webui checkpoint path dropdown only shows part of the checkpoints

[Patrick-Rud]

Reminder

• I have read the above rules and searched the existing issues.

System Info

• llamafactory version: 0.9.5.dev0
• Platform: Linux-6.6.87.2-microsoft-standard-WSL2-x86_64-with-glibc2.35
• Python version: 3.11.0
• PyTorch version: 2.9.1+cu128 (GPU)
• Transformers version: 4.52.4
• Datasets version: 4.0.0
• Accelerate version: 1.11.0
• PEFT version: 0.17.1
• GPU type: NVIDIA GeForce RTX 5060 Ti
• GPU number: 1
• GPU memory: 15.93GB
• TRL version: 0.24.0
• Bitsandbytes version: 0.49.0
• Default data directory: detected

Reproduction

bf16: true
cutoff_len: 2048
dataset: sample0.5
dataset_dir: data
ddp_timeout: 180000000
do_train: true
double_quantization: true
enable_thinking: false
finetuning_type: lora
flash_attn: auto
gradient_accumulation_steps: 16
include_num_input_tokens_seen: true
learning_rate: 5.0e-05
logging_steps: 5
lora_alpha: 16
lora_dropout: 0
lora_rank: 8
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 100000
model_name_or_path: /home/patrick/.cache/modelscope/hub/models/LLM-Research/Meta-Llama-3-8B-Instruct/
neat_packing: true
num_train_epochs: 2.0
optim: adamw_torch
output_dir: saves/Llama-3-8B-Instruct/lora/train_2026-01-13-01-29-26
packing: true
per_device_train_batch_size: 2
plot_loss: true
preprocessing_num_workers: 16
quantization_bit: 4
quantization_method: bnb
report_to: none
save_steps: 100
stage: sft
swanlab_mode: cloud
swanlab_project: llamafactory
swanlab_run_name: train0.5
template: llama3
trust_remote_code: true
use_swanlab: true
warmup_steps: 0

Others

I ctrl+c the webui service instead of aborting on the webui because the training stopped(full gpu memory usage but 0 gpu util) on this run. When I want to resume from the checkpoint, it's not on the dropdown list.

Why this happen? What's the alternative to resume training?