-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MicroServing Implementation #3064
base: main
Are you sure you want to change the base?
Conversation
engine_config, // | ||
model_configs, // | ||
trace_recorder), | ||
// EngineAction::BatchJumpForward(models, tokenizer, trace_recorder), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not sure whether we can delete batch jumpforward here.
@@ -51,7 +51,8 @@ def extract_creation_args(func: relax.Function) -> Dict[str, Any]: | |||
"rope_scaling": json.loads(args[10].value), | |||
"rope_ext_factors": args[11], | |||
"rotary_dim": args[12].value.value, | |||
"dtype": args[13].value, | |||
"enable_disaggregation": bool(args[13].value.value), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is it safe to remove enable_disaggregation
? The argument is not propagated after this pass
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No, this one is necessary. It's part of kwargs
and will be passed along
mlc-llm/python/mlc_llm/compiler_pass/dispatch_kv_cache_creation.py
Lines 103 to 108 in fffb785
kwargs = extract_creation_args(creation_func) | |
self.attach_kv_cache_metadata(kwargs) | |
bb = relax.BlockBuilder(new_mod) | |
self.create_tir_paged_kv_cache(bb, kwargs) | |
self.create_flashinfer_paged_kv_cache(bb, kwargs) |
python/mlc_llm/interface/router.py
Outdated
return error_protocol.create_error_response( | ||
HTTPStatus.BAD_REQUEST, message="The request has disconnected" | ||
) | ||
# TODO(Charlie): This is copied from engine.py -- why is it here? Non-streaming only has a single chunk right? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this question is not answered.
cc: @MasterJH5574 |
This PR introduces MicroServing API. MicroServing introduces simple yet effective REST APIs to support fine-grained sub-request level actions. A programmable router transforms user requests into sub-request calls, lifting fine-grained scheduling to the API level, thus enabling the dynamic reconfiguration of different orchestration patterns.
Co-authored-by: Ruihang Lai [email protected]
Co-authored-by: Charlie Ruan [email protected]
Co-authored-by: Yingcheng Wang [email protected]