File size: 10,855 Bytes
1faccd4 | 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 | Megatron-LM Backend
===================
Last updated: 12/01/2025.
We support Megatron Backend by implementing various workers for actor,
critic, reference, rollout and reward models. We also implement the
``3DHybridEngine`` using Megatron-LM and vLLM/SGLang in
`megatron_vllm.py <https://github.com/volcengine/verl/blob/main/verl/workers/sharding_manager/megatron_vllm.py>`_
and `megatron_sglang.py <https://github.com/volcengine/verl/blob/main/verl/workers/sharding_manager/megatron_sglang.py>`_.
**Pros**
- Support 5D parallelism (TP, EP, CP, DP, PP) and sequence parallelism
for best scalablility and throughput.
- 3D HybridEngine can significantly reduce peak memory usage and reduce
weight synchronize overhead between actor and rollout.
**Cons**
- Huggingface Models and Megatron checkpoints need tools for conversion.
Development Progress
--------------------
Note that [Deprecated] means that the feature is not supported in the latest
version of verl.
[To-Optimize] means that the feature is implemented but not optimized yet.
[WIP] means that the feature is working in progress.
[In-Release] means that the feature is ready and in review process,
coming at any time.
+---------------+-----------------------------------------------------------+
| [Deprecated] | Megatron 3D Parallelism with custom models |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron 0.11.0 ``GPTModel`` support |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron GRPO support |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron with vLLM 0.8.2, with per-tensor weights loading |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron with Context Parallel |
+---------------+-----------------------------------------------------------+
| [Done] | Qwen2MoE model support |
+---------------+-----------------------------------------------------------+
| [To-Optimize] | Megatron dist Checkpoint |
+---------------+-----------------------------------------------------------+
| [To-Optimize] | Huggingface and Megatron Checkpoint Converter |
+---------------+-----------------------------------------------------------+
| [To-Optimize] | Efficient fused linear, entropy and cross entropy |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron offload(param, grad, optimizer) |
+---------------+-----------------------------------------------------------+
| [Done] | Megatron Profiler |
+---------------+-----------------------------------------------------------+
| [In-Release] | Megatron 0.12.0, TE 2.2 with vLLM 0.8.3 and Fused Attn |
+---------------+-----------------------------------------------------------+
| [WIP] | Moonlight/DeepSeek-V3 model support |
+---------------+-----------------------------------------------------------+
| [WIP] | Expert Parallel support |
+---------------+-----------------------------------------------------------+
| [WIP] | Megatron support dynamic batch size |
+---------------+-----------------------------------------------------------+
| [To-Do] | Performance tuning |
+---------------+-----------------------------------------------------------+
| [MileStone] | Runnable with DeepSeek-V3 671B post-training |
+---------------+-----------------------------------------------------------+
Utils of Megatron Workers
-------------------------
MegatronWorker
^^^^^^^^^^^^^^
``MegatronWorker`` is the base class of different megatron worker
classes. In this class, ``get_megatron_global_info`` and
``get_megatron_rank_info`` function to retrieve the 3D parallel world
size and rank of each ``Worker`` running on specific GPU. These information
will be used in transfer protocol for Megatron Backend.
The following ``Worker`` class for different models will be utilized to
construct the ``WorkerGroup`` .
We implement various of APIs for each ``Worker`` class decorated by the
``@register(dispatch_mode=)`` . These APIs can be called by the ray
driver process. The data can be correctly collect and dispatch following
the ``dispatch_mode`` on each function. The supported dispatch_model
(i.e., transfer protocols) can be found in `decorator.py <https://github.com/volcengine/verl/blob/main/verl/single_controller/base/decorator.py>`_.
ActorRolloutRefWorker
^^^^^^^^^^^^^^^^^^^^^
This class is implemented for Actor/Rollout HybridEngine or for the
reference model to initialize their model and perform computation.
Actor/Rollout HybridEngine
''''''''''''''''''''''''''
1. HybridEngine, Actor and Rollout initialization API.
.. code:: python
@register(dispatch_mode=Dispatch.ONE_TO_ALL)
def init_model(self):
``ONE_TO_ALL``: when calling the ``init_model`` function from the driver
process, each worker (on a GPU) will execute the following model
initialization process.
The initialization details of HybridEngine, Actor and Rollout are
highlighted below:
1. ``MegatronPPOActor`` implements the simple PPO computation logics
when the model is built with Megatron, including compute log prob,
model update.
