由于手上只有4090，所以驱动比较老，如下所示，所以使用的是verl 0.4.1.x的版本(https://github.com/verl-project/verl/tree/v0.4.1.x)
1
2
3
4
5
6
7
8
9
10
11
12
13
| $ nvidia-smi
Thu Jan 29 15:55:02 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.67 Driver Version: 550.67 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 4090 On | 00000000:00:03.0 Off | Off |
| 30% 27C P8 13W / 450W | 1870MiB / 24564MiB | 0% Default |
| | | N/A |
...
|
Verl安装
这里使用容器进行安装，使用如下命令拉取容器
1
| docker pull verlai/verl:base-verl0.4-cu124-cudnn9.8-torch2.6-fa2.7.4
|
然后我使用的是vscode，所以可以直接连接进容器中，然后在容器中安装带vllm的verl
最终安装的各版本如下：
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
| root@10-60-189-161:/workspace/verl# pip3 list
Package Version Build Editable project location
---------------------------------------- --------------- -------- -------------------------
absl-py 2.1.0
accelerate 1.8.1
aiohappyeyeballs 2.3.5
aiohttp 3.10.1
aiohttp-cors 0.8.1
aiosignal 1.3.1
airportsdata 20250909
annotated-types 0.7.0
antlr4-python3-runtime 4.9.3
anyio 4.4.0
apex 0.1
argon2-cffi 23.1.0
argon2-cffi-bindings 21.2.0
arrow 1.3.0
asciitree 0.3.3
astor 0.8.1
asttokens 2.4.1
astunparse 1.6.3
async-lru 2.0.4
async-timeout 4.0.3
attrs 24.2.0
audioread 3.0.1
av 14.4.0
babel 2.16.0
beautifulsoup4 4.12.3
black 24.8.0
blake3 1.0.8
bleach 6.1.0
blis 0.7.11
blobfile 3.0.0
cachetools 5.4.0
catalogue 2.0.10
certifi 2024.7.4
cffi 1.17.0
cfgv 3.4.0
charset-normalizer 3.3.2
click 8.1.7
cloudpathlib 0.18.1
cloudpickle 3.0.0
cmake 3.30.2
codetiming 1.4.0
colorful 0.5.6
comm 0.2.2
compressed-tensors 0.9.3
confection 0.1.5
contourpy 1.2.1
cuda-bindings 12.9.0
cuda-python 12.5.0
cudf 24.6.0
cugraph 24.6.0
cugraph-dgl 24.6.0
cugraph-equivariant 24.6.0
cugraph-pyg 24.6.0
cugraph-service-client 24.6.0
cugraph-service-server 24.6.0
cuml 24.6.0
cupy-cuda12x 13.2.0
cycler 0.12.1
cymem 2.0.8
Cython 3.0.11
dask 2024.5.1
dask-cuda 24.6.0
dask-cudf 24.6.0
dask-expr 1.1.1
datasets 3.6.0
dbus-python 1.2.18
debugpy 1.8.5
decorator 5.1.1
defusedxml 0.7.1
Deprecated 1.3.1
depyf 0.18.0
dill 0.3.8
diskcache 5.6.3
distlib 0.3.9
distributed 2024.5.1
distributed-ucxx 0.38.0
distro 1.9.0
dm-tree 0.1.8
dnspython 2.7.0
einops 0.8.0
email_validator 2.2.0
entrypoints 0.4
exceptiongroup 1.2.2
execnet 2.1.1
executing 2.0.1
expecttest 0.1.3
fastapi 0.115.14
fastapi-cli 0.0.7
fasteners 0.19
fastjsonschema 2.20.0
fastrlock 0.8.2
filelock 3.20.3
flash_attn 2.7.4.post1
fonttools 4.53.1
fqdn 1.5.1
frozenlist 1.4.1
fsspec 2024.6.1
gast 0.6.0
