Qetuoadgjl Weights1: LLaVA Model Checkpoints from Pre-training and SFT

Five LLaVA model checkpoints uploaded by author xym93168 on Hugging Face in April 2026. The checkpoints document different training stages, including pre-training and supervised fine-tuning, with varying GPU configurations and batch sizes. Specific checkpoints include 'Pre_32gpu_llava_bs8_0111_1epoch/checkpoint-2181' and 'sft_8gpu_llava_bs08_0106_1epoch/checkpoint-2353'.

Use Cases

• Resume training from a specific checkpoint based on the provided pre-training or SFT stage.
• Compare model performance across different batch sizes (8 vs 16) mentioned in the checkpoint names.
• Benchmark training efficiency using the GPU configuration details (32 GPU vs 8 GPU) described.
• Analyze the impact of supervised fine-tuning (SFT) versus pre-training using the provided stage labels.

Strengths

• Checkpoints are explicitly labeled with their training stage (pre-train, sft, pre-train + sft).
• Training hyperparameters like batch size (8, 16) and GPU count (8, 32) are documented in the names.
• Multiple checkpoints from different training runs allow for comparative analysis.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's purpose and the exact model architecture (beyond 'LLaVA') are not detailed.

Provenance

Source

huggingface

Collection Method

Model checkpoints generated during training runs, uploaded by the author.

Time Range

Checkpoints generated in January 2026 (based on folder names '0105', '0106', '0111').

Freshness

Last updated 2026-04 08 04:57:47; freshness should be verified.

Geography

null