Enable Cross layers KV cache layout at NIXL Connector by liranschour · Pull Request #30207 · vllm-project/vllm

liranschour · 2025-12-07T13:17:01Z

Purpose

Enable NIXL Connector to us the new continuous cross layer KV cache layout described in RFC #27742 and implemented in #27743.
Demonstrate performance improvement of more the 2x in Tok/sec and TTFT due to dramatic reduction of fragmentation of transfer buffers.

Tested with P!=D with run_accuracy_test.sh P=1 D=2

Tested on H100 1P1D Qwen/Qwen3-0.6B:

branch	num reqs	input len	TTFT	ITL	tok/s	Desc/transfer
main	1000	16	18756.42	5.35	5288.41	56
kv_cross_layers	1000	16	8494.44	8.74	8572.56	1
main	128	1024	1660.84	9.40	37945.20	3528
kv_cross_layers	128	1024	686.98	9.26	55418.76	1
main	128	10240	11140.52	42.78	62339.74	34000
kv_cross_layers	128	10240	5226.71	14.41	117631.48	422

gemini-code-assist

Code Review

This pull request enables the use of a continuous cross-layer KV cache layout with the NIXL connector, which is a great performance optimization. The changes correctly signal the preference for this layout and add the necessary hooks to handle it.

However, I've found a critical issue in the implementation within NixlConnectorWorker.register_kv_caches. The existing logic does not correctly handle the shape of the cross-layer KV cache tensor, leading to incorrect calculation of num_blocks and num_layers. This will likely cause the KV cache transfer to fail for multi-layer models. I've added a detailed comment on this.

Once this issue is addressed, the PR should be in good shape.

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

mergify · 2025-12-07T13:21:20Z

Hi @liranschour, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2025-12-07T13:22:43Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

+        cross_layer_name = "ALL_LAYERS"
+
+        kv_caches = {cross_layer_name: kv_cache}
+        self.connector_worker.cross_layers = True
+        self.connector_worker.register_kv_caches(kv_caches)


Cross-layer caches break post-processing for heterogeneous blocks

The new cross-layer registration stores a single combined KV tensor under one key and sets cross_layers = True, but downstream post-processing still assumes split K/V caches whenever kv_topo.split_k_and_v is true. In the heterogeneous block-size path (block_size_ratio > 1) the code in blocksize_post_process/permute_device_kv indexes device_kv_caches as if each value were a (k,v) tuple, so with the cross-layer tensor it iterates over its first dimension and sample_cache = …[0] becomes a single block rather than a cache tensor. That path will raise shape errors or silently corrupt transfers as soon as heterogeneous block sizes are encountered, which are explicitly supported by this connector. The cross-layer registration should either propagate a non-split layout or adjust the post-processing to handle the combined tensor.

Useful? React with 👍 / 👎.

Signed-off-by: Liran Schour <lirans@il.ibm.com>

mergify · 2025-12-08T11:08:03Z

Hi @liranschour, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Signed-off-by: Liran Schour <lirans@il.ibm.com>

orozery

Looks good overall. Thanks!

Also wanted to share a summary I prepared on how this change effects the I/O size per different models:

Model	Old I/O size	New I/O size
deepseek-ai/DeepSeek-R1-Distill-Qwen-32B (tensor_parallel_size=2)	16 KB	2 MB
deepseek-ai/DeepSeek-V2-Lite-Chat (GPU block size=64)	72 KB	1.9 MB
meta-llama/Llama-3.1-8B-Instruct	32 KB	2 MB
meta-llama/Llama-3.2-1B-Instruct	16 KB	0.5 MB
meta-llama/Llama-3.1-70B-Instruct	8 KB	1.25 MB
mistralai/Mistral-7B-Instruct-v0.2	32 KB	2 MB
mistralai/Mistral-Small-24B-Instruct-2501	32 KB	2.5 MB
Qwen/Qwen2.5-3B-Instruct	8 KB	0.56 MB
Qwen/Qwen3-0.6B	32 KB	1.75 MB
Qwen/Qwen2.5-7B-Instruct	16 KB	0.87 MB
Qwen/Qwen3-4B-Instruct-2507	32 KB	2.25 MB
Qwen/Qwen2.5-1.5B-Instruct	8 KB	0.44 MB
Qwen/Qwen3-8B	28 KB	1.97 MB
Qwen/Qwen3-1.7B	32 KB	1.75 MB
Qwen/Qwen3-32B (tensor_parallel_size=2)	16 KB	2 MB

