Add rlhf support in launcher (#98)

* Add rlhf configs * Add some place holders * Add rlhf support * Fix rlhf path * Add hetjob var setting * Add het host in rlhf command --------- Co-authored-by: Bobby Chen <[email protected]>
NVIDIA · Aug 31, 2023 · cd6368e · cd6368e
1 parent 9e54708
commit cd6368e
Show file tree

Hide file tree

Showing 7 changed files with 935 additions and 13 deletions.
diff --git a/launcher_scripts/conf/config.yaml b/launcher_scripts/conf/config.yaml
@@ -10,6 +10,8 @@ defaults:
   - ia3_learning: null
   - evaluation: gpt3/evaluate_all
   - export: gpt3/export_gpt3
+  - rlhf_rm: gpt3/2b_rm
+  - rlhf_ppo: gpt3/2b_ppo
   - override hydra/job_logging: stdout
 
 hydra:
@@ -65,3 +67,5 @@ ia3_learning_config: ${hydra:runtime.choices.ia3_learning}
 evaluation_config: ${hydra:runtime.choices.evaluation}
 conversion_config: ${hydra:runtime.choices.conversion}
 export_config: ${hydra:runtime.choices.export}
+rlhf_rm_config: ${hydra:runtime.choices.rlhf_rm}
+rlhf_ppo_config: ${hydra:runtime.choices.rlhf_ppo}