feat: Support `mrope_section` with `rope_type: "yarn"` by raayandhar · Pull Request #13313 · sgl-project/sglang

raayandhar · 2025-11-15T02:07:03Z

Motivation

Support for this feature, as described: #13219

Modifications

New class: YaRNScalingMRotaryEmbedding to do MRoPE-enabled rotary embedding with YaRN context scaling. If we see a mrope_section and scaling_type == "yarn" then we can use this new class, otherwise we default back to YaRNScalingRotaryEmbedding.

Accuracy Tests

Added unit test

Benchmarking and Profiling

Checklist

Format your code according to the Format code with pre-commit.
Add unit tests according to the Run and add unit tests.
Update documentation according to Write documentations.
Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

python/sglang/srt/layers/rotary_embedding.py

raayandhar · 2025-11-15T22:39:35Z

Test locally with the same setup as #13238, on basic testing everything seems to be OK and functional now.

raayandhar · 2025-11-16T02:43:52Z

When it comes to testing, we can make changes to test_mrope.py to also test the rope scaling type by editing the config (and parameterizing it). However, when I test locally, this test seems to fail for me (just running pytest test/srt/rotary_embedding/test_mrope.py -v), I needed to add dummy server args and also change to _foward_native. I'm probably misunderstanding something about how to run this test. We could write a test similar to the cpu/test_rope.py?

yuan-luo · 2025-11-18T03:50:35Z

When it comes to testing, we can make changes to test_mrope.py to also test the rope scaling type by editing the config (and parameterizing it). However, when I test locally, this test seems to fail for me (just running pytest test/srt/rotary_embedding/test_mrope.py -v), I needed to add dummy server args and also change to _foward_native. I'm probably misunderstanding something about how to run this test. We could write a test similar to the cpu/test_rope.py?

@raayandhar Your running command have no problem. The test_mrope was broken due to some recent PR merged. We are trying to fix it. Currently have no workaround. Sorry for making confusion.

raayandhar · 2025-11-18T03:59:27Z

When it comes to testing, we can make changes to test_mrope.py to also test the rope scaling type by editing the config (and parameterizing it). However, when I test locally, this test seems to fail for me (just running pytest test/srt/rotary_embedding/test_mrope.py -v), I needed to add dummy server args and also change to _forward_native. I'm probably misunderstanding something about how to run this test. We could write a test similar to the cpu/test_rope.py?

@raayandhar Your running command have no problem. The test_mrope was broken due to some recent PR merged. We are trying to fix it. Currently have no workaround. Sorry for making confusion.

No worries, thanks so much for the clarification. Once there's a fix I'll add a test. I was able to get the test passing with the changes I mentioned above, if that helps. But that not be the correct way to do it.

raayandhar · 2025-11-21T01:20:34Z

When it comes to testing, we can make changes to test_mrope.py to also test the rope scaling type by editing the config (and parameterizing it). However, when I test locally, this test seems to fail for me (just running pytest test/srt/rotary_embedding/test_mrope.py -v), I needed to add dummy server args and also change to _foward_native. I'm probably misunderstanding something about how to run this test. We could write a test similar to the cpu/test_rope.py?

@raayandhar Your running command have no problem. The test_mrope was broken due to some recent PR merged. We are trying to fix it. Currently have no workaround. Sorry for making confusion.

Is there some timeline for the fix? I am also happy to try and help.

raayandhar · 2025-11-28T05:46:50Z

This PR can be re-reviewed once we merge #14082 and I add yarn scaling to the unit test.

raayandhar · 2025-11-28T07:44:47Z

Now that we have merged the fix, I adapted the test_mrope.py unit test to also test the yarn scaling:

UT test results (click to expand)