orozery · 2025-12-10T06:32:13Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

        if vllm_config.kv_transfer_config is None:
            raise ValueError("kv_transfer_config must be set for NixlConnector")
        self.kv_transfer_config = vllm_config.kv_transfer_config
+        self.cross_layers = False


I think that we should move this variable into TpKVTopology.
Then TpKVTopology.split_k_and_v can return False if we are in a cross layers case.
This should also address the bug that gemini pointed out in the current revision of this PR.

This change means we should defer the initialization of TpKVTopology from NixlConnectorWorker.__init__ to NixlConnectorWorker.register_kv_caches.

Done. Moved this logic under TpKVToplogy

orozery · 2025-12-10T06:34:17Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

        # TODO (NickLucche): Get kernel_block_size in a cleaner way
        # NHD default "view" for non-MLA cache
-        if self.device_type == "cpu":
+        if self.device_type == "cpu" or self.cross_layers:
            block_size_position = -2
        else:
            block_size_position = -2 if self.use_mla else -3


As I suggested to eliminate the self.cross_layers variable,
I suggest that we move this entire logic behind a new property TpKVTopology.block_size_position.
cc @NickLucche

Done. Moved this logic under TpKVToplogy

orozery · 2025-12-10T06:38:47Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

        # Attention backend and KV cache dtype affect memory layout
        "attn_backend_name": attn_backend_name,
        "cache_dtype": str(cache_config.cache_dtype),
+        "cross_layers": CROSS_LAYERS,


An option to consider:
Defer the computation of self.compat_hash to NixlConnectorWorker.register_kv_caches.
This will allow you to conditionally set this new cross_layers compatibility constraint only if cross layers is actually being used.

I deferred the compatibility computation to NixlConnectorWorker.register_kv_caches

Signed-off-by: Liran Schour <lirans@il.ibm.com>

mergify · 2025-12-11T09:08:38Z

Hi @liranschour, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

NickLucche

Thanks for the work @liranschour and @orozery !
Planning to review/run some benchmarks asap, in the meantime can we run tp_config_sweep.ch to verify heteroTP correctness with this layout? Thanks!

Also cc @xuechendi

NickLucche · 2025-12-11T09:42:47Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

@@ -74,6 +74,7 @@
 #   2: Add remote_request_id to kv_transfer_params
 #
 NIXL_CONNECTOR_VERSION: int = 2
+CROSS_LAYERS: bool = True


what's this for?

I will remove this. It is a leftover from previous version of the code.

Removed @NickLucche

@NickLucche I am able to run successfully tp_config_sweep_accuracy_test.sh only on Qwen/Qwen3-0.6B:

"GPU_MEMORY_UTILIZATION=0.6 PREFILLER_TP_SIZE=2 DECODER_TP_SIZE=2"

"GPU_MEMORY_UTILIZATION=0.6 PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2"

DeppSeek models are failing on my environment also on the main branch. So probably I have some error in the environment.

