[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights logic and bump SGLang to 0.4.9.post6 by hebiao064 · Pull Request #2720 · verl-project/verl

hebiao064 · 2025-07-24T00:53:52Z

Co-authored-by: CuiBo 82354186+SuperCB@users.noreply.github.com
Co-authored-by: GeLee 865038696@qq.com
Co-authored-by: 杨睿 yangruipis@163.com

What does this PR do?

Context

Builds on the earlier work from #2620 to further simplify and unify the update weights logic from @SuperCB and @GeLee-Q

Motivation

We've received numerous complaints that the update weights logic in VERL SGLang is difficult to understand and error-prone during development. For example, a recent refactor from VerlEngine to AsyncEngine accidentally broke the torch reduce patch logic, leading to a 3× slowdown in weight updates.

Additionally, when changing the tensor update logic from per-tensor to batched updates, we had to duplicate the logic across both Megatron SGLang Sharding Manager and FSDP SGLang Sharding Manager. This not only introduced unnecessary complexity, but also made maintenance and future improvements harder.

To address these issues, we propose simplifying and centralizing the update weights logic within core SGLang, exposing a unified API that is transparent to all RL engines. This abstraction reduces duplication, improves reliability, and accelerates future development.

Benefits

✅ Unified and transparent update weights API for all RL engines (FSDP, Megatron, etc.)
🚀 Speedup QWen3 MoE weight updates from 110s to 30s
🔧 Torch patch MP serializer to avoid inefficient P2P communication that slows down weight updates

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

Checklist Before Starting

Search for similar PRs. Paste at least one query link here: ...
Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
- {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
- If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
- {type} is in feat, fix, refactor, chore, test
- If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
- Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

Read the Contribute Guide.
Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
Add / Update the documentation.
Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

gemini-code-assist

Code Review

This pull request significantly simplifies the weight update logic in both FSDP and Megatron sharding managers by leveraging a new sgl_update_weights utility. The refactoring greatly improves code readability and maintainability by abstracting away complex distributed communication logic.

I've found one critical issue: a likely copy-paste error in the MegatronSGLangShardingManager where an incorrect device_mesh_key is used, which would lead to a runtime error. Other than that, the changes look solid and are a great improvement.

zhaochenyang20 · 2025-07-24T22:30:25Z

Nice work!

zhaochenyang20 · 2025-07-31T23:20:57Z

long river have crossed 🙏

vermouth1992 · 2025-08-01T11:09:35Z

Qwen2VL has bugs in transformers 4.53. We have to upgrade to 4.54 to use QWen2VL. We can ignore fsdp problem for now

hebiao064 · 2025-08-01T16:57:47Z

Qwen2VL has bugs in transformers 4.53. We have to upgrade to 4.54 to use QWen2VL. We can ignore fsdp problem for now

Would that also require vllm + transformer upgrade then?

I think we can follow ziyuan's suggestion #2720 (comment) for this PR and upgrade transformer/vllm in separate PR

hebiao064 · 2025-08-01T17:34:58Z

#2780 (comment) @vermouth1992

I fixed it in a separate PR: #2869

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

#2869) ### What does this PR do? Do not merge before: #2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

verl-project#2869) ### What does this PR do? Do not merge before: verl-project#2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

verl-project#2869) ### What does this PR do? Do not merge before: verl-project#2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

…s (#2869) ### What does this PR do? Do not merge before: verl-project/verl#2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

verl-project#2869) ### What does this PR do? Do not merge before: verl-project#2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

verl-project#2869) ### What does this PR do? Do not merge before: verl-project#2720 > Add **concise** overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review. Bump xformers, fixing patched model code ### Checklist Before Starting - [x] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

Simplify Update weights api with SGLang changes

ebf8cc7

hebiao064 requested review from SwordFaith, chenhaiq and zhaochenyang20 as code owners July 24, 2025 00:53

gemini-code-assist Bot reviewed Jul 24, 2025

View reviewed changes

Comment thread verl/workers/sharding_manager/megatron_sglang.py Outdated

Merge branch 'main' into bhe/refactor_update_weights

48843f3

hebiao064 changed the title ~~[fsdp,megatron,sglang] feat: [Do Not Merge] Simplify Update weights api with SGLang changes~~ [Do Not Merge] [fsdp,megatron,sglang] feat: Simplify Update weights api with SGLang changes Jul 24, 2025

fix

142479c

zhaochenyang20 mentioned this pull request Jul 25, 2025

RL SGLang Update Weight in Tensor Slow for MOE zhaochenyang20/Awesome-ML-SYS-Tutorial#169

