Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[BUG] Fix ds_chat regression #7014

Open
tjruwase opened this issue Feb 7, 2025 · 0 comments · May be fixed by #7015
Open

[BUG] Fix ds_chat regression #7014

tjruwase opened this issue Feb 7, 2025 · 0 comments · May be fixed by #7015
Assignees
@tjruwase
Labels
bug Something isn't working training

Comments

@tjruwase
Copy link
Contributor

tjruwase commented Feb 7, 2025

The module injection refactoring in PR #7004 broke existing code leading to ds-chat CI failures as below:

[rank2]:   File "/scratch/azureml/cr/j/1df407cd106142c2afb0cda94a311736/exe/wd/actions-runner/_work/DeepSpeed/DeepSpeed/unit-test-venv/lib/python3.10/site-packages/deepspeed/runtime/hybrid_engine.py", line 293, in create_inference_containers
[rank2]:     self._other_layers.append(self.inference_policies[child.__class__][0](
[rank2]: TypeError: LinearLayer.__init__() missing 2 required positional arguments: 'module' and 'mp_group'

To Reproduce
Run ds-chat workflow. Or after environment setup, you the following command will reproduce the issue.

deepspeed --num_gpus=1 main.py --data_path Dahoas/rm-static --data_split 2,4,4 --actor_model_name_or_path facebook/opt-125m --critic_model_name_or_path /tmp/ds_chat --num_padding_at_beginning 1 --per_device_generation_batch_size 4 --per_device_training_batch_size 4 --generation_batches 1 --ppo_epochs 1 --max_answer_seq_len 256 --max_prompt_seq_len 256 --actor_learning_rate 9.65e-6 --critic_learning_rate 5e-6 --num_train_epochs 1 --lr_scheduler_type cosine --gradient_accumulation_steps 1 --actor_weight_decay 0 --critic_weight_decay 0 --num_warmup_steps 100 --deepspeed --seed 1234 --actor_dropout 0.0 --print_answers --actor_zero_stage 3 --critic_zero_stage 3 --output_dir z3_he_true_offload_true_lora_true --enable_hybrid_engine --offload --actor_lora_dim 128 --actor_lora_module_name decoder.layers. --enable_test_mode --test_stop_step 5

Expected behavior
ds-chat CI should work.
@tjruwase tjruwase added bug Something isn't working training labels Feb 7, 2025
@tjruwase tjruwase self-assigned this Feb 7, 2025
@tjruwase tjruwase linked a pull request Feb 7, 2025 that will close this issue
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@tjruwase tjruwase

Labels
bug Something isn't working training
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Fix ds-chat CI regression deepspeedai/DeepSpeed
1 participant
@tjruwase