MCPcopy Create free account

hub / github.com/RL-Align/RL-Kernel / functions

Functions1,290 in github.com/RL-Align/RL-Kernel

⨍ Functions 1,290◇ Types & classes 193↳ Endpoints 3
Method__init__
(self, worker_factory: Any, *args: Any, **kwargs: Any)
rl_engine/executors/ray_actor_manager.py:185
Method__init__
( self, model_config: Optional[dict] = None, *, weight_bridge: Optional[Weight
rl_engine/executors/rollout.py:28
Method__init__
(self)
rl_engine/kernels/sampling.py:12
Method__init__
(self)
rl_engine/kernels/registry.py:86
Method__init__
(self)
rl_engine/kernels/ops/triton/loss/grpo_loss.py:69
Method__init__
(self)
rl_engine/kernels/ops/rocm/attention/flash_attn.py:24
Method__init__
No state; the op is a pure function over (q, k_cache, v_cache, k_new, v_new, ...).
rl_engine/kernels/ops/pytorch/attention/kv_cache.py:50
Method__init__
No state; the op is a pure function over (q, k, v, ...).
rl_engine/kernels/ops/pytorch/attention/standard_attn.py:43
Method__init__
(self)
rl_engine/kernels/ops/pytorch/linear/matmul.py:19
Method__init__
No state; the op is a pure function over (hidden, weight, bias).
rl_engine/kernels/ops/pytorch/linear/lm_head.py:34
Method__init__
(self)
rl_engine/kernels/ops/pytorch/linear/embedding.py:20
Method__init__
(self)
rl_engine/kernels/ops/pytorch/norm/rms_norm.py:15
Method__init__
(self)
rl_engine/kernels/ops/pytorch/activation/swiglu.py:55
Method__init__
(self)
rl_engine/kernels/ops/pytorch/loss/ratio_kl.py:16
Method__init__
(self)
rl_engine/kernels/ops/pytorch/loss/logp.py:14
Method__init__
(self)
rl_engine/kernels/ops/pytorch/loss/linear_logp.py:452
Method__init__
(self)
rl_engine/kernels/ops/pytorch/loss/grpo_loss.py:21
Method__init__
(self)
rl_engine/kernels/ops/pytorch/rotary_embedding/rope.py:21
Method__init__
(self)
rl_engine/kernels/ops/cuda/attention/flash_attn.py:16
Method__init__
(self)
rl_engine/kernels/ops/cuda/attention/prefix_shared_attn.py:20
Method__init__
(self)
rl_engine/kernels/ops/cuda/loss/logp.py:15
Method__init__
(self)
rl_engine/kernels/ops/cuda/loss/logp.py:95
Method__init__
(self)
rl_engine/kernels/ops/cuda/loss/linear_logp.py:260
Method__init__
(self, candidate: Any)
rl_engine/kernels/gtest/operator_specs.py:64
Method__init__
( self, model: torch.nn.Module, *, freeze: bool = True, eval_mode: boo
rl_engine/alignment/model_wrappers.py:116
Method__init__
(self)
rl_engine/platforms/constants.py:97
Method__init__
(self)
rl_engine/platforms/device.py:18
Method__init__
(self)
tests/test_stateless_training_contract.py:100
Method__init__
(self)
tests/test_stateless_training_contract.py:166
Method__init__
(self, logits: torch.Tensor, *, output_kind: str = "tensor")
tests/test_alignment_model_wrappers.py:21
Method__init__
(self, value)
tests/test_ray_actor_manager.py:27
Method__init__
(self, method)
tests/test_ray_actor_manager.py:32
Method__init__
(self, instance, options)
tests/test_ray_actor_manager.py:42
Method__init__
(self, fake_ray, actor_class, options=None)
tests/test_ray_actor_manager.py:51
Method__init__
(self)
tests/test_ray_actor_manager.py:69
Method__init__
(self, label="ray-rollout")
tests/test_ray_actor_manager.py:101
Method__init__
(self, publish_delta=1)
tests/test_ray_actor_manager.py:117
Method__init__
(self)
tests/test_ray_actor_manager.py:133
Method__init__
(self)
tests/test_ray_actor_manager.py:154
Method__init__
(self, shape)
tests/test_weight_sync_bridge.py:62
Method__init__
(self, **kwargs)
tests/test_weight_sync_bridge.py:884
Method__init__
(self, logits: torch.Tensor)
tests/test_paged_kv_baseline.py:22
Method__init__
(self, model, optimizer)
tests/test_deepspeed_training_worker.py:23
Method__init__
(self)
tests/test_deepspeed_training_worker.py:49
Method__init__
(self, parameters, modifier_rank=0)
tests/test_deepspeed_training_worker.py:67
Method__init__
(self)
tests/test_deepspeed_training_worker.py:118
Method__init__
(self, logits: torch.Tensor)
tests/test_stateless_executor.py:26
Method__init__
(self, rewards: torch.Tensor)
tests/test_stateless_executor.py:51
Method__init__
(self)
tests/test_stateless_executor.py:343
Method__init__
(self, **kwargs)
tests/test_vllm_rollout_sampler.py:24
Method__init__
(self, **kwargs)
tests/test_vllm_rollout_sampler.py:31
Method__init__
(self, config)
tests/test_vllm_rollout_sampler.py:252
Method__init__
(self, vocab_size: int, hidden_dim: int)
examples/grpo_single_gpu.py:50
Method__init__
(self, state_dict: Mapping[str, torch.Tensor])
benchmarks/benchmark_weight_sync_bridge.py:650
Method__init__
(self, device: torch.device | None = None, warmup: int = 3, repeat: int = 10)
benchmarks/profiler.py:164
Method__init__
(self, vocab_size: int, hidden_dim: int)
benchmarks/benchmark_stateless_executor.py:62
Method__init__
(self, vocab_size: int, hidden_dim: int)
