馃殌 The feature, motivation and pitch
pcp and dcp seem to share similar technical approaches. Both of them are part of the context parallelism share single cp world. Could we add relevant support for pcp as well through a similar technical approach like #40996 ? It may helpful for dsv4/glm5.2/qwen3.6 and other new model's pcp support
Alternatives
No response
Additional context
No response
Before submitting a new issue...
馃殌 The feature, motivation and pitch
pcp and dcp seem to share similar technical approaches. Both of them are part of the context parallelism share single cp world. Could we add relevant support for pcp as well through a similar technical approach like #40996 ? It may helpful for dsv4/glm5.2/qwen3.6 and other new model's pcp support
Alternatives
No response
Additional context
No response
Before submitting a new issue...