test swift for qwen3 math
This commit is contained in:
commit
8f168ecbef
2
.gitignore
vendored
Normal file
2
.gitignore
vendored
Normal file
@ -0,0 +1,2 @@
|
||||
output/
|
||||
result/
|
10
README.md
Normal file
10
README.md
Normal file
@ -0,0 +1,10 @@
|
||||
# 试验使用swift 对qwen3-8b进行grpo训练
|
||||
- 数据集:modelscope提供数学
|
||||
- 启动方法:external模式,因为在a6000上使用colocate方式启动会报内存不足,因此使用2张卡跑vllm,2张卡跑训练,基本都跑满
|
||||
```bash
|
||||
# start server
|
||||
sh swift_server.sh # 要等到看到vllm服务启动
|
||||
|
||||
# start client
|
||||
sh swift_client.sh # 启动训练任务
|
||||
```
|
16
swift_client.sh
Normal file
16
swift_client.sh
Normal file
@ -0,0 +1,16 @@
|
||||
CUDA_VISIBLE_DEVICES=2,3 \
|
||||
NPROC_PER_NODE=2 \
|
||||
swift rlhf \
|
||||
--rlhf_type grpo \
|
||||
--model /data1/yuyr/qwen3-8b \
|
||||
--dataset AI-MO/NuminaMath-TIR#5000 \
|
||||
--reward_funcs accuracy cosine \
|
||||
--use_vllm true \
|
||||
--vllm_mode server \
|
||||
--vllm_server_host localhost \
|
||||
--vllm_server_port 8000 \
|
||||
--per_device_train_batch_size 8 \
|
||||
--per_device_eval_batch_size 8 \
|
||||
--async_generate true \
|
||||
--num_generations 4 \
|
||||
--deepspeed zero3
|
5
swift_server.sh
Normal file
5
swift_server.sh
Normal file
@ -0,0 +1,5 @@
|
||||
CUDA_VISIBLE_DEVICES=0,1 \
|
||||
swift rollout \
|
||||
--model /data1/yuyr/qwen3-8b \
|
||||
--tensor_parallel_size 2 \
|
||||
--port 8000
|
Loading…
x
Reference in New Issue
Block a user