Signed-off-by: Liran Schour <lirans@il.ibm.com>

mergify · 2025-12-11T13:30:57Z

Hi @liranschour, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

Signed-off-by: Liran Schour <lirans@il.ibm.com>

mergify · 2025-12-14T14:31:30Z

Hi @liranschour, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

Tip

Is mypy or markdownlint failing?

mypy and markdownlint are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:

# For mypy (substitute "3.10" with the failing version if needed)
pre-commit run --hook-stage manual mypy-3.10
# For markdownlint
pre-commit run --hook-stage manual markdownlint

orozery · 2025-12-15T07:21:14Z

vllm/distributed/kv_transfer/kv_connector/utils.py

    attn_backend: type[AttentionBackend]
    engine_id: str
    remote_block_size: dict[str, int]
+    cross_layers: bool


nit: maybe change to cross_layers_blocks / cross_layers_kv_cache or something else that is a bit more self-explanatory?
Also applies for the other cross_layers introduced in this PR.

orozery · 2025-12-15T07:23:47Z

vllm/distributed/kv_transfer/kv_connector/utils.py

        return self.remote_block_size[self.engine_id]

+    def block_size_position(self, device_type: str) -> int:
+        if device_type == "cpu" or self.cross_layers:


I think that device_type is nixl-connector specific. Can you move it out from here to the nixl-connector?

orozery · 2025-12-15T07:28:36Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

        logger.debug("Detected attention backend %s", self.backend_name)
        logger.debug("Detected kv cache layout %s", self.kv_cache_layout)

-        self.compat_hash = compute_nixl_compatibility_hash(


I think it's better to leave initialization to None here, so that all fields are declared in __init__.
See for example how it's done in gpu_model_runner.py:

# Lazy initializations # self.model: nn.Module # Set after load_model # Initialize in initialize_kv_cache self.kv_caches: list[torch.Tensor] = [] # Initialize in initialize_kv_cache_tensors self.cross_layers_kv_cache: torch.Tensor | None = None self.cross_layers_attn_backend: type[AttentionBackend] | None = None

orozery · 2025-12-15T07:30:36Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

    def register_kv_caches(self, kv_caches: dict[str, torch.Tensor]):
        """Register the KV Cache data in nixl."""

+        backend = get_attn_backend(


nit: introduce self.attn_backend in __init__ and re-use it here?

orozery · 2025-12-15T07:37:19Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

+            is_mla=self.use_mla,
+            total_num_kv_heads=self.model_config.get_total_num_kv_heads(),
+            attn_backend=backend,
+            cross_layers=next(iter(kv_caches)) == "ALL_LAYERS",


There is a small edge-case if a model has a layer named ALL_LAYERS :)
In the offloading connector we detect a cross layers kv cache differently:

vllm/vllm/v1/kv_offload/worker/cpu_gpu.py

Lines 197 to 205 in 3327807

test_shape = attn_backend.get_kv_cache_shape(

num_blocks=1234, block_size=16, num_kv_heads=8, head_size=256

)

has_layers_dim = False

if len(gpu_shape) != len(test_shape):

# cross-layers tensor

# shape is (num_blocks, ...)

assert len(gpu_shape) == len(test_shape) + 1

Actually, maybe you can introduce TpKVTopology.tensor_shape initialized from the kv_caches tensor shape, and then set _cross_layers in TpKVTopology.__post_init__ ?

orozery · 2025-12-15T07:47:37Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

+        self.enforce_compat_hash = self.kv_transfer_config.get_from_extra_config(
+            "enforce_handshake_compat", True
+        )


This can be moved back to __init__.

orozery · 2025-12-15T07:49:36Z

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

+            attn_backend=backend,
+            cross_layers=next(iter(kv_caches)) == "ALL_LAYERS",
+        )
+        self._use_pallas = self.kv_topo._use_pallas


We can either move this back to __init__ (setting it using attn_backend == AttentionBackendEnum.PALLAS),
or remove it all together and switch self._use_pallas to self.kv_topo.use_pallas (new property).

