUT_ROOT`)
+
+---
+
+## Tips
+
+- If something “can’t find file/path”, check the **constants at the top** of the script first.
+- Keep outputs named consistently (`responses.json`) — several scripts rely on that.
+- For multi-GPU training, make sure your CUDA devices are visible and `torchrun/accelerate` sees them.