Repository Issue Activity (beta)

#80 Full-Scale Experiment with exact_accuracy 0

#69 Support for AMD ROCm devices additionally to NVIDIA CUDA one.

#63 Question about carry state management across different samples during training

#10 arcprize.org leaderboard

#106 Question about baseline CoT prompts and settings (o3-mini-high, DeepSeek R1, Claude 3.7 8K)

#82 Reproducibility results of Sudoku-Extreme-1k

#46 The input to CastedSparseEmbedding is the index of the input in the overall data. What is the role of this embedding?

#104 How to use checkpoints tp predict?

#101 Replace bare except with explicit exception in model loading (evaluate.py)

#100 Is HRM suitable for text reasoning tasks on DNA sequences?

#98 not an issue but a fork

#41 import adam_atan2_backend ModuleNotFoundError: No module named 'adam_atan2_backend'

#97 Truc

#25 pip install adam_atan2 success then run code get No module named 'adam_atan2_backend'

#67 I would like to ask the model parameter 27, why does checkpoint have 2.2G?

#53 Universal Device Support Implementation

#90 Has anyone managed to run evaluate.py with the provided ARC-AGI-2 checkpoint?

#95 Optionally, arbitrary mean and fan-in scaling supported.

#91 Enhance Weight Initialization: Add Correct Truncated Normal with Variance Compensation

#94 Documentation clearly explains rationale, usage, and mathematical background.

#93 Variance after initialization matches requested std.

#92 New initializer is available and tested.

#12 Reproducibility results of Sudoku-Extreme and ARC-AGI 1

#88 Contribution Guide, Interested in helping

#86 Tokenised App