| # ============================================================ | |
| # General & System Configurations | |
| # ============================================================ | |
| project_name: "Emasters_Group-2_CapstoneProject" | |
| seed: 42 | |
| device: "auto" # Options: 'auto', 'cuda', 'mps', 'cpu' | |
| dtype: "float16" # Options: 'float16', 'bfloat16', 'float32' | |
| # ============================================================ | |
| # Model & Tokenizer Configs | |
| # ============================================================ | |
| model_name: "Qwen/Qwen2.5-Coder-0.5B-Instruct" | |
| lora_adapter_path: "models/checkpoints/lora_finetuned" | |
| # ============================================================ | |
| # Embedding Model Configs | |
| # ============================================================ | |
| embedding_model: "BAAI/bge-small-en-v1.5" # used only when embedding.source == "external" | |
| embedding: | |
| source: "external" # Options: 'code_lm' or 'external' | |
| external_model: "BAAI/bge-small-en-v1.5" | |
| max_length: 512 | |
| batch_size: 32 | |
| # ============================================================ | |
| # Retrieval Settings (semantic + AST/structural hybrid search) | |
| # ============================================================ | |
| retrieval: | |
| hybrid: true | |
| structural_weight_alpha: 0.7 | |
| structural_language: "mongodb" | |
| topk_sweep: [1, 2, 3, 5, 8] | |
| # ============================================================ | |
| # Large-LLM baseline (for "small code LM vs. LLM" comparison) | |
| # ============================================================ | |
| llm_baseline: | |
| backend: "api" # Options: 'local' or 'api' | |
| local_model: "Qwen/Qwen2.5-Coder-0.5B-Instruct" | |
| api_model: "gemini-3.6-flash" | |
| api_key_env: "GEMINI_API_KEY" | |
| max_new_tokens: 256 | |
| temperature: 0.2 | |
| # ============================================================ | |
| # Dataset Paths | |
| # ============================================================ | |
| data_paths: | |
| raw_dir: "data/raw" | |
| spider: "spider_real_execution_mongo.json" | |
| # ============================================================ | |
| # FAISS Vector & AST Index Paths | |
| # ============================================================ | |
| indices_paths: | |
| faiss_index: "data/indices/faiss_semantic.index" | |
| ast_store: "data/indices/ast_store.pkl" | |
| # ============================================================ | |
| # Text Generation Parameters | |
| # ============================================================ | |
| generation: | |
| max_length: 512 | |
| temperature: 0.2 | |
| top_k: 50 | |
| top_p: 0.95 | |
| num_return_sequences: 1 | |
| # ============================================================ | |
| # Evaluation Settings | |
| # ============================================================ | |
| evaluation: | |
| device: "cpu" # CodeBERTScore device | |
| bertscore_model: "microsoft/codebert-base" | |
| bertscore_local_path: "models/codebert-base" | |
| timeout_seconds: 3 | |
| # ============================================================ | |
| # LoRA Training Settings | |
| # ============================================================ | |
| lora: | |
| r: 16 | |
| alpha: 32 | |
| dropout: 0.05 | |
| batch_size: 1 | |
| learning_rate: 0.0002 | |
| epochs: 1 | |
| output_dir: "models/checkpoints/lora_finetuned" | |
| trainable_fraction: 1.0 # Fraction of LoRA parameters to train (0.0 to 1.0) | |
| # ============================================================ | |
| # Indexing Settings | |
| # ============================================================ | |
| indexing: | |
| top_k: 3 | |
| vector_dim: 1024 # Must match embedding model's output dimension | |
| # ============================================================ | |
| # Outputs & Evaluation Paths | |
| # ============================================================ | |
| outputs: | |
| logs_dir: "output/logs" | |
| metrics_path: "output/metrics/evaluation_metrics.json" | |
| plots_dir: "output/plots" | |
| generated_dir: "output/generated" | |