Issues

• Add DeepSeek-R1-Distill-Qwen (1.5B / 7B / 14B / 32B) model configs

  Lightning-AI/litgpt#2348 · Filter by author jkatdare opened on Oct 1, 2026

• Finetuning with gradient accumulation weights micro-batches, not tokens, so the objective depends on micro_batch_size

  Lightning-AI/litgpt#2346 · Filter by author jkatdare opened on Oct 1, 2026
  2

• Sliding window layers rebuild a dense (T, T) attention mask with a Python loop on every training forward pass

  Lightning-AI/litgpt#2344 · Filter by author jkatdare opened on Oct 1, 2026

• Proposal: deploy LitGPT on Nebius Serverless Endpoints

  Lightning-AI/litgpt#2341 · Filter by author SalikovAlex opened on Sep 28, 2026

• fix_and_load_json drops a comma that is inside a string

  Lightning-AI/litgpt#2339 · Filter by author SashaMIT opened on Sep 28, 2026

• EvalPort: portable interchange for lm_eval results.json

  Lightning-AI/litgpt#2337 · Filter by author adhabnr-ux opened on Sep 27, 2026

• Configurable per-layer LoRA ranks

  enhancement
  Lightning-AI/litgpt#2335 · Filter by author kor80 opened on Sep 25, 2026

• LLM.generate() can silently select a token excluded by top_k on CPU when sampling probabilities are float16

  Lightning-AI/litgpt#2323 · Filter by author rookieLiu2018 opened on Sep 8, 2026
  1

• litgpt pretrain crashes or deadlocks when a rank's val_dataloader yields zero batches (multi-GPU FSDP)

  Lightning-AI/litgpt#2319 · Filter by author pconsul00 opened on Sep 2, 2026
  1

• Speculative decoding drops the first generated token

  Lightning-AI/litgpt#2307 · Filter by author Excelius-Wang opened on Aug 25, 2026

• Reang

  question
  Lightning-AI/litgpt#2306 · Filter by author izereang opened on Aug 25, 2026

• Stop sequence matching mishandles overlapping prefixes and buffered tokens

  Lightning-AI/litgpt#2302 · Filter by author tandede opened on Aug 19, 2026

• No checkpoints saved during pretraining in Google Colab despite log.log_model and save_interval set.

  bug
  Lightning-AI/litgpt#2223 · Filter by author harshitgavita-07 opened on Mar 26, 2026
  3

• mask_cache in kv-cache path seems to force attn_mask, preventing flash attention

  question
  Lightning-AI/litgpt#2220 · Filter by author sea-with-sakura opened on Mar 23, 2026
  1

• Gradient Clipping Doesn't Work in Finetuning

  question
  Lightning-AI/litgpt#2191 · Filter by author AI-God-Dev opened on Jan 15, 2026
  3

• Memory Management Issues

  Lightning-AI/litgpt#2190 · Filter by author AI-God-Dev opened on Jan 15, 2026

• Unsafe Checkpoint Loading - This Is Bad

  Lightning-AI/litgpt#2189 · Filter by author AI-God-Dev opened on Jan 15, 2026
  1

• Feature: Support for Multimodality

  enhancement
  Lightning-AI/litgpt#2173 · Filter by author sanggusti opened on Dec 23, 2025
  1

• Request to support MedGemma and MedSigLip

  enhancement
  Lightning-AI/litgpt#2172 · Filter by author sanggusti opened on Dec 23, 2025
  1

• difference during inference with or without kvcache

  bug
  Lightning-AI/litgpt#2165 · Filter by author guangyuli-uoe opened on Nov 27, 2025
  4

• Cut cross entropy for pretraining and finetuning

  enhancement
  Lightning-AI/litgpt#2163 · Filter by author Aceticia opened on Nov 19, 2025

• Support for SSM models (Mamba, Mamba2)

  enhancement
  Lightning-AI/litgpt#2158 · Filter by author radulescupetru opened on Nov 14, 2025
  2

• jsonargparse==4.42.0 causing test_config_help failure

  Lightning-AI/litgpt#2150 · Filter by author bhimrazy opened on Oct 30, 2025

• Request to support OLMoE

  enhancement
  Lightning-AI/litgpt#2133 · Filter by author yangdongchao opened on Sep 12, 2025
  2

• Initial and final evaluation in finetune scripts do not accumulate over devices

  bug
  Lightning-AI/litgpt#2116 · Filter by author mseeger opened on Aug 30, 2025
  4