Skip to content

feat: Add option to pass QAICInferenceSession to TextGeneration #356

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.

Sign up for GitHub

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

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
+13 −4
Draft

feat: Add option to pass QAICInferenceSession to TextGeneration #356

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 13 additions & 4 deletions QEfficient/generation/text_generation_inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,6 +322,8 @@ def cloud_ai_100_exec_kv(
automation=False,
prompt_to_lora_id_mapping: Optional[List[int]] = None,
is_tlm: bool = False,
session: Optional[QAICInferenceSession] = None,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we add this same param in latency_stats_bertstyle function as well so we don't need to create new session there also.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Will add

print_latency_stats: bool = True,
):
"""
This method generates output until ``eos`` or ``generation_len`` by executing the compiled ``qpc`` on ``Cloud AI 100`` Hardware cards.
Expand All @@ -342,6 +344,8 @@ def cloud_ai_100_exec_kv(
:Write_io_dir (str): Path to write the input and output files. ``Defaults to None``.
:automation (bool): If true, it prints input, output, and performance stats. ``Defaults to False``.
:prompt_to_lora_id_mapping (List[int]): Mapping to associate prompts with their respective LoRA adapter.
:session (QAICInferenceSession): Pre-initialized QAICInferenceSession object. ``Defaults to None``.
:print_latency_stats (bool): If True, it prints latency statistics. ``Defaults to True``.

Returns:
:CloudAI100ExecInfo: Object holding execution output and performance details.
Expand Down Expand Up @@ -372,6 +376,7 @@ def cloud_ai_100_exec_kv(
write_io_dir=write_io_dir,
full_batch_size=full_batch_size,
is_tlm=is_tlm,
session=session,
)
if full_batch_size is None:
exec_info = [
Expand All @@ -395,8 +400,8 @@ def cloud_ai_100_exec_kv(
exec_info = generate_text.generate(
prompt=prompt, generation_len=generation_len, prompt_to_lora_id_mapping=prompt_to_lora_id_mapping
)

print_latency_stats_kv(prompt, exec_info=exec_info, automation=automation)
if print_latency_stats:
print_latency_stats_kv(prompt, exec_info=exec_info, automation=automation)
return exec_info


Expand All @@ -411,13 +416,16 @@ def __init__(
enable_debug_logs: bool = False,
write_io_dir: Optional[str] = None,
is_tlm: Optional[int] = None,
session: Optional[QAICInferenceSession] = None,
) -> None:
self._ctx_len = ctx_len
self._write_io_dir = write_io_dir
self.is_tlm = is_tlm

# Load QPC
self._session = QAICInferenceSession(qpc_path, device_id, enable_debug_logs=enable_debug_logs)
self._session = (
session if session else QAICInferenceSession(qpc_path, device_id, enable_debug_logs=enable_debug_logs)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we store this session somewhere while creating it, so we fetch it from the path directly, somewhat like we do for qpc?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand, session is an object of QAICInferenceSession. I need to pass it as a parameter. How can I get it from a path?

Copy link
Contributor

@asmigosw asmigosw May 7, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean user would not be able to pass this object, so make we make use of object created this time and not have it as a param. Maybe we can make QAICInferenceSession like a singleton class something like below:

class QAICInferenceSession:
    session = None

    def __new__(cls, *args, **kwargs):
        if not cls.session:
            cls.session= super(QAICInferenceSession, cls).__new__(cls)
        return `cls.session

)

# Fetch the variables from the QPC
self._vocab_size = self._fetch_vocab_size() # Fetch Vocab size
Expand Down Expand Up @@ -905,9 +913,10 @@ def __init__(
enable_debug_logs: bool = False,
write_io_dir: Optional[str] = None,
is_tlm: bool = False,
session: Optional[QAICInferenceSession] = None,
) -> None:
self._qaic_model = QEffTextGenerationBase(
tokenizer, qpc_path, full_batch_size, ctx_len, device_id, enable_debug_logs, write_io_dir, is_tlm
tokenizer, qpc_path, full_batch_size, ctx_len, device_id, enable_debug_logs, write_io_dir, is_tlm, session
)
self._full_batch_size = self._qaic_model.full_batch_size
self._tokenizer = self._qaic_model.tokenizer
Expand Down
Loading