Closed

hebiao064 added 2 commits July 31, 2025 13:06

Merge branch 'volcengine:main' into bhe/refactor_update_weights

8a916d8

Update image in CI

e5ca652

hebiao064 requested a review from eric-haibin-lin as a code owner July 31, 2025 21:55

hebiao064 changed the title ~~[Do Not Merge] [fsdp,megatron,sglang] feat: Simplify Update weights api with SGLang changes~~ [fsdp,megatron,sglang] feat: Simplify Update weights api with SGLang changes and bump SGLang to 0.4.9.post6 Jul 31, 2025

fix ci

f66eeb1

zhaochenyang20 mentioned this pull request Jul 31, 2025

[fsdp, megatron, sglang] fix: Fixed a bug in the update_weight process where the GPU ID was being passed incorrectly. #2620

Closed

7 tasks

hebiao064 mentioned this pull request Jul 31, 2025

[Feature] Performance boost for VerlEngine's update_weights_from_tensor sgl-project/sglang#6762

Closed

2 tasks

ETOgaosion reviewed Aug 1, 2025

View reviewed changes

Comment thread docker/verl0.5-cu126-torch2.7-fa2.7.4/Dockerfile.app.sglang0.4.9.post6.mcore0.12.app Outdated

hebiao064 added 2 commits August 1, 2025 03:16

fix

9ca3e8d

fix

d77bced

hebiao064 changed the title ~~[fsdp,megatron,sglang] feat: Simplify Update weights api with SGLang changes and bump SGLang to 0.4.9.post6~~ [fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights logic and bump SGLang to 0.4.9.post6 Aug 1, 2025

hebiao064 added 2 commits August 1, 2025 04:17

fix

3f18a92

fix

0263e0f

zhaochenyang20 approved these changes Aug 1, 2025

View reviewed changes

fix

a7b5bfc

fix

bc05aa3

hebiao064 added 2 commits August 1, 2025 19:33

fix

03ebe85

fix

e521bff

hebiao064 mentioned this pull request Aug 1, 2025

[sglang] chore: bump transformer formers 4.54.0 and fix QWen VL issues #2869

Merged

7 tasks

zhaochenyang20 merged commit 53f9b2b into verl-project:main Aug 2, 2025
49 of 51 checks passed

yellowbee686 pushed a commit to yellowbee686/verl that referenced this pull request Aug 2, 2025

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

d639c96

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

Juniper1021 pushed a commit to Juniper1021/verl that referenced this pull request Aug 7, 2025

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

5a10445

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

hebiao064 mentioned this pull request Aug 24, 2025

sglang is 2.x slower than vllm in param sync #3173

Open

whatadayG pushed a commit to whatadayG/verl that referenced this pull request Sep 5, 2025

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

9a7da92

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

chenjiaoAngel added a commit to chenjiaoAngel/verl that referenced this pull request Nov 14, 2025

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

4fc1f3c

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

TimurTaepov pushed a commit to giorgossideris/verl that referenced this pull request Dec 20, 2025

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

ed977ee

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

vyomakesh0728 added a commit to vyomakesh0728/verl that referenced this pull request Jan 22, 2026

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights l…

1257697

…ogic and bump SGLang to 0.4.9.post6 (verl-project#2720)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights logic and bump SGLang to 0.4.9.post6#2720

[fsdp,megatron,sglang] feat: Accelerate and Simplify Update weights logic and bump SGLang to 0.4.9.post6#2720
zhaochenyang20 merged 14 commits intoverl-project:mainfrom
hebiao064:bhe/refactor_update_weights

hebiao064 commented Jul 24, 2025 •

edited

Loading

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

zhaochenyang20 commented Jul 24, 2025

Uh oh!

zhaochenyang20 commented Jul 31, 2025

Uh oh!

Uh oh!

vermouth1992 commented Aug 1, 2025

Uh oh!

hebiao064 commented Aug 1, 2025

Uh oh!

hebiao064 commented Aug 1, 2025 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

hebiao064 commented Jul 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Context

Motivation

Benefits

Checklist Before Starting

Test

API and Usage Example

Design & Code Changes

Checklist Before Submitting

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

zhaochenyang20 commented Jul 24, 2025

Uh oh!

zhaochenyang20 commented Jul 31, 2025

Uh oh!

Uh oh!

vermouth1992 commented Aug 1, 2025

Uh oh!

hebiao064 commented Aug 1, 2025

Uh oh!

hebiao064 commented Aug 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

hebiao064 commented Jul 24, 2025 •

edited

Loading

hebiao064 commented Aug 1, 2025 •

edited

Loading