gguf 0.17.1
gitdb 4.0.12
GitPython 3.1.44
google-api-core 2.25.1
google-auth 2.40.3
googleapis-common-protos 1.70.0
grpcio 1.73.1
h11 0.14.0
hf_transfer 0.1.9
hf-xet 1.1.5
httpcore 1.0.5
httptools 0.6.4
httpx 0.27.0
huggingface-hub 0.33.1
hydra-core 1.3.2
hypothesis 5.35.1
identify 2.6.12
idna 3.7
igraph 0.11.6
importlib_metadata 7.2.1
iniconfig 2.0.0
intel-openmp 2021.4.0
interegular 0.3.3
ipykernel 6.29.5
ipython 8.26.0
isoduration 20.11.0
isort 5.13.2
jedi 0.19.1
Jinja2 3.1.6
jiter 0.12.0
joblib 1.4.2
json5 0.9.25
jsonpointer 3.0.0
jsonschema 4.23.0
jsonschema-specifications 2023.12.1
jupyter_client 8.6.2
jupyter_core 5.7.2
jupyter-events 0.10.0
jupyter-lsp 2.2.5
jupyter_server 2.14.2
jupyter_server_terminals 0.5.3
jupyterlab 4.2.4
jupyterlab_code_formatter 3.0.2
jupyterlab_pygments 0.3.0
jupyterlab_server 2.27.3
jupyterlab-tensorboard-pro 4.0.0
jupytext 1.16.4
kiwisolver 1.4.5
kvikio 24.6.0
langcodes 3.4.0
language_data 1.2.0
lark 1.2.2
lazy_loader 0.4
librmm 24.6.0
librosa 0.10.1
liger_kernel 0.5.10
lightning-thunder 0.2.0.dev0
lightning-utilities 0.11.6
lintrunner 0.12.5
llguidance 0.7.30
llvmlite 0.44.0
lm-format-enforcer 0.10.12
locket 1.0.0
looseversion 1.3.0
lxml 6.0.0
marisa-trie 1.2.0
Markdown 3.7
markdown-it-py 3.0.0
MarkupSafe 2.1.5
mathruler 0.1.0
matplotlib 3.9.2
matplotlib-inline 0.1.7
mdit-py-plugins 0.4.1
mdurl 0.1.2
mistral_common 1.9.0
mistune 3.0.2
mkl 2021.1.1
mkl-devel 2021.1.1
mkl-include 2021.1.1
mock 5.1.0
mpmath 1.3.0
msgpack 1.0.8
msgspec 0.20.0
multidict 6.0.5
multiprocess 0.70.16
murmurhash 1.0.10
mypy-extensions 1.0.0
nbclient 0.10.0
nbconvert 7.16.4
nbformat 5.10.4
nest-asyncio 1.6.0
networkx 3.3
ninja 1.11.1.1
nodeenv 1.9.1
notebook 7.2.1
notebook_shim 0.2.4
numba 0.61.2
numcodecs 0.11.0
numpy 2.2.6
nvfuser 0.2.6a0+1d02b13
nvidia-cublas-cu12 12.4.5.8
nvidia-cuda-cupti-cu12 12.4.127
nvidia-cuda-nvrtc-cu12 12.4.127
nvidia-cuda-runtime-cu12 12.4.127
nvidia-cudnn-cu12 9.1.0.70
nvidia-cudnn-frontend 1.5.2
nvidia-cufft-cu12 11.2.1.3
nvidia-curand-cu12 10.3.5.147
nvidia-cusolver-cu12 11.6.1.9
nvidia-cusparse-cu12 12.3.1.170
nvidia-cusparselt-cu12 0.6.2
nvidia-dali-cuda120 1.40.0 16741769
nvidia-ml-py 12.575.51
nvidia-modelopt 0.15.0
nvidia-nccl-cu12 2.21.5
nvidia-nvimgcodec-cu12 0.3.0.5
nvidia-nvjitlink-cu12 12.4.127
nvidia-nvtx-cu12 12.4.127
nvidia-pyindex 1.0.9
nvtx 0.2.5
nx-cugraph 24.6.0
omegaconf 2.3.0
onnx 1.16.0
openai 2.16.0
opencensus 0.11.4
opencensus-context 0.1.3
opencv-python-headless 4.11.0.86
