Repository Issue Activity (beta)

#1242 False positive warning about mixed precision in `merge_lora.py`

#2223 No checkpoints saved during pretraining in Google Colab despite log.log_model and save_interval set.

#2306 Reang

#2319 litgpt pretrain crashes or deadlocks when a rank's val_dataloader yields zero batches (multi-GPU FSDP)

#1102 Make `save_hyperparameters()` robust against different CLI entry points

#2307 Speculative decoding drops the first generated token

#2302 Stop sequence matching mishandles overlapping prefixes and buffered tokens

#2189 Unsafe Checkpoint Loading - This Is Bad

#2158 Support for SSM models (Mamba, Mamba2)

#2263 MCP to help OSS devs understand Lightning-AI/litgpt

#2116 Initial and final evaluation in `finetune` scripts do not accumulate over devices

#1871 failure converting pretrained litgpt checkpoints to HF format: a reproducible example

#2191 Gradient Clipping Doesn't Work in Finetuning

#1221 Calculate loss at beginning and end of training

#1082 Add TinyStories to the pretraining docs

#756 Is there a library to serve the model trained via lit-gpt?

#722 Git-LFS doesn't work for downloading `RedPajama-Data-1T` dataset

#2238 `sample()` with `temperature=0` does not produce greedy decoding when `top_p > 0`

#1809 Support for mini-omni and mini-omni2 pre training, fine tuning on custom dataset.

#1957 Multiple redundant calls to generate_example() when using multiple GPUs

#2192 Error Handling is... Trusting

#491 RuntimeError: Cannot writeback when the parameter shape changes

#2222 Description

#1402 Pretraining example from readme fails in Colab

#2182 Sliding-window KV cache + prefill longer than window breaks causality (ring-buffer overwrite + mask not position-aware)