(sglang) ynez% pytest test/srt/rotary_embedding/test_mrope.py -v
=========================================================================== test session starts ============================================================================
platform linux -- Python 3.12.3, pytest-9.0.1, pluggy-1.6.0 -- /home/raayan/projects/sglang/.venv/bin/python3
cachedir: .pytest_cache
rootdir: /home/raayan/projects/sglang/test
configfile: pytest.ini
plugins: anyio-4.11.0
collected 48 items                                                                                                                                                         

test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                          [  2%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                         [  4%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                       [  6%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                          [  8%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                         [ 10%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype0-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                       [ 12%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                          [ 14%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                         [ 16%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                       [ 18%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                          [ 20%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                         [ 22%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-11-dtype1-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                       [ 25%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                        [ 27%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                       [ 29%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                     [ 31%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                        [ 33%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                       [ 35%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype0-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                     [ 37%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                        [ 39%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                       [ 41%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                     [ 43%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                        [ 45%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                       [ 47%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_default-8192-dtype1-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                     [ 50%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                             [ 52%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                            [ 54%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                          [ 56%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                             [ 58%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                            [ 60%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype0-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                          [ 62%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                             [ 64%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                            [ 66%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                          [ 68%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                             [ 70%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                            [ 72%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-11-dtype1-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                          [ 75%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                           [ 77%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                          [ 79%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                        [ 81%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                           [ 83%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                          [ 85%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype0-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                        [ 87%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-1-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                           [ 89%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-1-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                          [ 91%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-1-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                        [ 93%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-2-model_info0-Qwen/Qwen2-VL-7B-Instruct] PASSED                                           [ 95%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-2-model_info1-Qwen/Qwen2-VL-72B-Instruct] PASSED                                          [ 97%]
test/srt/rotary_embedding/test_mrope.py::test_mrope[mrope_yarn-8192-dtype1-2-model_info2-Qwen/Qwen2.5-VL-72B-Instruct] PASSED                                        [100%]

============================================================================= warnings summary =============================================================================
.venv/lib/python3.12/site-packages/_pytest/config/__init__.py:1397
  /home/raayan/projects/sglang/.venv/lib/python3.12/site-packages/_pytest/config/__init__.py:1397: PytestConfigWarning: Unknown config option: asyncio_mode
  
    self._warn_or_fail_if_strict(f"Unknown config option: {key}\n")

<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
===================================================================== 48 passed, 3 warnings in 53.33s ======================================================================
sys:1: DeprecationWarning: builtin type swigvarlink has no __module__ attribute
(sglang) ynez%

I think this PR is ready for review / CI now @yhyang201

cc @yuan-luo

yuan-luo · 2025-12-02T03:49:55Z

Please paste the lmms_eval result.

yhyang201 · 2025-12-02T09:01:30Z

/tag-and-rerun-ci

raayandhar · 2025-12-02T18:13:19Z

Please paste the lmms_eval result.

Running python test/srt/models/test_vlm_models.py --mem-fraction-static 0.7 (which runs lmms_eval) with --tp 4 and changing to batch size 1 (instead of 32) on Qwen2.5-VL-3B-Instruct with the following changes to it's rope_scaling config section:

    "rope_scaling": {
      "rope_type": "yarn", 
      "factor": 2.0,
      "original_max_position_embeddings": 256000,
      "mrope_section": [
        16,
        24,
        24
      ]
    },

introduced rope_type, factor and doubled original_max_position_embeddings.

We are passing with score of 0.4644, but you can expand below:

Test results (click to expand)

openai_compatible (model_version="./models/Qwen2.5-VL-3B-Instruct-yarn",tp=1), gen_kwargs: (), limit: None, num_fewshot: None, batch_size: 1
| Tasks  |Version|Filter|n-shot| Metric |   |Value |   |Stderr|
|--------|------:|------|-----:|--------|---|-----:|---|------|
|mmmu_val|      0|none  |     0|mmmu_acc|↑  |0.4644|±  |   N/A|