2. ``vLLMRollout`` support generation with vLLM. We modify the vLLM
Engine and make it executed under SPMD to fit into our
``WorkerGroup`` design.
See `source code <https://github.com/volcengine/verl/blob/main/verl/workers/megatron_workers.py#L63>`_ for more information.
.. code:: python
# build actor model
self.actor = MegatronPPOActor(config=self.config.actor,
model_config=self.actor_model_config,
megatron_config=megatron_config,
actor_module=self.actor_module,
actor_optimizer=self.actor_optimizer,
actor_optimizer_config=self.actor_optim_config)
# build rollout
# rollout initialization
rollout = vLLMRollout(actor_module=params,
config=self.config.rollout,
tokenizer=self.tokenizer,
model_hf_config=self.actor_model_config,
train_tp=mpu.get_tensor_model_parallel_world_size())
...
1. Generate sequence and recompute log prob
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_PP_AS_DP_PROTO)
def generate_sequences(self, prompts: DataProto):
- ``Dispatch.MEGATRON_PP_AS_DP_PROTO``: The PP dimension of the actor
model will be regarded as DP dimension. Then the driver process will
dispatch and collect the data according to this reorganization. This
is because, in HybridEngine, the actor weight, which usually applied
larger 3D parallel sizes, will be gathered along the PP dimension and
TP dimension. Therefore, the corresponding data should be dispatched
and collected through the 3D parallel group of the rollout model,
rather than the actor model. However, the world_size and rank
information can only be retrieved from ``get_megatron_global_info`` and
``get_megatron_rank_info``, which records the 3D information for the
actor model. Moreover, the data resharding inside TP dimension will be
processed within the HybridEngine.
- In this function, the rollout model will perform auto-regressive
generation and the actor model will recompute the old log prob for the
generated response.
3. Update actor model
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_COMPUTE_PROTO)
def update_actor(self, data: DataProto):
- ``Dispatch.MEGATRON_COMPUTE_PROTO``: User passes the data partitioned
by DP dimension. The data is dispatched to all tp/pp ranks within the
same dp group, and ultimately only collects output data from tp=0 and
the last pp.
- Update the actor model weight using PPO & entropy loss.
..note::
Currently, training Tensor Parallel Size can be different from inference
Tensor Parallel Size.
ReferenceModel
''''''''''''''
1. Reference model initialization
The reference model is initialized using the same function as the actor
model without initializing the HybridEngine and Optimizer. Then the
actor model is also wrapped by the ``MegatronPPOActor``.
2. Compute reference log prob
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_COMPUTE_PROTO)
def compute_ref_log_prob(self, data: DataProto):
- In this function, the reference model will call the compute log prob
function in ``MegatronPPOActor`` to compute the reference log prob.
CriticWorker and RewardWorker
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1. Model initialization
Quite similar to reference model. The CriticWorker will perform
additional initialization for the Optimizer.
2. Compute Values for CriticWorker
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_COMPUTE_PROTO)
def compute_values(self, data: DataProto):
3. Update Critic
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_COMPUTE_PROTO)
def update_critic(self, data: DataProto):
4. Compute Reward
.. code:: python
@register(dispatch_mode=Dispatch.MEGATRON_COMPUTE_PROTO)
def compute_rm_score(self, data: DataProto):
Utils of Train Optimization
---------------------------
Offload
^^^^^^^
When resources are tight, the offload method can lower GPU memory
usage, helping training and inference frameworks work well under verl.
It moves parameters, gradients, and optimizers to CPU memory and only
loads them back to the GPU when needed.
If you want to use the offload, you can add the following parameters
for the actor and ref separately.
.. code:: python
# For the actor
actor_rollout_ref.actor.megatron.param_offload=True \
actor_rollout_ref.actor.megatron.grad_offload=True \
actor_rollout_ref.actor.megatron.optimizer_offload=True \
# For the ref w/o grad and optimizer
actor_rollout_ref.ref.megatron.param_offload=True \
For the critic, you can include these parameters.
.. code:: python
# For the critic
critic.megatron.param_offload=True \
critic.megatron.grad_offload=True \
critic.megatron.optimizer_offload=True \
Related MCore Document
----------------------
There is also a detailed document of using MCore to train different
kinds of models, please refer to `MCore Document <https://github.com/volcengine/verl/blob/main/verl/models/mcore/readme.md>`_.
|