opentelemetry-api 1.26.0
opentelemetry-exporter-otlp 1.26.0
opentelemetry-exporter-otlp-proto-common 1.26.0
opentelemetry-exporter-otlp-proto-grpc 1.26.0
opentelemetry-exporter-otlp-proto-http 1.26.0
opentelemetry-exporter-prometheus 0.47b0
opentelemetry-proto 1.26.0
opentelemetry-sdk 1.26.0
opentelemetry-semantic-conventions 0.47b0
opentelemetry-semantic-conventions-ai 0.4.13
opt-einsum 3.3.0
optree 0.16.0
orjson 3.10.18
outlines 0.1.11
outlines_core 0.1.26
overrides 7.7.0
packaging 23.2
pandas 2.2.2
pandocfilters 1.5.1
parso 0.8.4
partd 1.4.2
partial-json-parser 0.2.1.1.post7
pathspec 0.12.1
peft 0.15.2
pexpect 4.9.0
pillow 10.4.0
pip 25.1.1
platformdirs 4.2.2
pluggy 1.5.0
ply 3.11
polygraphy 0.49.12
pooch 1.8.2
pre_commit 4.2.0
preshed 3.0.9
prometheus_client 0.20.0
prometheus-fastapi-instrumentator 7.1.0
prompt_toolkit 3.0.47
proto-plus 1.26.1
protobuf 4.25.8
psutil 6.0.0
ptyprocess 0.7.0
PuLP 2.9.0
pure_eval 0.2.3
py-cpuinfo 9.0.0
py-spy 0.4.0
pyarrow 20.0.0
pyasn1 0.6.1
pyasn1_modules 0.4.2
pybind11 2.13.4
pybind11_global 2.13.4
pycocotools 2.0+nv0.8.0
pycountry 24.6.1
pycparser 2.22
pycryptodomex 3.23.0
pydantic 2.11.7
pydantic_core 2.33.2
pydantic-extra-types 2.11.0
pyext 0.7
Pygments 2.18.0
PyGObject 3.42.1
pylatexenc 2.10
pylibcugraph 24.6.0
pylibcugraphops 24.6.0
pylibraft 24.6.0
pylibwholegraph 24.6.0
pynvjitlink 0.2.3
pyparsing 3.1.2
pytest 8.1.1
pytest-flakefinder 1.1.0
pytest-rerunfailures 14.0
pytest-shard 0.1.2
pytest-xdist 3.6.1
python-dateutil 2.9.0.post0
python-dotenv 1.1.1
python-hostlist 1.23.0
python-json-logger 2.0.7
python-multipart 0.0.20
pytz 2023.4
PyYAML 6.0.2
pyzmq 26.1.0
qwen-vl-utils 0.0.11
raft-dask 24.6.0
rapids-dask-dependency 24.6.0a0
ray 2.47.1
referencing 0.35.1
regex 2024.7.24
requests 2.32.3
rfc3339-validator 0.1.4
rfc3986-validator 0.1.1
rich 13.7.1
rich-toolkit 0.14.7
rmm 24.6.0
rpds-py 0.20.0
rsa 4.9.1
ruff 0.12.1
safetensors 0.5.3
scikit-learn 1.5.1
scipy 1.14.0
Send2Trash 1.8.3
sentencepiece 0.2.0
sentry-sdk 2.32.0
setproctitle 1.3.6
setuptools 70.3.0
shellingham 1.5.4
six 1.16.0
smart-open 7.0.4
smmap 5.0.2
sniffio 1.3.1
sortedcontainers 2.4.0
soundfile 0.12.1
soupsieve 2.6
soxr 0.4.0
spacy 3.7.5
spacy-legacy 3.0.12
spacy-loggers 1.0.5
srsly 2.4.8
stack-data 0.6.3
starlette 0.46.2
sympy 1.13.1
tabulate 0.9.0
tbb 2021.13.1
tblib 3.0.0
tensorboard 2.16.2
tensorboard-data-server 0.7.2
tensordict 0.6.2
tensorrt 10.3.0
terminado 0.18.1
texttable 1.7.0
thinc 8.2.5
threadpoolctl 3.5.0
