[Main][Ops] Make triton rope support index_selecting from cos_sin_cache by Angazenn · Pull Request #5450 · vllm-project/vllm-ascend

Angazenn · 2025-12-28T13:29:57Z

What this PR does / why we need it?

This PR extends original rope_triton_forward and split_qkv_rmsnorm_rope to support cos_sin_cache && positions as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are:

avoiding pre-computation of cos sin before model execution, which helps to remove redundant codes.
allowing eagle3 draft model to have different rope parameters with main model (see [Bug]: Potential accuracy && accept rate degradation if rope parameters in eagle3 draft model is different from main model. #6612 ). This help to recover accept rate && accuracy in that case.

In addition, this kernel change only introduces very small performance degradation. Those index_select or chunk operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81).

Highlights

RoPE Cache Unification: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling.
Triton Kernel Integration: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations.
Custom Operation Registration: Registered rope_forward_oot as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation.
Refactored RoPE Forward Pass: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

vLLM version: v0.13.0
vLLM main: vllm-project/vllm@5326c89

Additional test on Qwen3-235b accuracy:

Aime2024	GSM8K	Livecodebench
83.33	96.26	70.23

github-actions · 2025-12-28T13:30:05Z

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

A PR should do only one thing, smaller PRs enable faster reviews.
Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

gemini-code-assist

Code Review

This pull request refactors the Rotary Positional Embedding (RoPE) implementation to use a combined cos_sin_cache, which is a good simplification. However, I've found a few critical issues. The new Triton kernel for RoPE does not correctly use the token positions, leading to incorrect calculations. Additionally, an old RoPE function (rope_forward_triton) is now broken due to using undefined variables. Finally, the shape of the tensors returned by the new RoPE implementation is incorrect, which will likely cause errors in downstream attention operations. These issues need to be addressed before this PR can be merged.

github-actions · 2026-01-05T06:32:29Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

wangxiyuan · 2026-01-05T12:02:17Z

Any progress? If this PR is still alive, please rebase to main and make CI happy, otherwise you can close it. Thanks

github-actions · 2026-01-16T13:02:08Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

github-actions · 2026-02-02T11:17:42Z

This pull request has conflicts, please resolve those before we can evaluate the pull request.

Signed-off-by: Angazenn <supperccell@163.com>

…cache (#6602) ### What this PR does / why we need it? This PR adapts #5450, #6523 to v0.13.0, to fix #6612 . This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see #6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. --------- Signed-off-by: Angazenn <supperccell@163.com>

…to qwen3next_rebase * 'main' of https://github.com/vllm-project/vllm-ascend: [Docs] Fix GLM-5 deploy command (vllm-project#6711) [npugraph_ex]enable npugraph_ex by default (vllm-project#6664) [doc]add GLM5.md (vllm-project#6709) [Model] GLM5 adaptation (vllm-project#6642) [Bugfix] Update target probs to target logits in rejection sample (vllm-project#6685) [Main][Ops] Make triton rope support index_selecting from cos_sin_cache (vllm-project#5450) [CI]fix nightly multi node test error for wait for pod ready (vllm-project#6675) [main to main] upgrade main 0210 (vllm-project#6673) [main][Quant] Remove unused rotation functions and parameters from W4A4 LAOS quantization (vllm-project#6648) [Test][BugFix] Fix torch.rand usage in triton penalty test (vllm-project#6680) Add Worker Interface:check_health (vllm-project#6681)

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com> Signed-off-by: momochenchuw <chenchuw@huawei.com>

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com>

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com> Signed-off-by: zrj026 <zhangrunjiang026@gmail.com>

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com>

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com> Signed-off-by: zrj026 <zhangrunjiang026@gmail.com>

…he (vllm-project#5450) ### What this PR does / why we need it? This PR extends original `rope_triton_forward` and `split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as inputs. This fully aligns to vLLM RoPE api interface. Compared with earlier implementation for RoPE, the benefits are: 1. avoiding pre-computation of `cos` `sin` before model execution, which helps to remove redundant codes. 2. allowing eagle3 draft model to have different rope parameters with main model (see vllm-project#6612 ). This help to recover accept rate && accuracy in that case. In addition, this kernel change only introduces very small performance degradation. Those `index_select` or `chunk` operations are now changed into simple memory access in triton kernel (For example, https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81). **Highlights** - **RoPE Cache Unification**: Replaced separate _sin and _cos global tensors with a unified cos_sin_cache and explicit positions tensor for Rotary Positional Embeddings (RoPE), streamlining data handling. - **Triton Kernel Integration**: Updated Triton kernels (split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the cos_sin_cache and positions for more efficient and integrated RoPE calculations. - **Custom Operation Registration**: Registered `rope_forward_oot` as a new custom operation, allowing its use in fused compilation passes and providing a dedicated entry point for the new RoPE implementation. - **Refactored RoPE Forward Pass**: Modified the rope_forward_oot function to accept the new cos_sin_cache and positions arguments, enabling a more flexible and integrated RoPE application within the system. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? - vLLM version: v0.13.0 - vLLM main: vllm-project/vllm@5326c89 Additional test on Qwen3-235b accuracy: | Aime2024 | GSM8K | Livecodebench | | -------- | -------- | -------- | | 83.33 | 96.26 | 70.23 | --------- Signed-off-by: Angazenn <supperccell@163.com>

