2/25/25

Fine-tuning Results
Created on February 26|Last edited on February 26
Comment
﻿
Note: the runs below are as follows:
Blue = all substructures, unrefined hyperparameters
Gray = all substructures, refined
Purple = per substructure, unrefined
Red = per substructure, refined
Validation set﻿
eval/steps_per_second
eval/steps_per_second
2004006008001ktrain/global_step123456
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
eval/samples_per_second
eval/samples_per_second
2004006008001ktrain/global_step6789101112
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
eval/runtime
eval/runtime
2004006008001ktrain/global_step2468
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
qwen2.5-vl-3b-instruct-full-sft
Run set6
﻿
Train set﻿
Run set6
﻿
System﻿
Add a comment