(sync): sync codes to github. - #155

Merged

PanAndy merged 19 commits into

mainfrom

sync/sync_to_github_0826

Aug 26, 2025

Commits on Aug 26, 2025

(feat): support pp + tie embedding.

liu-zichen
authored and
PanAndy
committed
3af6bb8
(fix): Skip current step when final_response_mask.sum() == 0.

ToasterSC
authored and
PanAndy
committed
7361501
(fix): addressed all issues where the response mask sum was zero, both at the batch level and within individual domains.

ToasterSC
authored and
PanAndy
committed
02261f0
(fix): fix mcore val.
PanAndy
committed
a3ab7e3
(feat): add rlvr rollout pipeline for just val.
PanAndy
committed
bd8e054
(fix): Fix VLM collator empty multi_modal_data.

guoshengCS
authored and
PanAndy
committed
e3df1e0
(fix): remove fp8 quantization patch in vLLM 0.10.0.

hydrozhao
authored and
PanAndy
committed
022d782
(docs): Add docs for vlm rlvr pipeline.

guoshengCS
authored and
PanAndy
committed
a61bcc7
(fix): set RAY_get_check_signal_interval_milliseconds to 1 to avoid queueing of rpc in Ray.

hydrozhao
authored and
PanAndy
committed
075678a
(fix): Fix the issue where DataProto could not collect metrics from all ranks.

Schnabel-8
authored and
PanAndy
committed
564d8e4
(fix): Fix math timeout and code extract.

sydney170
authored and
PanAndy
committed
b93f34c
(feat): support deepspeed ulysses.

chocoded
authored and
PanAndy
committed
cc7aa99
(fix): fix vllm 0.10.0.

hydrozhao
authored and
PanAndy
committed
d88574b
(docs): Add documentation for the DPO pipeline.

wenxuan.jwx
authored and
PanAndy
committed
d0a3f42
(feat): Optimize redundant imports and adjust timeout.

sydney170
authored and
PanAndy
committed
303a639
(fix): llm_as_judge with vllm strategy should init cuda before torch.cuda.reset_max_memory_allocated.

HuangJoJo
authored and
PanAndy
committed
4f92512
(docs): update readme taoSR1.
PanAndy
committed
e747fff
(fix): fix rlvr reward num_nodes infer.
PanAndy
committed
0f9b841
(fix): fix val setting.
PanAndy
committed
5420fa4

Navigation Menu

Uh oh!

(sync): sync codes to github. - #155

(sync): sync codes to github.#155
PanAndy merged 19 commits into
mainfrom
sync/sync_to_github_0826

Commits on Aug 26, 2025

(feat): support pp + tie embedding.

(fix): Skip current step when final_response_mask.sum() == 0.

(fix): addressed all issues where the response mask sum was zero, both at the batch level and within individual domains.

(fix): fix mcore val.

(feat): add rlvr rollout pipeline for just val.

(fix): Fix VLM collator empty multi_modal_data.

(fix): remove fp8 quantization patch in vLLM 0.10.0.

(docs): Add docs for vlm rlvr pipeline.

(fix): set RAY_get_check_signal_interval_milliseconds to 1 to avoid queueing of rpc in Ray.

(fix): Fix the issue where DataProto could not collect metrics from all ranks.

(fix): Fix math timeout and code extract.

(feat): support deepspeed ulysses.

(fix): fix vllm 0.10.0.

(docs): Add documentation for the DPO pipeline.

(feat): Optimize redundant imports and adjust timeout.

(fix): llm_as_judge with vllm strategy should init cuda before torch.cuda.reset_max_memory_allocated.

(docs): update readme taoSR1.

(fix): fix rlvr reward num_nodes infer.

(fix): fix val setting.