thriftpy2 0.5.0
tiktoken 0.9.0
tinycss2 1.3.0
tokenizers 0.21.2
tomli 2.0.1
toolz 0.12.1
torch 2.6.0
torchaudio 2.6.0
torchdata 0.11.0
torchvision 0.21.0
tornado 6.2
tqdm 4.66.5
traitlets 5.14.3
transformers 4.53.0
treelite 4.1.2
triton 3.2.0
typer 0.12.4
types-dataclasses 0.6.6
types-python-dateutil 2.9.0.20240316
typing_extensions 4.12.2
typing-inspection 0.4.1
tzdata 2024.1
ucx-py 0.38.0
ucxx 0.38.0
uri-template 1.3.0
urllib3 2.0.7
uvicorn 0.35.0
uvloop 0.21.0
verl 0.4.1 /workspace/verl
virtualenv 20.31.2
vllm 0.8.5
wandb 0.20.1
wasabi 1.1.3
watchfiles 1.1.0
wcwidth 0.2.13
weasel 0.4.1
webcolors 24.8.0
webencodings 0.5.1
websocket-client 1.8.0
websockets 15.0.1
Werkzeug 3.0.3
wheel 0.44.0
wrapt 1.16.0
xdoctest 1.0.2
xformers 0.0.29.post2
xgrammar 0.1.18
xxhash 3.5.0
yarl 1.9.4
zarr 2.18.2
zict 3.0.0
zipp 3.19.2
|
Verl示例
- 首先下载数据集到
./data/gsm8k
1
| python3 examples/data_preprocess/gsm8k.py --local_dir ./data/gsm8k
|
- 然后下载模型到本地缓存
/root/.cache/huggingface/hub/models--Qwen--Qwen2.5-0.5B-Instruct/中
1
| python3 -c "import transformers; transformers.pipeline('text-generation', model='Qwen/Qwen2.5-0.5B-Instruct')"
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
| PYTHONUNBUFFERED=1 python3 -m verl.trainer.main_ppo \
data.train_files=./data/gsm8k/train.parquet \
data.val_files=./data/gsm8k/test.parquet \
data.train_batch_size=256 \
data.max_prompt_length=512 \
data.max_response_length=256 \
actor_rollout_ref.model.path=Qwen/Qwen2.5-0.5B-Instruct \
actor_rollout_ref.actor.optim.lr=1e-6 \
actor_rollout_ref.actor.ppo_mini_batch_size=64 \
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4 \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
actor_rollout_ref.rollout.gpu_memory_utilization=0.4 \
actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=4 \
critic.optim.lr=1e-5 \
critic.model.path=Qwen/Qwen2.5-0.5B-Instruct \
critic.ppo_micro_batch_size_per_gpu=4 \
algorithm.kl_ctrl.kl_coef=0.001 \
trainer.logger=['console'] \
trainer.val_before_train=False \
trainer.default_hdfs_dir=null \
trainer.n_gpus_per_node=1 \
trainer.nnodes=1 \
trainer.save_freq=10 \
trainer.test_freq=10 \
trainer.total_epochs=15 2>&1 | tee verl_demo.log
|
运行结果如下，可以看到确实开始了迭代：
Verl在vscode中的Debug配置
首先需要在vscode中下载Ray Distributed Debugger
然后需要先使用ray start --head来拉起ray集群
然后点击Ray Distributed Debugger插件添加对应ray集群的地址以完成连接
然后需要在需要debug的ray任务的代码中加入breakpoint()关键字
- 然后直接照常运行作业即可，运行到breakpoint时会自动连接到Ray Distributed Debugger，这时点击该debugger按钮即可进行debug
参考资料