github-actions Bot added the module:ops label Dec 28, 2025

gemini-code-assist Bot reviewed Dec 28, 2025

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

Comment thread vllm_ascend/ops/triton/rope.py Outdated

Comment thread vllm_ascend/ops/rotary_embedding.py

Angazenn force-pushed the cos_sin branch from 653c0e2 to a590270 Compare December 28, 2025 13:39

whx-sjtu requested changes Dec 29, 2025

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

whx-sjtu requested changes Dec 29, 2025

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

Comment thread vllm_ascend/ops/triton/rope.py Outdated

whx-sjtu reviewed Dec 29, 2025

View reviewed changes

Comment thread vllm_ascend/ops/rotary_embedding.py Outdated

Angazenn force-pushed the cos_sin branch from b1d2abd to 4c0ba28 Compare December 30, 2025 16:38

github-actions Bot added the merge-conflicts label Jan 5, 2026

Angazenn force-pushed the cos_sin branch from 4c0ba28 to 8b6e81c Compare January 7, 2026 13:35

github-actions Bot removed the merge-conflicts label Jan 7, 2026

github-actions Bot added the merge-conflicts label Jan 16, 2026

Angazenn force-pushed the cos_sin branch from dc1d379 to 0839976 Compare February 2, 2026 08:09

github-actions Bot added merge-conflicts and removed merge-conflicts labels Feb 2, 2026

Angazenn force-pushed the cos_sin branch 2 times, most recently from 6b0ec47 to df33b34 Compare February 5, 2026 11:23

github-actions Bot removed the merge-conflicts label Feb 5, 2026

Angazenn force-pushed the cos_sin branch 2 times, most recently from e6a1b57 to b3f5744 Compare February 5, 2026 11:27

Angazenn marked this pull request as ready for review February 5, 2026 11:28

Angazenn requested review from MengqingCao, realliujiaxu, wangxiyuan, yiz-liu and zzzzwwjj as code owners February 5, 2026 11:28

Angazenn marked this pull request as draft February 9, 2026 07:45

Angazenn marked this pull request as ready for review February 9, 2026 07:46

whx-sjtu changed the title ~~[draft]rope support cos_sin_cache~~ [Ops] Make triton rope support index_selecting from cos_sin_cache Feb 9, 2026

modify nightly

486c92f

Signed-off-by: Angazenn <supperccell@163.com>

Angazenn force-pushed the cos_sin branch from d8c17a5 to 486c92f Compare February 9, 2026 08:07

wangxiyuan approved these changes Feb 9, 2026

View reviewed changes

optimize

f530321

Signed-off-by: Angazenn <supperccell@163.com>

Angazenn force-pushed the cos_sin branch from 3caa277 to f530321 Compare February 9, 2026 14:26

Angazenn added 2 commits February 10, 2026 00:51

fix lint

b2fec7b

Signed-off-by: Angazenn <supperccell@163.com>

add nightly

7a805f3

Signed-off-by: Angazenn <supperccell@163.com>

Angazenn changed the title ~~[Ops] Make triton rope support index_selecting from cos_sin_cache~~ [Main][Ops] Make triton rope support index_selecting from cos_sin_cache Feb 10, 2026

Angazenn mentioned this pull request Feb 10, 2026

[v0.13.0][Ops] Make triton rope support index_selecting from cos_sin_cache #6602

Merged

winson-00178005 reviewed Feb 10, 2026

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

winson-00178005 reviewed Feb 10, 2026

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

winson-00178005 reviewed Feb 10, 2026

View reviewed changes

Comment thread vllm_ascend/ops/triton/rope.py Outdated

optimize

79faae3

Signed-off-by: Angazenn <supperccell@163.com>

whx-sjtu reviewed Feb 11, 2026

View reviewed changes

Comment thread vllm_ascend/ops/register_custom_ops.py Outdated

change registered name

76b31ca

Signed-off-by: Angazenn <supperccell@163.com>

whx-sjtu approved these changes Feb 11, 2026

View reviewed changes

whx-sjtu merged commit c0c2eb6 into vllm-project:main Feb 11, 2026
27 checks passed

Angazenn mentioned this pull request Mar 2, 2026

[Bug]: Potential accuracy && accept rate degradation if rope parameters in eagle3 draft model is different from main model. #6612

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Main][Ops] Make triton rope support index_selecting from cos_sin_cache#5450

[Main][Ops] Make triton rope support index_selecting from cos_sin_cache#5450
whx-sjtu merged 7 commits intovllm-project:mainfrom
Angazenn:cos_sin

Angazenn commented Dec 28, 2025 •

edited

Loading

Uh oh!

github-actions Bot commented Dec 28, 2025

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Jan 5, 2026

Uh oh!

wangxiyuan commented Jan 5, 2026

Uh oh!

github-actions Bot commented Jan 16, 2026

Uh oh!

github-actions Bot commented Feb 2, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

Angazenn commented Dec 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this PR does / why we need it?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

github-actions Bot commented Dec 28, 2025

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Jan 5, 2026

Uh oh!

wangxiyuan commented Jan 5, 2026

Uh oh!

github-actions Bot commented Jan 16, 2026

Uh oh!

github-actions Bot commented Feb 2, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Angazenn commented Dec 28, 2025 •

edited

Loading