[Model] Support Qwen3.8-Flash-Next (#53896)

Signed-off-by: huanghaoyan.hhy <huanghaoyan.hhy@alibaba-inc.com>
Signed-off-by: andyluo7 <andy.luo@amd.com>
Signed-off-by: lishunyang12 <lishunyang03@gmail.com>
Signed-off-by: Harry Huang <huanghaoyan.hhy@alibaba-inc.com>
Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: KungYork <firelikest@163.com>
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: andyluo7 <43718156+andyluo7@users.noreply.github.com>
Co-authored-by: Isotr0py <mozf@inferact.ai>
Co-authored-by: andyluo7 <andy.luo@amd.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: lishunyang12 <lishunyang03@gmail.com>
Co-authored-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Harry Huang
2026-08-31 05:57:56 +00:00
committed by GitHub
co-authored by zjy0516 KungYork Thien Tran Yifan Qiao andyluo7 Isotr0py andyluo7 OpenAI Codex lishunyang12 Vadim Gimpelson Isotr0py Lucas Wilkinson Claude Opus 5 OpenAI Codex
parent 687db59744
commit e126687a9a
124 changed files with 19424 additions and 450 deletions
+44
View File
@@ -6,6 +6,7 @@ Run `pytest tests/quantization/test_fp8.py --forked`.
"""
import logging
from types import SimpleNamespace
import pytest
import regex as re
@@ -31,6 +32,10 @@ from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
from vllm.model_executor.layers.quantization.online.fp8 import (
Fp8PerTensorOnlineLinearMethod,
)
from vllm.model_executor.layers.quantization.utils import flashinfer_utils
from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
prepare_fp8_moe_layer_for_fi,
)
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
process_fp8_input_tensor_strategy_moe,
)
@@ -50,6 +55,45 @@ MODELS = [
]
def test_prepare_gated_trtllm_fp8_moe_weights_pads_each_projection(monkeypatch):
monkeypatch.setattr(
flashinfer_utils,
"rotate_weights_for_fi_trtllm_fp8_per_tensor_moe",
lambda *args: None,
)
intermediate = 17
padded_intermediate = 32
hidden_size = 4
gate = torch.ones((1, intermediate, hidden_size), dtype=torch.float8_e4m3fn)
up = torch.full_like(gate, 2)
w13 = torch.cat((gate, up), dim=1)
w2 = torch.ones((1, hidden_size, intermediate), dtype=torch.float8_e4m3fn)
layer = SimpleNamespace(
activation=SimpleNamespace(is_gated=True),
moe_config=SimpleNamespace(
is_act_and_mul=True,
intermediate_size_per_partition=intermediate,
),
)
padded_w31, _, _, _ = prepare_fp8_moe_layer_for_fi(
layer,
w13,
w2,
w13_scale=torch.ones(1),
w13_input_scale=torch.ones(1),
w2_scale=torch.ones(1),
w2_input_scale=torch.ones(1),
is_trtllm=True,
)
expected = w13.new_zeros((1, 2 * padded_intermediate, hidden_size))
expected[:, :intermediate] = up
expected[:, padded_intermediate : padded_intermediate + intermediate] = gate
assert layer.moe_config.intermediate_size_per_partition == padded_intermediate
assert torch.equal(padded_w31, expected)
def test_static_fp8_moe_input_scales_remain_scalar() -> None:
a1_scale, a2_scale = process_fp8_input_tensor_strategy_moe(
torch.tensor([0.25, 0.5]),