已关闭
[ignore] [ci-test] Bad patch 2 bad context #79
hanboyou创建于 2025年12月22日关闭于 2025年12月24日
[ignore] [ci-test] Bad patch 2 bad context #79
已关闭
hanboyou创建于 2025年12月22日关闭于 2025年12月24日
共 2 个文件变更+12-6
@@ -1,5 +1,11 @@
1# Qwen3系列模型 RL训练优化实践样例1# Qwen3系列模型 RL训练优化实践样例
2- 2+[2025/12/21 00:33:59.173 GMT+08:00] - [Missing reference] Patch '0001-feature-model_converter.patch' is not mentioned in README.md.
3+[2025/12/21 00:33:59.173 GMT+08:00] - [Missing reference] Patch '0006-vllm_ascend-feature-chunk_moe_eager.patch' is not mentioned in README.md.
4+[2025/12/21 00:33:59.173 GMT+08:00] - [Missing reference] Patch '0011-verl-feature-enable_rollout_rebalance.patch' is not mentioned in README.md.
5+[2025/12/21 00:33:59.173 GMT+08:00] - [Missing reference] Patch '0001-mindspeed-bugfix-bad_context.patch' is not mentioned in README.md.
6+[2025/12/21 00:33:59.173 GMT+08:00] - [Missing reference] Patch '0001-megatron-bugfix-ssssss.patch' is not mentioned in README.md.
7+llm_rl/qwen3/patches/megatron/0001-megatron-bugfix-bad_path.patch
8+0001-megatron-bugfix-state_ten_verification.patch
3## 概述9## 概述
4本样例针对Qwen3-235B-A22B和Qwen3-32B模型,基于[veRL开源框架](https://github.com/volcengine/verl),使用veRL原生支持的MindSpeed和vLLM-Ascend框架,完成RL训练全流程的优化适配。10本样例针对Qwen3-235B-A22B和Qwen3-32B模型,基于[veRL开源框架](https://github.com/volcengine/verl),使用veRL原生支持的MindSpeed和vLLM-Ascend框架,完成RL训练全流程的优化适配。
5 11 
Rllm_rl/qwen3/patches/mindspeed/0001-mindspeed-bugfix-builder.patch→llm_rl/qwen3/patches/mindspeed/0001-mindspeed-bugfix-bad_context.patch+5-5
@@ -14,13 +14,13 @@ index 8cacac2..1b10806 100644
14+++ b/llm_rl/qwen3/mindspeed/op_builder/builder.py14+++ b/llm_rl/qwen3/mindspeed/op_builder/builder.py
15@@ -52,7 +52,7 @@ class MindSpeedOpBuilder(ABC):15@@ -52,7 +52,7 @@ class MindSpeedOpBuilder(ABC):
16 16
17- def cxx_args(self):17+ ========== mismatching context ==========
18- args = ['-fstack-protector-all', '-Wl,-z,relro,-z,now,-z,noexecstack', '-fPIC', '-pie',18+ ========== mismatching context ==========
19- '-s', '-fvisibility=hidden', '-D_FORTIFY_SOURCE=2', '-O2']19- '-s', '-fvisibility=hidden', '-D_FORTIFY_SOURCE=2', '-O2']
20+ '-s', '-fvisibility=hidden', '-D_FORTIFY_SOURCE=2', '-O2', '-include cstring']20+ '-s', '-fvisibility=hidden', '-D_FORTIFY_SOURCE=2', '-O2', '-include cstring']
21- return args21+ ========== mismatching context ==========
22- 22+ ========== mismatching context ==========
23- def extra_ldflags(self):23+ 
24-- 24--
252.50.1.windows.1252.50.1.windows.1
26 26