modelscope / modelscope/ms-swift

pt训练时,如果max length > 模型的max position embedding会怎么样

Open
#7,018 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

stale
Dominant language
Python
Stars
15.7k
Forks
1.7k
Avg merge
1d 16h
Merged PRs (30d)
136

Description

pt训练时,如果max length > 模型的max position embedding会怎么样,我目前用如下参数训练时,发现不会报错,但没有配置任何scaling参数,请问目前的框架会进行怎样的处理:
Qwen3-8B-Base max position embedding: 32768

swift pt \ --model $model_path \ --model_type qwen3 \ --train_type full \ --dataset ${DATA_PATHS[@]} \ --val_dataset ${VAL_DATA_PATHS[@]} \ --dataset_shuffle true \ --shuffle_buffer_size 10000 \ --streaming true \ --torch_dtype bfloat16 \ --per_device_train_batch_size 1 \ --per_device_eval_batch_size 1 \ --learning_rate 1e-5 \ --gradient_accumulation_steps 2 \ --sequence_parallel_size 8 \ --packing true \ --eval_steps 5 \ --save_steps 500 \ --save_total_limit 2 \ --logging_steps 1 \ --deepspeed zero3 \ --max_length 64000 \ --max_steps 10000 \ --warmup_ratio 0.05 \ --dataloader_num_workers 4 \ --dataset_num_proc 8 \ --save_only_model true \

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by tracing how the training command handles --max_length 64000 when the model reports max position embedding 32768; check the resulting sequence-length and position-embedding behavior. Done means documenting whether the framework errors, truncates, or applies scaling, with a reproducible conclusion for the supplied command.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.