benchmarks/benchmark_stateless_executor.py:81
Method__post_init__
(self)
rl_engine/executors/deepspeed_trainer.py:49
Method__post_init__
(self)
rl_engine/executors/stateless_executor.py:36
Method__post_init__
(self)
rl_engine/executors/paged_kv_baseline.py:45
Method__post_init__
(self)
rl_engine/executors/vllm_sampler.py:22
Method__str__
(self)
tests/test_weight_sync_bridge.py:54
Method_batch_from_token_groups
( self, token_groups: Sequence[Sequence[int]], rollout: RolloutStageResult, )
rl_engine/executors/training_contract.py:135
Function_bwd_kernel
( Q, K, V, sm_scale, Out, DO, DQ, DK, DV, L, M, Delta, str
rl_engine/kernels/ops/triton/triton_attn.py:186
Function_bwd_preprocess
( Out, DO, Delta, stride_oz, stride_oh, stride_om, stride_on, stride_doz,
rl_engine/kernels/ops/triton/triton_attn.py:135
Function_cuda_ipc_manifest_child
(manifest: Any, queue: Any)
benchmarks/benchmark_weight_sync_bridge.py:370
Function_cuda_vmm_manifest_child
(manifest: Any, queue: Any)
benchmarks/benchmark_weight_sync_bridge.py:406
Function_enough_gpu_memory
(num_bytes: int)
tests/test_embedding.py:169
Function_enough_gpu_memory
Return True only if CUDA is present and has free memory with headroom.
tests/test_lm_head.py:303
Function_extract_logits
(model_output: Any)
rl_engine/executors/deepspeed_trainer.py:411
Function_fwd_kernel
( Q, K, V, sm_scale, L, M, Out, stride_qz, stride_qh, stride_qm, s
rl_engine/kernels/ops/triton/triton_attn.py:7
Function_group_norm_kernel
( rewards_ptr, bounds_ptr, # int32[num_groups + 1], CSR-style group offsets adv_ptr, # float32[N
rl_engine/kernels/ops/triton/loss/grpo_loss.py:31
Function_info_on_rank
Experimental: Log only on a specific distributed rank. Useful for multi-node RL training to avoid log flooding.
rl_engine/utils/logger.py:30
Function_info_once
Log INFO message only once across the same logger instance.
rl_engine/utils/logger.py:20
Function_linear_logp_fwd_kernel
One program per token-block. Streams the vocab in BLOCK_V tiles, folding each ``hidden @ Wblk^T`` tile into an online-softmax state without ever
rl_engine/kernels/ops/triton/loss/linear_logp.py:24
Function_make_attention_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:93
Function_make_embedding_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:160
Function_make_kv_cache_attention_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:183
Function_make_linear_logp_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:116
Function_make_lm_head_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:171
Function_make_logp_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:105
Function_make_matmul_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:81
Function_make_rms_norm_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:69
Function_make_rope_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:130
Function_make_silu_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:141
Function_make_swiglu_inputs
( args: argparse.Namespace, dtype: torch.dtype, device: torch.device )
rl_engine/kernels/gtest/operator_inputs.py:150
Function_noop_deleter
(_ptr)
rl_engine/executors/bridge.py:922
Function_ratio_kl_bwd_kernel
( policy_ptr, action_ptr, mask_ptr, ratio_ptr, diff_ptr, logz_ptr, grad_ratio_ptr,
rl_engine/kernels/ops/triton/loss/ratio_kl.py:87
Function_ratio_kl_fwd_kernel
( policy_ptr, ref_ptr, action_ptr, mask_ptr, old_ptr, ratio_ptr, kl_ptr, diff_
rl_engine/kernels/ops/triton/loss/ratio_kl.py:22
Function_run_logp_fused_workload
( profiler: PerformanceProfiler, args: argparse.Namespace, device: torch.device, dtype: torch.
benchmarks/profiler.py:637
Function_run_logp_native_workload
( profiler: PerformanceProfiler, args: argparse.Namespace, device: torch.device, dtype: torch.
benchmarks/profiler.py:611
Function_run_sampling_native_workload
( profiler: PerformanceProfiler, args: argparse.Namespace, device: torch.device, dtype: torch.
benchmarks/profiler.py:684
Function_shared_memory_manifest_child
(manifest, queue)
tests/test_weight_sync_bridge.py:304
Function_shared_memory_manifest_child
(manifest: Any, queue: Any)
benchmarks/benchmark_weight_sync_bridge.py:133
Function_shared_memory_manifest_tamper_child
(manifest, queue)
tests/test_weight_sync_bridge.py:373
Function_shared_memory_manifest_tamper_child
(manifest: Any, queue: Any)
benchmarks/benchmark_weight_sync_bridge.py:151
Function_shared_memory_tensor_pickle_child
(imported, queue)
tests/test_weight_sync_bridge.py:298
Function_tp_linear_logp_gloo_worker
(rank, world_size, init_method, result_queue)
tests/test_linear_logp.py:64
Function_warn_once
Log WARNING message only once.
rl_engine/utils/logger.py:25
Functionadapter
(model_outputs, batch_inputs)
tests/test_stateless_executor.py:249
← previousnext →701–800 of 1,290, ranked by callers