[algo] feat: support GRPO algorithm (#124)

- Implement KL loss, GRPO outcome adv, and utilize bon rollouts
- Provide scripts for deepseek and qwen on GSM8k. Can provide more for
other datasets.
- Support seq balance
- Train using qwen2-7b, GSM8k score can reach 0.89

examples/grpo_trainer/run_deepseek7b_llm.sh

@@ -0,0 +1,39 @@
+set -x
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.val_batch_size=1312 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    actor_rollout_ref.model.path=deepseek-ai/deepseek-llm-7b-chat \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size=128 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=False \
+    actor_rollout_ref.actor.fsdp_config.grad_offload=False \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size=256 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size=256 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console','wandb'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='deepseek_llm_7b_function_rm' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 $@

examples/grpo_trainer/run_deepseek7b_llm_seq_balance.sh

@@ -0,0 +1,38 @@
+set -x
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.val_batch_size=1312 \
+    data.max_prompt_length=512 \
+    data.max_response_length=512 \
+    actor_rollout_ref.model.path=deepseek-ai/deepseek-llm-7b-chat \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.use_dynamic_bsz=True \
+    actor_rollout_ref.actor.ppo_max_token_len_per_gpu=24000 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=False \
+    actor_rollout_ref.actor.fsdp_config.grad_offload=False \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console','wandb'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='deepseek_llm_7b_function_rm_seq_packing' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 $@

examples/grpo_trainer/run_qwen2-7b.sh

@@ -0,0 +1,41 @@
+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.val_batch_size=1312 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    actor_rollout_ref.model.path=Qwen/Qwen2-7B-Instruct \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.ppo_micro_batch_size=128 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=False \
+    actor_rollout_ref.actor.fsdp_config.grad_offload=False \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size=256 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.log_prob_micro_batch_size=256 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console','wandb'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='qwen2_7b_function_rm' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 $@

examples/grpo_trainer/run_qwen2-7b_seq_balance.sh

@@ -0,0 +1,41 @@
+set -x
+
+export VLLM_ATTENTION_BACKEND=XFORMERS
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=grpo \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=1024 \
+    data.val_batch_size=1312 \
+    data.max_prompt_length=512 \
+    data.max_response_length=1024 \
+    actor_rollout_ref.model.path=Qwen/Qwen2-7B-Instruct \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
+    actor_rollout_ref.actor.use_dynamic_bsz=True \
+    actor_rollout_ref.actor.ppo_max_token_len_per_gpu=24000 \
+    actor_rollout_ref.actor.use_kl_loss=True \
+    actor_rollout_ref.actor.kl_loss_coef=0.001 \
+    actor_rollout_ref.actor.kl_loss_type=low_var_kl \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.fsdp_config.param_offload=False \
+    actor_rollout_ref.actor.fsdp_config.grad_offload=False \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
+    actor_rollout_ref.rollout.n=5 \
+    actor_rollout_ref.ref.fsdp_config.param_offload=True \
+    algorithm.kl_ctrl.kl_coef=0.001 \
+    trainer.critic_warmup=0 \
+    trainer.logger=['console','wandb'] \
+    trainer.project_name='verl_grpo_example_gsm8k' \
+    trainer.experiment_name='qwen2_7b_function_rm_kl1e-3' \
+    +trainer.val_before_train=False \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=15 $@

examples/ppo_trainer/run_qwen2-7b_rm_seq_balance.sh

@@ -58,4 +58,4 @@ python3 -m verl.trainer.main_ppo \
     trainer.nnodes=1 \
     trainer.save_freq=-1 \
     trainer.test_freq=5 \
-    trainer.total_epochs=100 $@
+    trainer.total_epochs=15 $@

examples/ppo_trainer/run_qwen2-7b_seq_balance.sh

@@ -49,4 +49,4 @@ python3 -m verl.trainer.main_ppo \
     trainer.nnodes=1 \
     trainer.save_freq=-1 \
     trainer.test_freq=5 \
-    trainer.total_epochs=100 $@
+    trainer.total_epochs=15 $@

tests/__init__.py

@@ -0,0 +1,13 @@
+# Copyright 2024 Bytedance Ltd. and/or its affiliates
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

tests/e2e/arithmetic_sequence/rl/config/ray_trainer.yaml

@@ -28,6 +28,9 @@ actor_rollout_ref:
     grad_clip: 1.0
     clip_ratio: 0.2
     entropy_coeff: 0.0
+    use_kl_loss: False # True for GRPO
+    kl_loss_coef: 0.001 # for grpo
+    kl_loss_type: low_var_kl # for grpo
     ppo_epochs: 1
     shuffle: False
     ulysses_sequence_parallel_size: 1 # sp size

