sentence-transformers How to use hasinthakapiyumal/bge-code-v1-ai-pattern-tuned with sentence-transformers:
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("hasinthakapiyumal/bge-code-v1-ai-pattern-tuned")
sentences = [
"This code implements a Reinforcement Learning (RL) agent, specifically a Deep Q-Network (DQN) or its Double DQN variant, designed for interaction with OpenAI Gym environments. It employs an experience replay buffer for efficient off-policy learning and manages exploration-exploitation through initial random actions and decaying noise. A key AI pattern is the use of a distributional RL approach, likely C51, evident from the visualization of predicted value distributions across a range of atoms during policy rendering.",
"The code exhibits an optimization-driven 3D bounding-box estimation pattern: BoxRegressor minimizes a reprojection residual of 8 keypoints plus size and distance regularizers, using a pseudo-inverse camera-initialization and a discrete RotY sweep evaluated with least_squares. It embodies a geometric projection pattern via get_keypoints that builds the 3D corners under rotation, projects them through the camera matrix to image coordinates, and uses those projections as the residual basis. Data handling follows a PyTorch-style pattern with multiple Dataset classes that load KITTI data, apply per-sample augmentations (random flips, crops, color normalization), and normalize/prepare size, keypoints, and distance tensors for learning.",
"This code showcases a PyTorch-based, object-oriented training framework with a BaseModel that unifies compilation, iterative training (train_step, train_epoch), evaluation, checkpointing, and early stopping driven by a Monitor. It implements a modular regularization strategy via get_regularizer, supporting float, L1/L2, and L1_L2 patterns and applying penalties selectively to embedding vs. non-embedding parameters. The MultiTaskModel demonstrates a true multi-task pattern: per-task outputs and losses organized in a ModuleList, task-wise activation mappings, and per-task metric evaluation with aggregated reporting and group-aware metrics guided by a feature_map.",
"This code implements a system for Large Language Model (LLM) inference, loading a pre-trained causal language model and tokenizer to process user queries. A central AI pattern is \"AutoThink,\" which enables controllable generation through activation steering. This involves applying steering vectors from a specified dataset to a target model layer, guided by explicit \"pattern strengths\" to direct the LLM's output towards desired cognitive attributes like depth, accuracy, and self-correction."
]
embeddings = model.encode(sentences)
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [4, 4] 
