From de40c736d01eddd7b5cbb904fe417e1a038dbab5 Mon Sep 17 00:00:00 2001
From: YeAnbang <anbangy2@outlook.com>
Date: Mon, 9 Jun 2025 09:37:28 +0800
Subject: [PATCH] fix bug, tested

---
 .gitignore                                              | 4 ++++
 applications/ColossalChat/coati/distributed/consumer.py | 4 +---
 applications/ColossalChat/coati/distributed/producer.py | 2 +-
 3 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/.gitignore b/.gitignore
index 0503f3c95..e603f5015 100644
--- a/.gitignore
+++ b/.gitignore
@@ -167,3 +167,7 @@ applications/ColossalChat/wandb
 applications/ColossalChat/model
 applications/ColossalChat/eval
 applications/ColossalChat/rollouts
+applications/ColossalChat/*.txt
+applications/ColossalChat/*.db
+applications/ColossalChat/stdin
+applications/ColossalChat/*.zip
diff --git a/applications/ColossalChat/coati/distributed/consumer.py b/applications/ColossalChat/coati/distributed/consumer.py
index c7f424e0a..40d278e84 100644
--- a/applications/ColossalChat/coati/distributed/consumer.py
+++ b/applications/ColossalChat/coati/distributed/consumer.py
@@ -132,9 +132,7 @@ class BaseConsumer:
                             format_acc = raw_batch["format_acc"][:, :, 0]
                             ans_acc = raw_batch["ans_acc"][:, :, 0]
                             response_len = (
-                                raw_batch["response_idx"][:, :, 1]
-                                - raw_batch["response_idx"][:, :, 0]
-                                + 1
+                                raw_batch["response_idx"][:, :, 1] - raw_batch["response_idx"][:, :, 0] + 1
                             ).type(torch.float32)
                             effective_group_mask = None
                             if self.filter_range is not None and self.grpo_config.get("dynamic_batching", True):
diff --git a/applications/ColossalChat/coati/distributed/producer.py b/applications/ColossalChat/coati/distributed/producer.py
index 135c8db0e..23542f1c6 100644
--- a/applications/ColossalChat/coati/distributed/producer.py
+++ b/applications/ColossalChat/coati/distributed/producer.py
@@ -291,7 +291,7 @@ class BaseProducer:
                     reward_model_output = self.reward_model(
                         outputs["input_ids"].view((-1, outputs["input_ids"].size(-1))),
                         gt_answer=gt_answer,
-                        response_idx=outputs["response_idx"],
+                        response_idx=outputs["response_idx"].view((-1, 2)),
                     )
                 outputs["reward"] = (
                     torch.tensor([value[0] for value in reward_model_output])