Repository Issue Activity (beta)

#220 use `gpudebug` for tracing/profiling

#161 Week 1 Day 5 Qwen3-4B reference test is flaky with MLX 0.32

#108 Flash attention is not faster than vanilla attention

#118 📋 Documentation Enhancement Suggestion

#116 A simple CUDA adaptation of the tutorial

#82 Is there any timeline for week3?

#30 Is there any timeline for week2+ tasks?

#84 Support for MLA (Multi-Head Latent Attention) Task

#50 Might be better to add more information about how MLX extension works.

#63 test_task_3_qwen2_grouped_query_attention uses rope_traditional=True as reference

#27 Can not pass all the test when check installation

#45 precision issues in M1

#41 Weights shape issue in `src/tiny_llm_ref/attention.py:SimpleMultiHeadAttention`

#40 refactor: support multiple models

#34 Migrate to Qwen 3

#37 feat: Distributed inference (TP/PP)

#28 Ref `Qwen2MultiHeadAttention` Impl seems not have enough precision

#32 The reference top k sampler is incorrect

#23 week-1-day-3 task_3 test case not implemented

#22 check.py does not exist

#17 The reference solution cannot pass all the *week 1* tests when checking the Installation

#15 The step to run build_ext.sh is missing from the docs

#1 Wrong test script in Week1

#4 Question about pdm