Rename: --hooks to --forward-hooks (#13994)

docs/advanced_features/forward_hooks.md

@@ -20,14 +20,14 @@ Hooks are configured via a `ServerArgs` field:
 class ServerArgs:
     ...
     # For forward hooks
-    hooks: Optional[List[dict[str, Any]]] = None
+    forward_hooks: Optional[List[dict[str, Any]]] = None
 ````
 
 In JSON form, a minimal configuration looks like:
 
 ```jsonc
 {
-  "hooks": [
+  "forward_hooks": [
     {
       "name": "outer_linear_hooks",
       "target_modules": ["outer.0", "outer.1"],
@@ -42,7 +42,7 @@ In JSON form, a minimal configuration looks like:
 
 #### Top-level fields
 
-* `hooks` (optional list of objects)
+* `forward_hooks` (optional list of objects)
   Each element is a hook spec describing:
 
   * Which modules to target
@@ -53,7 +53,7 @@ In JSON form, a minimal configuration looks like:
 
 ### Hook spec schema
 
-Each entry in `hooks` is a JSON object with the following shape:
+Each entry in `forward_hooks` is a JSON object with the following shape:
 
 ```jsonc
 {
@@ -149,16 +149,16 @@ The first two cause initialization to fail fast with a descriptive error; the la
 Hooks are registered in `ModelRunner.initialize()`:
 
 ```python
-if server_args.hooks:
-    register_hooks(self.model, server_args.hooks)
+if server_args.forward_hooks:
+    register_forward_hooks(self.model, server_args.forward_hooks)
 ```
 
-The actual registration logic is implemented by `register_hooks`:
+The actual registration logic is implemented by `register_forward_hooks`:
 