…0207) Signed-off-by: Liran Schour <lirans@il.ibm.com> Signed-off-by: liranschour <liranschour@users.noreply.github.com> Co-authored-by: Or Ozeri <or@ozery.com> Signed-off-by: mohammad najafi <mohammad.najafi@amd.com>

…0207) Signed-off-by: Liran Schour <lirans@il.ibm.com> Signed-off-by: liranschour <liranschour@users.noreply.github.com> Co-authored-by: Or Ozeri <or@ozery.com> Signed-off-by: 陈建华 <1647430658@qq.com>

…0207) Signed-off-by: Liran Schour <lirans@il.ibm.com> Signed-off-by: liranschour <liranschour@users.noreply.github.com> Co-authored-by: Or Ozeri <or@ozery.com>

…roject#30207)" This reverts commit 64e3d67.

…roject#30207)" This reverts commit 64e3d67. Signed-off-by: Or Ozeri <oro@il.ibm.com>

…" (#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

…" (#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86)

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> Signed-off-by: PiratePai <416932041@qq.com> Signed-off-by: Pai <416932041@qq.com>

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86)

…0207) Signed-off-by: Liran Schour <lirans@il.ibm.com> Signed-off-by: liranschour <liranschour@users.noreply.github.com> Co-authored-by: Or Ozeri <or@ozery.com>

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86)

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86) feat(otel): production-ready OpenTelemetry logging (streaming, SGLang-compatible format) \nSquashed commits:\n- feat: Production-ready OpenTelemetry logging with streaming support\n- fix(otel): match SGLang logging format with proper kvlistValue structure\n feat(openai): streaming improvements for completions API - max_tokens: null support - remove asserts in serving_completion.py so null max_tokens uses computed max_model_len - prompt_length value - UTF-8 streaming fix - hold back any delta containing U+FFFD replacement char in incremental detokenizer to prevent streaming corrupted multi-byte chars like "�️�️" for emojis - gpt-oss special tokens - default skip_special_tokens=False for harmony models in chat completions when caller doesn't explicitly set it, so protocol framing tokens are preserved in output Tested on /v1/completions stream=true endpoint: - max_tokens:null streams successfully - stream_options.include_usage returns usage chunks - emoji/UTF-8 streaming produces clean output (no U+FFFD) - skip_special_tokens:false accepted without error feat(kimi): comprehensive Kimi K2 tool call streaming fixes - Fix same-tool-multiple-times streaming using re.finditer for correct match indexing - Fix previous_texts[i] accumulation for named tool_choice streaming - Add Kimi K2 marker detection and extraction helpers - Handle raw JSON output when reasoning parser returns it as reasoning content - Add per-tool name tracking (tool_name_sent_arr) for parallel tool calls - Strip all tool markers from leaked content after section ends - Add finish-time handling for tools whose names weren't sent during streaming - Handle string vs dict arguments in remaining args logic - Update KIMI_K2_VLLM_CHANGES.md with comprehensive porting documentation Test results: - 15/15 parallel tool calls - 10/10 concurrent streaming requests - 121/121 full tool calling suite checks - ~95% edge case tests (failures are model behavior, not bugs) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86)

liranschour requested review from ApostaC and NickLucche as code owners December 7, 2025 13:17

mergify bot added the kv-connector label Dec 7, 2025

gemini-code-assist bot reviewed Dec 7, 2025

View reviewed changes

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py Show resolved Hide resolved

chatgpt-codex-connector bot reviewed Dec 7, 2025

View reviewed changes

liranschour added 2 commits December 7, 2025 13:31

Cross layers implementation

5195265

Signed-off-by: Liran Schour <lirans@il.ibm.com>

Fix linting

0f36888

Signed-off-by: Liran Schour <lirans@il.ibm.com>

liranschour force-pushed the nixl_kv_cont_cross_layers branch from d75ca47 to 0f36888 Compare December 7, 2025 13:37

mergify bot added the v1 label Dec 8, 2025

liranschour force-pushed the nixl_kv_cont_cross_layers branch from b3b5d45 to 0f36888 Compare December 8, 2025 14:34

