Fix(MInference): fix the search pattern feature

[Huiqiang Jiang (iofu728)]

What does this PR do?

Fixes #110 resupport the search pattern feature.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Was this discussed/approved via a Github issue? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

Huiqiang Jiang (@iofu728)

[Copilot]

Pull Request Overview

This PR restores the “search pattern” workflow by introducing an is_search mode, wiring through new configuration flags, and adjusting model patching logic to skip loading existing patterns when searching.

• Remove accelerate from core requirements.
• Add is_search, starting_layer, and config_path to attention kwargs and propagate them through forward passes.
• Adjust model patching to bypass loading best patterns in search mode and update experiment scripts and deps.

Reviewed Changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
setup.py	Drop accelerate from INSTALL_REQUIRES.
minference/modules/minference_forward.py	Implement search logic (is_search flag), load/save configs.
minference/modules/forward.py	Pass num_hidden_layers into prefill_kwargs.
minference/models_patch.py	Skip loading best_pattern when in search mode.
minference/minference_configuration.py	Embed is_search, starting_layer, and config_path into attn_kwargs.
experiments/infinite_bench/run_infinitebench.py	Standardize _attn_implementation usage.
experiments/infinite_bench/requirements.txt	Add evaluation dependencies (jieba, nltk, rouge, etc.).

Comments suppressed due to low confidence (4)

setup.py:37

• [nitpick] Removing accelerate from runtime requirements may affect users; please update the README or installation docs to explain whether accelerate is now optional or required via extras.

-    "accelerate",

minference/modules/minference_forward.py:636

• Introduce a new is_search code path; add tests to cover both is_search=True (including config file creation and the error when the search is complete) and is_search=False to ensure backward compatibility.

+    is_search = prefill_kwargs["attn_forward_config"].get("is_search", False)

minference/minference_configuration.py:72

• [nitpick] The new attn_kwargs now include is_search, starting_layer, and config_path; please update the class docstring or configuration reference to explain these parameters and their effects.

+        self.attn_kwargs = {

minference/models_patch.py:85

• self.config does not define an is_search attribute; this should reference self.config.attn_kwargs["is_search"] (or add self.is_search = is_search in the config class) to avoid an AttributeError.

if not self.config.is_search:

[Copilot]

[nitpick] Using print in a library module can be noisy and hard to control; consider using a logger (e.g., Python's logging module) at DEBUG or INFO level.

Suggested change

	print(f"--- Searching layer {layer_idx} ---")
	logging.info(f"--- Searching layer {layer_idx} ---")