 ```python
-def register_hooks(model: nn.Module, hook_specs: List[dict[str, Any]]) -> None:
+def register_forward_hooks(model: nn.Module, hook_specs: List[dict[str, Any]]) -> None:
     """
-    hook_specs is a list of dicts from server_args.hooks.
+    hook_specs is a list of dicts from server_args.forward_hooks.
     Attaches forward hooks to the matching modules.
     """
     name_to_module = dict(model.named_modules())
@@ -255,7 +255,7 @@ In JSON:
 
 ```jsonc
 {
-  "hooks": [
+  "forward_hooks": [
     {
       "name": "capture_outer",
       "target_modules": ["outer.0", "outer.1"],
@@ -279,7 +279,7 @@ This will:
 
 ### Summary
 
-* Define `hooks` as a list of specs in `ServerArgs` to turn on the feature.
+* Define `forward_hooks` as a list of specs in `ServerArgs` to turn on the feature.
 
 * Each spec:
 

docs/advanced_features/server_arguments.md

@@ -403,7 +403,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
 ## Forward hooks
 | Argument | Description | Defaults | Options |
 | --- | --- | --- | --- |
-| `--hooks` | JSON-formatted list of hook specifications. Each element must include `target_modules` (list of glob patterns matched against `model.named_modules()` names) and `hook_factory` (Python import path to a factory, e.g. `my_package.hooks:make_hook`). An optional `name` field is used for logging, and an optional `config` object is passed as a `dict` to the factory. | `None` | Type: JSON list |
+| `--forward-hooks` | JSON-formatted list of forward hook specifications. Each element must include `target_modules` (list of glob patterns matched against `model.named_modules()` names) and `hook_factory` (Python import path to a factory, e.g. `my_package.hooks:make_hook`). An optional `name` field is used for logging, and an optional `config` object is passed as a `dict` to the factory. | `None` | Type: JSON list |
 
 ## Debug tensor dumps
 | Argument | Description | Defaults | Options |

python/sglang/srt/model_executor/hook_manager.py

@@ -8,9 +8,9 @@ import torch.nn as nn
 logger = logging.getLogger(__name__)
 
 
-def register_hooks(model: nn.Module, hook_specs: List[dict[str, Any]]) -> None:
+def register_forward_hooks(model: nn.Module, hook_specs: List[dict[str, Any]]) -> None:
     """
-    hook_specs is a list of dicts from server_args.hooks.
+    hook_specs is a list of dicts from server_args.forward_hooks.
     Attaches forward hooks to the matching modules.
     """
     name_to_module = dict(model.named_modules())

python/sglang/srt/model_executor/model_runner.py

@@ -122,7 +122,7 @@ from sglang.srt.mem_cache.memory_pool import (
 from sglang.srt.model_executor.cpu_graph_runner import CPUGraphRunner
 from sglang.srt.model_executor.cuda_graph_runner import CudaGraphRunner
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
-from sglang.srt.model_executor.hook_manager import register_hooks
+from sglang.srt.model_executor.hook_manager import register_forward_hooks
 from sglang.srt.model_executor.npu_graph_runner import NPUGraphRunner
 from sglang.srt.model_executor.piecewise_cuda_graph_runner import (
     PiecewiseCudaGraphRunner,
@@ -535,8 +535,8 @@ class ModelRunner:
             self.graph_mem_usage = 0
             self.init_attention_backend()
 
-        if server_args.hooks:
-            register_hooks(self.model, server_args.hooks)
+        if server_args.forward_hooks:
+            register_forward_hooks(self.model, server_args.forward_hooks)
 
         # auxiliary hidden capture mode. TODO: expose this to server args?
         if self.spec_algorithm.is_eagle3() and not self.is_draft_worker:

python/sglang/srt/server_args.py

@@ -588,7 +588,7 @@ class ServerArgs:
     mm_enable_dp_encoder: bool = False
 
     # For forward hooks
-    hooks: Optional[List[dict[str, Any]]] = None
+    forward_hooks: Optional[List[dict[str, Any]]] = None
 
     def __post_init__(self):
         """
@@ -3870,10 +3870,10 @@ class ServerArgs:
 
         # For registering hooks
         parser.add_argument(
-            "--hooks",
+            "--forward-hooks",
             type=json_list_type,
-            default=None,
-            help="The hooks to be attached.",
+            default=ServerArgs.forward_hooks,
+            help="JSON-formatted forward hook specifications to attach to the model.",
         )
 
     @classmethod

test/srt/test_model_hooks.py

@@ -4,7 +4,7 @@ import json
 import torch
 import torch.nn as nn
 
-from sglang.srt.model_executor.hook_manager import register_hooks
+from sglang.srt.model_executor.hook_manager import register_forward_hooks
 from sglang.srt.server_args import ServerArgs
 from sglang.test.test_utils import CustomTestCase
 
@@ -46,7 +46,7 @@ class TinyModel(nn.Module):
 
 
 class TestAttachHooks(CustomTestCase):
-    """Tests for ModelRunner.register_hooks / resolve_callable integration."""
+    """Tests for register_forward_hooks / resolve_callable integration."""
 
     def setUp(self):
         HOOK_CALLS.clear()
@@ -67,7 +67,7 @@ class TestAttachHooks(CustomTestCase):
         ]
 
         model = TinyModel()
-        register_hooks(model, hook_specs)
+        register_forward_hooks(model, hook_specs)
 
         x = torch.randn(3, 4)
         _ = model(x)
@@ -92,7 +92,7 @@ class TestAttachHooks(CustomTestCase):
             }
         ]
 
-        register_hooks(model, hook_specs)
+        register_forward_hooks(model, hook_specs)
 
         x = torch.randn(3, 4)
         _ = model(x)
@@ -103,7 +103,7 @@ class TestAttachHooks(CustomTestCase):
     def test_cli_hooks_reach_model(self):
         """
         Ensure that when hooks are provided via CLI, they are parsed into
-        ServerArgs, passed to ModelRunner.register_hooks, and actually
+        ServerArgs, passed to register_forward_hooks, and actually
         run during a forward pass.
         """
         parser = argparse.ArgumentParser()
@@ -121,17 +121,17 @@ class TestAttachHooks(CustomTestCase):
         cli_args = [
             "--model-path",
             "Qwen/Qwen2-7B-Instruct",  # Dummy value; not used in this test
-            "--hooks",
+            "--forward-hooks",
             json.dumps(hooks_spec),
         ]
 
         args = parser.parse_args(cli_args)
         server_args = ServerArgs.from_cli_args(args)
 
-        self.assertEqual(server_args.hooks, hooks_spec)
+        self.assertEqual(server_args.forward_hooks, hooks_spec)
 
         model = TinyModel()
-        register_hooks(model, server_args.hooks)
+        register_forward_hooks(model, server_args.forward_hooks)
 
         x = torch.randn(3, 4)
         _ = model(x)