已关闭
[ignore] [ci-test] Bad patch 1 bad path #67
hanboyou创建于 2025年12月13日关闭于 2025年12月24日
[ignore] [ci-test] Bad patch 1 bad path #67
已关闭
共 3 个文件变更+10-5
| @@ -1,5 +1,10 @@ | |||
| 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_patch.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 | ||
| 3 | ## 概述 | 8 | ## 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。 | 9 | 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 | 10 | ||
| @@ -10,7 +10,7 @@ | |||
| 10 | # distributed under the License is distributed on an "AS IS" BASIS, | 10 | # distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | # See the License for the specific language governing permissions and | 12 | # See the License for the specific language governing permissions and |
| 13 | -# limitations under the License. | 13 | +# limitations under the License.j |
| 14 | 14 | ||
| 15 | #!/bin/bash | 15 | #!/bin/bash |
| 16 | set -o pipefail | 16 | set -o pipefail |
Rllm_rl/qwen3/patches/megatron/0001-megatron-bugfix-state_ten_verification.patch→llm_rl/qwen3/patches/megatron/0001-megatron-bugfix-bad_path.patch+3-3
| @@ -9,10 +9,10 @@ thereby improving code robustness. | |||
| 9 | llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py | 2 ++ | 9 | llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py | 2 ++ |
| 10 | 1 file changed, 2 insertions(+) | 10 | 1 file changed, 2 insertions(+) |
| 11 | 11 | ||
| 12 | -diff --git a/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py b/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py | 12 | +diff --git a/x/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py b/x/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py |
| 13 | index a00d1a3..62fe20f 100644 | 13 | index a00d1a3..62fe20f 100644 |
| 14 | ---- a/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py | 14 | +--- a/x/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py |
| 15 | -+++ b/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py | 15 | ++++ b/x/llm_rl/qwen3/megatron/core/optimizer/distrib_optimizer.py |
| 16 | class DistributedOptimizer(MixedPrecisionOptimizer): | 16 | class DistributedOptimizer(MixedPrecisionOptimizer): |
| 17 | # The optimizer state of STEP is handled | 17 | # The optimizer state of STEP is handled |
| 18 | # specifically and is read from param_groups. | 18 | # specifically and is read from param_groups. |