tests/e2e/check_results.py

@@ -48,5 +48,5 @@ def extract_reward_from_line(line):
                 best_reward = reward
 
     print(f'Best reward is {best_reward}')
-    assert best_reward > 0.2, f'Best reward must be greater than 0.3. best_reward: {best_reward}'
+    assert best_reward > 0.2, f'Best reward must be greater than 0.2. best_reward: {best_reward}'
     print('Check passes')

verl/trainer/config/ppo_trainer.yaml

@@ -27,6 +27,9 @@ actor_rollout_ref:
     grad_clip: 1.0
     clip_ratio: 0.2
     entropy_coeff: 0.001
+    use_kl_loss: False # True for GRPO
+    kl_loss_coef: 0.001 # for grpo
+    kl_loss_type: low_var_kl # for grpo
     ppo_epochs: 1
     shuffle: False
     ulysses_sequence_parallel_size: 1 # sp size

verl/trainer/ppo/core_algos.py

@@ -20,6 +20,7 @@
 
 import numpy as np
 import torch
+from collections import defaultdict
 
 import verl.utils.torch_functional as verl_F
 
@@ -106,6 +107,54 @@ def compute_gae_advantage_return(token_level_rewards: torch.Tensor, values: torc
     return advantages, returns
 
 
+# NOTE(sgm): this implementation only consider outcome supervision, where the reward is a scalar.
+def compute_grpo_outcome_advantage(token_level_rewards: torch.Tensor,
+                                   eos_mask: torch.Tensor,
+                                   index: torch.Tensor,
+                                   epsilon: float = 1e-6):
+    """
+    Compute advantage for GRPO, operating only on Outcome reward 
+    (with only one scalar reward for each response).
+    Args:
+        token_level_rewards: `(torch.Tensor)`
+            shape: (bs, response_length)
+        eos_mask: `(torch.Tensor)`
+            shape: (bs, response_length)
+    
+    Returns:
+        advantages: `(torch.Tensor)`
+            shape: (bs, response_length)
+        Returns: `(torch.Tensor)`
+            shape: (bs, response_length)
+    """
+    response_length = token_level_rewards.shape[-1]
+    non_zero_mask = (token_level_rewards != 0)
+    scores = (token_level_rewards * non_zero_mask).sum(dim=-1)
+
+    id2score = defaultdict(list)
+    id2mean = {}
+    id2std = {}
+
+    with torch.no_grad():
+        bsz = scores.shape[0]
+        for i in range(bsz):
+            id2score[index[i]].append(scores[i])
+        for idx in id2score:
+            if len(id2score[idx]) == 1:
+                id2mean[idx] = torch.tensor(0.0)
+                id2std[idx] = torch.tensor(1.0)
+            elif len(id2score[idx]) > 1:
+                id2mean[idx] = torch.mean(torch.tensor(id2score[idx]))
+                id2std[idx] = torch.std(torch.tensor([id2score[idx]]))
+            else:
+                raise ValueError(f"no score in prompt index: {idx}")
+        for i in range(bsz):
+            scores[i] = (scores[i] - id2mean[index[i]]) / (id2std[index[i]] + epsilon)
+        scores = scores.unsqueeze(-1).tile([1, response_length]) * eos_mask
+
+    return scores, scores
+
+
 def compute_rewards(token_level_scores, old_log_prob, ref_log_prob, kl_ratio):
     kl = old_log_prob - ref_log_prob
     return token_level_scores - kl * kl_ratio
@@ -210,6 +259,14 @@ def kl_penalty(logprob: torch.FloatTensor, ref_logprob: torch.FloatTensor, kl_pe
     if kl_penalty == "mse":
         return 0.5 * (logprob - ref_logprob).square()
 
+    # J. Schulman. Approximating kl divergence, 2020.
+    # # URL http://joschu.net/blog/kl-approx.html.
+    if kl_penalty == 'low_var_kl':
+        kl = ref_logprob - logprob
+        ratio = torch.exp(kl)
+        kld = (ratio - kl - 1).contiguous()
+        return torch.clamp(kld, min=-10, max=10)
+
     if kl_penalty == "full":
         # so, here logprob and ref_logprob should contain the logits for every token in vocabulary
         raise NotImplementedError