Result
: {'results': {'mmmu_val': {'alias': 'mmmu_val', 'mmmu_acc,none': 0.46444, 'mmmu_acc_stderr,none': 'N/A', 'mmmu_acc_pass_at_k,none': [], 'mmmu_acc_pass_at_k_stderr,none': [], 'submission,none': [], 'submission_stderr,none': []}}, 'group_subtasks': {'mmmu_val': []}, 'configs': {'mmmu_val': {'task': 'mmmu_val', 'dataset_path': 'lmms-lab/MMMU', 'test_split': 'validation', 'full_docs': False, 'process_results_use_image': False, 'doc_to_visual': '<function mmmu_doc_to_visual at 0x7ac92c32afc0>', 'doc_to_text': '<function mmmu_doc_to_text at 0x7ac92c32be20>', 'doc_to_target': 'answer', 'doc_to_messages': '<function mmmu_doc_to_messages at 0x7ac92c160ea0>', 'process_results': '<function mmmu_process_results at 0x7ac92c162020>', 'description': '', 'target_delimiter': ' ', 'fewshot_delimiter': '\n\n', 'num_fewshot': 0, 'metric_list': [{'metric': 'mmmu_acc', 'aggregation': '<function mmmu_aggregate_results at 0x7ac92c163240>', 'higher_is_better': True}], 'output_type': 'generate_until', 'generation_kwargs': {'max_new_tokens': 16, 'until': ['\n\n']}, 'repeats': 1, 'should_decontaminate': False, 'metadata': {'version': 0.0, 'interleaved_format': False}, 'lmms_eval_specific_kwargs': {'default': {'prompt_type': 'format', 'multiple_choice_prompt': "Answer with the option's letter from the given choices directly.", 'open_ended_prompt': 'Answer the question using a single word or phrase.'}, 'prompt_type': 'format', 'multiple_choice_prompt': "Answer with the option's letter from the given choices directly.", 'open_ended_prompt': 'Answer the question using a single word or phrase.'}}}, 'versions': {'mmmu_val': 0.0}, 'n-shot': {'mmmu_val': 0}, 'higher_is_better': {'mmmu_val': {'mmmu_acc': True}}, 'n-samples': {'mmmu_val': {'original': 900, 'effective': 900}}, 'config': {'model': 'openai_compatible', 'model_args': 'model_version="./models/Qwen2.5-VL-3B-Instruct-yarn",tp=1', 'batch_size': '1', 'batch_sizes': [], 'device': None, 'use_cache': None, 'limit': None, 'bootstrap_iters': 100000, 'gen_kwargs': '', 'random_seed': 0, 'numpy_seed': 1234, 'torch_seed': 1234, 'fewshot_seed': 1234}, 'git_hash': 'c55de164b', 'date': '20251203_020157', 'task_hashes': {'mmmu_val': '614600386b06b53646ff0656be64e14e0f33e49b2e8eb7e528d482ff0ba6e7ae'}, 'model_source': 'openai_compatible', 'model_name': '"./models/Qwen2.5-VL-3B-Instruct-yarn"', 'model_name_sanitized': 'models__Qwen2.5-VL-3B-Instruct-yarn__', 'system_instruction': None, 'system_instruction_sha': None, 'fewshot_as_multiturn': False, 'chat_template': None, 'chat_template_sha': None, 'start_time': 16373248.667930858, 'end_time': 16373599.762910033, 'total_evaluation_time_seconds': '351.09497917443514'}
Model ./models/Qwen2.5-VL-3B-Instruct-yarn achieved accuracy: 0.4644
Cleaning up process 3946576
/usr/lib/python3.12/subprocess.py:1127: ResourceWarning: subprocess 3946576 is still running
  _warn("subprocess %s is still running" % self.pid,
ResourceWarning: Enable tracemalloc to get the object allocation traceback
.
----------------------------------------------------------------------
Ran 1 test in 394.662s

OK

raayandhar · 2025-12-02T19:38:30Z

Caught a small extra print, but before that nearly all the CI was passing, only one that failed was not related to my changes.

raayandhar · 2025-12-09T05:02:37Z

@yhyang201 could you take another look?

yhyang201 · 2025-12-19T05:53:05Z

/rerun-failed-ci

raayandhar · 2025-12-20T21:31:47Z

@yhyang201 could you take a look at the CI and let me know what you think? The failures seem unrelated.

yuan-luo · 2026-01-03T02:06:40Z

Could you fix the VLM CI? Thanks.
https://github.com/sgl-project/sglang/actions/runs/20534806282/job/58991668621?pr=13313