Add cross layers compatibility check

8d36b4b

Signed-off-by: Liran Schour <lirans@il.ibm.com>

orozery requested changes Dec 10, 2025

View reviewed changes

Move cross_layers logic into TpKVTopology

2a20197

Signed-off-by: Liran Schour <lirans@il.ibm.com>

liranschour force-pushed the nixl_kv_cont_cross_layers branch from 4bf547e to 2a20197 Compare December 11, 2025 08:33

NickLucche reviewed Dec 11, 2025

View reviewed changes

Code review minor fix

073b30e

Signed-off-by: Liran Schour <lirans@il.ibm.com>

liranschour force-pushed the nixl_kv_cont_cross_layers branch from 85d5cf0 to 073b30e Compare December 11, 2025 12:52

Linting...

b403a9e

Signed-off-by: Liran Schour <lirans@il.ibm.com>

liranschour force-pushed the nixl_kv_cont_cross_layers branch from 67d23fe to b403a9e Compare December 14, 2025 14:27

liranschour requested a review from orozery December 14, 2025 14:29

orozery requested changes Dec 15, 2025

View reviewed changes

liranschour force-pushed the nixl_kv_cont_cross_layers branch 4 times, most recently from ad45e7c to 02718ad Compare December 17, 2025 10:14

nicole-lihui mentioned this pull request Jan 26, 2026

[Bug]: PD report DeepseekV32 AssertionError: num_kv_heads == 1 #33074

Open

1 task

chaunceyjiang mentioned this pull request Jan 26, 2026

[Bugfix] Fix DeepseekV32 AssertionError: num_kv_heads == 1 #33086

Closed

5 tasks

liranschour mentioned this pull request Jan 28, 2026

Fix prefer_cross_layers check #33231

Closed

5 tasks

orozery added a commit to orozery/vllm that referenced this pull request Jan 28, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (vllm-p…

3a9f961

…roject#30207)" This reverts commit 64e3d67.

orozery mentioned this pull request Jan 28, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (#30207)" #33241

Merged

orozery added a commit to orozery/vllm that referenced this pull request Jan 28, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (vllm-p…

a37185f

…roject#30207)" This reverts commit 64e3d67. Signed-off-by: Or Ozeri <oro@il.ibm.com>

NickLucche pushed a commit that referenced this pull request Jan 28, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (#30207)…

2e8de86

…" (#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

khluu added a commit that referenced this pull request Jan 28, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (#30207)…

fe18ce4

…" (#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> (cherry picked from commit 2e8de86)

apd10 pushed a commit to apd10/vllm that referenced this pull request Jan 31, 2026

Revert "Enable Cross layers KV cache layout at NIXL Connector (vllm-p…

5cf644d

…roject#30207)" (vllm-project#33241) Signed-off-by: Or Ozeri <oro@il.ibm.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>

ZhanqiuHu mentioned this pull request Feb 18, 2026

Revert "[Bugfix] Disable TRTLLM attention with KV transfer enabled (#33192)" #34832

Merged

5 tasks

	test_shape = attn_backend.get_kv_cache_shape(
	num_blocks=1234, block_size=16, num_kv_heads=8, head_size=256
	)

	has_layers_dim = False
	if len(gpu_shape) != len(test_shape):
	# cross-layers tensor
	# shape is (num_blocks, ...)
	assert len(gpu_shape) == len(test_shape) + 1

Uh oh!

Conversation

liranschour commented Dec 7, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

mergify bot commented Dec 7, 2025

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Dec 7, 2025

Choose a reason for hiding this comment

Uh oh!

mergify bot commented Dec 8, 2025

Uh oh!

orozery left a comment

Choose a reason for hiding this comment

Uh oh!

orozery Dec 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

orozery Dec 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mergify bot commented Dec 11, 2025

Uh oh!

NickLucche left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mergify bot commented Dec 11, 2025

Uh oh!

mergify bot commented Dec 14, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

liranschour commented Dec 7, 2025 •

edited by github-actions bot

Loading

orozery Dec 10, 2025 •

edited

Loading

orozery Dec 10, 2025 •

edited

Loading