[ROCm] [Model] Enable TML inkling (#48841)

Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
TJian
2026-07-27 10:05:46 +00:00
committed by GitHub
co-authored by OpenAI Codex
parent bc3629b1c4
commit 0906123953
33 changed files with 9213 additions and 9 deletions
@@ -0,0 +1,20 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import torch
from torch import nn
from vllm.models.inkling.amd.sconv_swa_attn import InklingConvState
def test_runtime_sconv_block_size_tracks_unified_cache_page():
"""The cache planner may enlarge W=4 blocks to match attention pages."""
owner = InklingConvState.__new__(InklingConvState)
nn.Module.__init__(owner)
owner.block_size = 4
owner.kv_cache = torch.tensor([])
assert owner.cache_block_size == 4
owner.kv_cache = torch.empty(2, 1, 32, 1024)
assert owner.cache_block_size == 32