Signed-off-by: Raayan Dhar raayan.dhar@gmail.com <raayan.dhar@gmail.com>

raayandhar · 2026-01-05T19:33:55Z

Could you fix the VLM CI? Thanks. https://github.com/sgl-project/sglang/actions/runs/20534806282/job/58991668621?pr=13313

I don't think the CI is related to my changes. I just tested on main and got the same failure (0.3833 below 0.4):

Test results (click to expand)

======================================================================
ERROR: test_vlm_mmmu_benchmark (__main__.TestVLMModels.test_vlm_mmmu_benchmark)
Test VLM models against MMMU benchmark.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/raayan/projects/sglang/python/sglang/test/kits/mmmu_vlm_kit.py", line 311, in _run_vlm_mmmu_test
    self.assertGreaterEqual(
AssertionError: 0.38333 not greater than or equal to 0.4 : Model Qwen/Qwen2.5-VL-3B-Instruct accuracy (0.3833) below expected threshold (0.4000)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/raayan/projects/sglang/python/sglang/srt/utils/common.py", line 2508, in retry
    return fn()
           ^^^^
  File "/home/raayan/projects/sglang/python/sglang/test/test_utils.py", line 1722, in <lambda>
    lambda: super(CustomTestCase, self)._callTestMethod(method),
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Test failed for Qwen/Qwen2.5-VL-3B-Instruct: 0.38333 not greater than or equal to 0.4 : Model Qwen/Qwen2.5-VL-3B-Instruct accuracy (0.3833) below expected threshold (0.4000)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/raayan/projects/sglang/python/sglang/test/test_utils.py", line 1721, in _callTestMethod
    retry(
  File "/home/raayan/projects/sglang/python/sglang/srt/utils/common.py", line 2516, in retry
    raise Exception(f"retry() exceed maximum number of retries.")
Exception: retry() exceed maximum number of retries.

----------------------------------------------------------------------
Ran 1 test in 440.477s

FAILED (errors=1)
(sglang) raayan@uril-1:~/projects/sglang$ git status
On branch main
Your branch is up to date with 'origin/main'.

Also, the model config does not use YaRN scaling, so it should not hit my path anyways. The root cause of this CI failing should be separate from my changes.

raayandhar · 2026-01-06T01:37:46Z

/rerun-failed-ci

yuan-luo · 2026-01-06T16:37:04Z

Could you paste a model leveraging this flag and the acc gets no drop? This is a critical PR, we are very cautious to merge. Thank you very much.

raayandhar · 2026-01-06T16:42:09Z

Could you paste a model leveraging this flag and the acc gets no drop? This is a critical PR, we are very cautious to merge. Thank you very much.

Sure, I can do so later today. Actually earlier I also pasted the lmm_evals result as well for a model using this flag. Is that sufficient or were you thinking of something else?

#13313 (comment)

yuan-luo · 2026-01-14T02:50:59Z

/rerun-failed-ci

yuan-luo · 2026-01-14T02:51:21Z

Let's have all the GPU CI passed.

yuan-luo · 2026-01-14T02:53:10Z

Overall, LGTM.

yuan-luo · 2026-01-15T02:24:27Z

/rerun-failed-ci

yuan-luo · 2026-01-19T01:56:28Z

/rerun-failed-ci

yuan-luo · 2026-01-23T05:59:13Z

/rerun-failed-ci

yuan-luo · 2026-01-25T05:58:22Z

/rerun-failed-ci

yhyang201 · 2026-01-26T04:31:01Z

/rerun-failed-ci

Signed-off-by: raayandhar <raayan.dhar@gmail.com>

yuan-luo · 2026-02-04T07:55:09Z

/rerun-failed-ci

…3313) Signed-off-by: Raayan Dhar raayan.dhar@gmail.com <raayan.dhar@gmail.com> Signed-off-by: raayandhar <raayan.dhar@gmail.com>

raayandhar force-pushed the mrope_section-with-yarn branch from 49312c4 to 3c28a31 Compare November 15, 2025 02:09

