tested after rebasing, fix importance sampling bug

YeAnbang · YeAnbang · commit 32b21486708f · 2025-08-06T06:15:15.000Z
diff --git a/applications/ColossalChat/examples/requirements.txt b/applications/ColossalChat/examples/requirements.txt
@@ -1,4 +1,4 @@
 pandas>=1.4.1
 sentencepiece
-colossalai==0.4.7
+colossalai>=0.4.7
 prompt_toolkit
diff --git a/applications/ColossalChat/requirements.txt b/applications/ColossalChat/requirements.txt
@@ -1,4 +1,4 @@
-transformers==4.39.3
+transformers>=4.39.3
 tqdm
 datasets==2.14.7
 loralib
diff --git a/applications/ColossalChat/rl_example.py b/applications/ColossalChat/rl_example.py
@@ -152,7 +152,7 @@
         "-ei",
         "--eval-interval",
         type=int,
-        default=100,
+        default=-1,
         help="Interval for evaluation. Evaluate every ei training steps.",
     )
     parser.add_argument(
diff --git a/colossalai/shardformer/modeling/qwen3.py b/colossalai/shardformer/modeling/qwen3.py
@@ -273,6 +273,7 @@ def qwen3_for_causal_lm_forward(
         hidden_states: Optional[torch.FloatTensor] = None,
         stage_index: Optional[List[int]] = None,
         shard_config: ShardConfig = None,
+        **kwargs
     ):
         r"""
         Args:

Original file line number	Diff line number	Diff line change
`@@ -152,7 +152,7 @@`
`152`	`152`	`"-ei",`
`153`	`153`	`"--eval-interval",`
`154`	`154`	`type=int,`
`155`		`- default=100,`
	`155`	`+ default=-1,`
`156`	`156`	`help="Interval for evaluation. Evaluate every ei training steps.",`
`157`	`157`	`)`
`158`	`158`	`parser.add_argument(`