Issues

• Allocate KV cache buffers in pages

  enhancementgood first issue
  Feature

  awslabs/keys_values#137 · Filter by author mseeger opened on Jul 8, 2026
  1

• Replace Lightning Fabric by elementary setup

  enhancement
  Feature

  awslabs/keys_values#135 · Filter by author mseeger opened on Jul 1, 2026
  7

  

• Consider query-only test time training

  enhancement
  Feature

  awslabs/keys_values#132 · Filter by author mseeger opened on Jun 18, 2026
  1

• Consider modern RoPE variants

  enhancementgood first issue
  Feature

  awslabs/keys_values#131 · Filter by author mseeger opened on Jun 18, 2026

• Dashboards for training and evaluation loss

  enhancementgood first issue
  Feature

  awslabs/keys_values#104 · Filter by author mseeger opened on May 6, 2026
  1

• Implement LoRA improvements

  enhancementgood first issue
  Feature

  awslabs/keys_values#86 · Filter by author mseeger opened on Apr 10, 2026

• Create auto-tuning script to find best parameter combinations

  enhancementgood first issue
  Feature

  awslabs/keys_values#79 · Filter by author mseeger opened on Apr 1, 2026
  6

• Roll back small number of KV cache updates

  enhancement
  Feature

  awslabs/keys_values#76 · Filter by author mseeger opened on Mar 27, 2026
  4

• Backward CPU offloading: Asynchronous transfer

  enhancement
  Feature

  awslabs/keys_values#62 · Filter by author mseeger opened on Mar 14, 2026
  16

• Configuration of fine-tuning scripts by YAML file

  enhancementgood first issue
  Feature

  awslabs/keys_values#50 · Filter by author mseeger opened on Mar 6, 2026
  1

• KV cache seems not reset before applied to new sequence

  bug
  Bug

  awslabs/keys_values#45 · Filter by author mseeger opened on Mar 2, 2026
  1

• Evaluation script fails when started independently >2 times

  bug
  Bug

  awslabs/keys_values#42 · Filter by author mseeger opened on Feb 27, 2026
  1

• Can KV reordering be avoided with flex_attention?

  enhancement
  Feature

  awslabs/keys_values#39 · Filter by author mseeger opened on Feb 24, 2026
  5

• Clean up flex_attention integration, extend_kv=False case

  enhancement
  Task

  awslabs/keys_values#34 · Filter by author mseeger opened on Feb 19, 2026
  3

• Refactor node annotations in autograd_hooks.py

  enhancement
  Task

  awslabs/keys_values#31 · Filter by author mseeger opened on Feb 16, 2026
  3

• Try to use PyTorch DataLoader again

  enhancementgood first issue
  Feature

  awslabs/keys_values#27 · Filter by author mseeger opened on Feb 10, 2026

• Implement speculative decoding

  enhancement
  Feature

  awslabs/keys_values#22 · Filter by author mseeger opened on Feb 1, 2026
  1

• Complete code for GPU memory estimation. Use for decision making

  enhancement
  Feature

  awslabs/keys_values#20 · Filter by author mseeger opened on Feb 1, 2026

• Implement context parallelism (RingAttention) as KVCache

  enhancement
  Feature

  awslabs/keys_values#19 · Filter by author mseeger opened on Jan 31, 2026
  5

  

• Implement FastGen as meta policy

  enhancement
  Feature

  awslabs/keys_values#17 · Filter by author mseeger opened on Jan 31, 2026

• Implement test to quantify the impact of different features on gradient approximation

  enhancementgood first issue
  Feature

  awslabs/keys_values#15 · Filter by author mseeger opened on Jan 31, 2026

• Do finetune scripts work with AMP?

  question
  Task

  awslabs/keys_values#14 · Filter by author mseeger opened on Jan 31, 2026

• Support dynamic RoPE scaling during inference call

  enhancement
  Feature

  awslabs/keys_values#12 · Filter by author mseeger opened on Jan 31, 2026
  1

• Support different cache types for each layer. Gradient computations when all or some caches are of type DenseKVCache

  enhancement
  Feature

  awslabs/keys_values#10 · Filter by author mseeger opened on Jan 31, 2026
  1

• Improve recovery from OOM errors mechanism

  enhancement
  Feature

  awslabs/keys_values#8 · Filter by author mseeger opened on Jan 31, 2026