yuan-luo reviewed Nov 15, 2025

View reviewed changes

python/sglang/srt/layers/rotary_embedding.py Show resolved Hide resolved

raayandhar force-pushed the mrope_section-with-yarn branch from 3c28a31 to 7dc040c Compare November 15, 2025 21:24

raayandhar marked this pull request as ready for review November 15, 2025 23:09

raayandhar requested review from BBuf, Edwardf0t1, Fridge003, HaiShaw, Ying1123, ch-wan, ispobock, kushanam and merrymercy as code owners November 15, 2025 23:09

yhyang201 self-assigned this Nov 18, 2025

raayandhar force-pushed the mrope_section-with-yarn branch from 1e3c382 to 5e41c35 Compare November 28, 2025 03:41

raayandhar force-pushed the mrope_section-with-yarn branch from 5e41c35 to 7cbf330 Compare November 28, 2025 07:34

github-actions bot added the run-ci label Dec 2, 2025

raayandhar requested a review from yuan-luo December 6, 2025 20:49

Merge branch 'main' into mrope_section-with-yarn

19bcb5d

raayandhar added 2 commits January 5, 2026 10:28

Merge branch 'main' into mrope_section-with-yarn

2672491

fix unit test

6908180

Signed-off-by: Raayan Dhar raayan.dhar@gmail.com <raayan.dhar@gmail.com>

yuan-luo approved these changes Jan 14, 2026

View reviewed changes

Merge branch 'main' into mrope_section-with-yarn

5495481

raayandhar added 2 commits January 19, 2026 15:52

Merge branch 'main' into mrope_section-with-yarn

fd21f1a

Merge branch 'main' into mrope_section-with-yarn

d8600d7

raayandhar and others added 3 commits January 30, 2026 15:09

merge

7b2162a

Signed-off-by: raayandhar <raayan.dhar@gmail.com>

Merge branch 'main' into mrope_section-with-yarn

9f4f63e

Merge branch 'main' into mrope_section-with-yarn

6e5a890

yhyang201 approved these changes Feb 8, 2026

View reviewed changes

ispobock merged commit 92cdd39 into sgl-project:main Feb 14, 2026
180 of 192 checks passed

Conversation

raayandhar commented Nov 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Uh oh!

Uh oh!

raayandhar commented Nov 15, 2025

Uh oh!

raayandhar commented Nov 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

yuan-luo commented Nov 18, 2025

Uh oh!

raayandhar commented Nov 18, 2025

Uh oh!

raayandhar commented Nov 21, 2025

Uh oh!

raayandhar commented Nov 28, 2025

Uh oh!

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

Uh oh!

Uh oh!

yuan-luo commented Dec 2, 2025

Uh oh!

yhyang201 commented Dec 2, 2025

Uh oh!

raayandhar commented Dec 2, 2025

Uh oh!

raayandhar commented Dec 2, 2025

Uh oh!

raayandhar commented Dec 9, 2025

Uh oh!

yhyang201 commented Dec 19, 2025

Uh oh!

raayandhar commented Dec 20, 2025

Uh oh!

yuan-luo commented Jan 3, 2026

Uh oh!

raayandhar commented Jan 5, 2026

Uh oh!

raayandhar commented Jan 6, 2026

Uh oh!

yuan-luo commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

raayandhar commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

yuan-luo commented Jan 14, 2026

Uh oh!

yuan-luo commented Jan 14, 2026

Uh oh!

yuan-luo commented Jan 14, 2026

Uh oh!

yuan-luo commented Jan 15, 2026

Uh oh!

yuan-luo commented Jan 19, 2026

Uh oh!

yuan-luo commented Jan 23, 2026

Uh oh!

yuan-luo commented Jan 25, 2026

Uh oh!

yhyang201 commented Jan 26, 2026

Uh oh!

yuan-luo commented Feb 4, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

raayandhar commented Nov 15, 2025 •

edited

Loading

raayandhar commented Nov 16, 2025 •

edited

Loading

raayandhar commented Nov 28, 2025 •

edited

Loading

yuan-luo commented Jan 6, 2026 •

edited

Loading

raayandhar commented Jan 6, 2026 •

edited

Loading