checkpoint-engine/patches/vllm_fp8.patch at main · kip-cxj/checkpoint-engine · GitHub

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
From cb707077b0df8bb02adadc7eda63c97a8c6d6c76 Mon Sep 17 00:00:00 2001
From: weixiao-huang <hwx.simle@gmail.com>
Date: Sun, 7 Sep 2025 14:56:43 +0800
Subject: [PATCH] [BugFix] use _wrap_parameter_or_copy instead of using
 Parameter and add missing scale attributes

Signed-off-by: weixiao-huang <hwx.simle@gmail.com>
---
 .../model_executor/layers/quantization/fp8.py | 40 ++++++++++++++-----
 .../layers/quantization/kv_cache.py           |  7 ++++
 2 files changed, 36 insertions(+), 11 deletions(-)

diff --git a/vllm/model_executor/layers/quantization/fp8.py b/vllm/model_executor/layers/quantization/fp8.py
index 65e0b7062..46f1885cc 100644
--- a/vllm/model_executor/layers/quantization/fp8.py
+++ b/vllm/model_executor/layers/quantization/fp8.py
@@ -65,6 +65,25 @@ def _is_col_major(x: torch.Tensor) -> bool:
     return x.stride(0) == m * n and x.stride(1) == 1 and x.stride(2) == m


+def _wrap_parameter_or_copy(layer: torch.nn.Module, name: str,
+                            weight: torch.Tensor):
+    layer_weight = getattr(layer, name)
+    if isinstance(layer_weight, Parameter):
+        # If it is already a Parameter, we assume it is the right shape
+        # directly copy it from weight to keep pointer unchanged in CUDA Graph
+        layer_weight.copy_(weight)
+    else:
+        # torch.compile() cannot use Parameter subclasses.
+        # but these weights are already Parameter
+        # so this can be compatible with torch.compile
+        param = Parameter(weight, requires_grad=False)
+        if hasattr(layer_weight, "weight_loader"):
+            # keep the weight_loader attribute to make sure
+            # the weight can be loaded correctly in weight update
+            param.weight_loader = layer_weight.weight_loader
+        setattr(layer, name, param)
+
+
 class Fp8Config(QuantizationConfig):
     """Config class for FP8."""

@@ -387,10 +406,9 @@ class Fp8LinearMethod(LinearMethodBase):

             weight = self._maybe_pad_weight(weight)

-            # Torch.compile cannot use Parameter subclasses.
-            layer.weight = Parameter(weight, requires_grad=False)
-            layer.weight_scale_inv = Parameter(weight_scale_inv,
-                                               requires_grad=False)
+            _wrap_parameter_or_copy(layer, "weight", weight)
+            _wrap_parameter_or_copy(layer, "weight_scale_inv",
+                                    weight_scale_inv)

         # If checkpoint not serialized fp8, quantize the weights.
         elif not self.quant_config.is_checkpoint_fp8_serialized:
@@ -740,13 +758,13 @@ class Fp8MoEMethod(FusedMoEMethodBase):
                 w2_weight = layer.w2_weight
                 w2_weight_scale_inv = layer.w2_weight_scale_inv

-            # torch.compile() cannot use Parameter subclasses.
-            layer.w13_weight = Parameter(w13_weight, requires_grad=False)
-            layer.w13_weight_scale_inv = Parameter(w13_weight_scale_inv,
-                                                   requires_grad=False)
-            layer.w2_weight = Parameter(w2_weight, requires_grad=False)
-            layer.w2_weight_scale_inv = Parameter(w2_weight_scale_inv,
-                                                  requires_grad=False)
+            _wrap_parameter_or_copy(layer, "w13_weight", w13_weight)
+            _wrap_parameter_or_copy(layer, "w13_weight_scale_inv",
+                                    w13_weight_scale_inv)
+            _wrap_parameter_or_copy(layer, "w2_weight", w2_weight)
+            _wrap_parameter_or_copy(layer, "w2_weight_scale_inv",
+                                    w2_weight_scale_inv)
+
             if self.rocm_aiter_moe_enabled:
                 # reshaping weights is required for aiter moe kernel.
                 shuffled_w13, shuffled_w2 = shuffle_weights(
diff --git a/vllm/model_executor/layers/quantization/kv_cache.py b/vllm/model_executor/layers/quantization/kv_cache.py
index e5604670f..0483e9b32 100644
--- a/vllm/model_executor/layers/quantization/kv_cache.py
+++ b/vllm/model_executor/layers/quantization/kv_cache.py
@@ -48,6 +48,13 @@ class BaseKVCacheMethod(QuantizeMethodBase):
             f"{self.__class__.__name__}.apply should not be called.")

     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
+        # update weights may miss these attributes, we create it if not present
+        if not hasattr(layer, "q_scale"):
+            assert not hasattr(layer, "k_scale")
+            assert not hasattr(layer, "v_scale")
+            assert not hasattr(layer, "prob_scale")
+            self.create_weights(layer)
+
         # If the kv-cache dtype is auto, we enforce the k/v_scale to be 1.0
         # regardless whether the kv-scale is available in the checkpoint.
         # No need to process kv scales after loading if we are going to
--
2.39.3 (Apple Git-146)