SecAlign-70B support: configs, quantization, multi-GPU (#1)

- **PEFT adapter merging.** `model_loader.py` auto-detects PEFT adapters (e.g. `facebook/Meta-SecAlign-8B`), merges on CPU in bf16, and caches the merged model to disk. No config flags needed.

- **Configurable quantization.** `quantization:` field in YAML or `--quantization` on CLI, accepting `nf4`, `fp4`, or `int8`. Replaces the old `load_in_4bit` boolean.

- **Multi-GPU sharding.** `device_map:` in configs or `--device-map` on CLI. Config value is now correctly read from YAML presets (was previously ignored).

- **CLI overrides.** New `--model`, `--device-map`, `--quantization` flags to override preset values from the command line.

- **SecAlign injection presets.** Configs for prompt injection on Meta-SecAlign-8B and 70B (default + Optuna-tuned), using new `AlpacaInjectionSource` — generates 3-role prompts from AlpacaFarm data with trusted/untrusted separation.

- **Fixes.** `BenchmarkRunner.summarize()` crash when all runs are skipped. System prompt suppression now works correctly (`""` suppresses model defaults, `None` omits the turn).

Co-authored-by: Peter Romov <peter@romov.com>
Co-authored-by: Alexander Panfilov <39771221+kotekjedi@users.noreply.github.com>

pyproject.toml

@@ -3,7 +3,7 @@ name = "claudini"
 version = "0.1.0"
 description = "Claudini: Automated Research Discovers State-of-the-Art Adversarial Attack Algorithms"
 readme = "README.md"
-requires-python = ">=3.12"
+requires-python = ">=3.14"
 dependencies = [
     "accelerate>=1.13.0",
     "numpy>=2.0",
@@ -14,6 +14,9 @@ dependencies = [
     "transformers>=4.40",
     "typer>=0.9",
     "datasets>=2.14",
+    "huggingface_hub>=0.20",
+    "bitsandbytes>=0.46.1",
+    "peft>=0.18.1",
 ]
 
 [build-system]