Distributed/ND-Parallel (#2977)

This commit is contained in:
salman
2025-07-31 20:25:02 +01:00
committed by GitHub
parent 7b68dfafd7
commit 294c7fe7a6
49 changed files with 712 additions and 835 deletions

View File

@@ -298,7 +298,7 @@ def oai_gsm8k_transform(cfg, *args, **kwargs):
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_linear": True,
"sequence_parallel_degree": 2,
"context_parallel_size": 2,
"flash_attention": True,
"sequence_len": 1024,
"special_tokens": {