Lightning-AI / litgpt

Pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.

https://lightning.ai

Lightning-AI/litgpt Issues

Will CycleIterator forward to dataset on resume for pretrain?
Updated 7 days ago1
Pretraining example from readme fails in Colab
Updated 7 days ago3
test_tinyllama issue with LitData and `iterate_over_all`
Updated 7 days ago1
Streamline LitGPT API
Updated 7 days ago5
Remove old and unused LLMs
Updated 8 days ago
Add release workflow
Closed 8 days ago2
LoRA matrices dropout
Updated 8 days ago
how to solve this debug
Updated 9 days ago
LoRA multi-GPU no longer works if applying LoRA selectively
Closed 9 days ago3
PermissionError: [WinError 5]
Updated 9 days ago8
Cannot copy out of meta tensor; no data!
Closed 9 days ago2
LR scheduler can result in a division by 0
Updated 9 days ago
Address frozen parameter warning with FSDP on nightly torch
Updated 9 days ago
Thunder test exceeds 5 min timeout
Closed 10 days ago1
LlamaMOE. The order of softmax and topK
Updated 10 days ago1
Is it correct to keep using adapter_kv_cache during training in litgpt/adapter.py?
Updated 10 days ago3
fabric.print only works on sys.stderr, does not print inference result
Updated 10 days ago1
Customizable loss function & inference step?
Closed 10 days ago2
ValueError: Cannot attend to 3063, block size is only 2048
Closed 10 days ago1
How to use custom dataset for evaluate?
Updated 12 days ago
Test readme commands as part of the CI
Closed 13 days ago5
tokenizer.py
Closed 13 days ago2
litgpt download doesn't work
Closed 16 days ago7
How to specify which GPU to use？
Closed 15 days ago2
After some iteration in pretraining a LLM, IndexError is raised related to dataset chunking
Updated 15 days ago
Why FSDPStrategy is so slow-down when I use multi-machine
Updated 16 days ago4
A potential bug for multi-GPU training
Updated 16 days ago5
Failed to load the finetuned model with `AutoModelForCausalLM.from_pretrained(name, state_dict=state_dict)`
Updated 16 days ago4
Add support for memory-efficient and faster optimizers
Updated 16 days ago1
combine FSDP with selective activation checkpointing
Updated 18 days ago
OOM Error: CUDA out of memory when finetuning llama3-8b
Closed 19 days ago3
The `litgpt evaluate` command attempts to download config files from gated repos
Closed 20 days ago1
Conversion to HF checkpoint should generate a checkpoint format that can be loaded directly
Updated 20 days ago1
ValueError: 'Meta-Llama-3-8B-Instruct' is not a supported config name
Closed 20 days ago4
On the relationship with `torchtune`
Updated 20 days ago1
--checkpoint-dir 'xx' is missing the files: ['model_config.yaml']
Closed 21 days ago2
merging intermediate adapter
Closed 21 days ago
Nucleus (top-p) sampling
Updated 22 days ago2
Continual pretraining for custom data is not working. Not recognizing TextFiles as a data attribute.
Closed 22 days ago3
Add support for phi-3-mini
Closed 22 days ago2
Finetune and Pretrain Path Discrepancy
Closed 22 days ago2
Blockwise quantization only supports 16/32-bit floats, but got torch.uint8 ( `bnb.nf4` quantisation is not working)
Updated 22 days ago15
Qwen1.5 Family Support
Updated 23 days ago
Docs for non-cli usage
Updated 25 days ago3
Pretraining example is not working
Updated 25 days ago1
simple API interface for initializing and running model inference
Updated a month ago
Question about inference time (resource-tables) : tps quantized llm < non quantized llm
Closed a month ago5
error: Unrecognized arguments: --val_split_fraction 0.1
Closed a month ago6
RuntimeError: Error(s) in loading state_dict for LlamaModel: size mismatch for model.layers.0.self_attn.q_proj.weight: copying a param with shape torch.Size([768, 2048]) from checkpoint, the shape in current model is torch.Size([2048, 2048]).
Updated a month ago
VRAM of LoRA fine-tuning
Closed a month ago