fix(exceptions): narrow broad Exception clauses to specific types

[sowndappan5]

Summary

This PR refactors various overly broad except Exception: clauses across the codebase to catch specific, anticipated exception types (such as ImportError, AttributeError, OSError, ValueError, etc.).

This prevents the accidental swallowing of critical execution/control flow exceptions like KeyboardInterrupt and SystemExit, aligning with pythonic best practices and improving general debuggability.

Addresses #8093.

Details of Changes

• Core Utilities:
  • deepspeed/utils/numa.py: Catch (FileNotFoundError, subprocess.SubprocessError) when checking for numactl command availability.
• Runtime & Engine:
  • deepspeed/runtime/engine.py: Catch OSError during directory creation, and catch BaseException (and re-raise) during module compilation to ensure hooks cleanup properly under all conditions.
  • deepspeed/runtime/hybrid_engine.py: Catch (ImportError, AttributeError) when querying transformers.
  • deepspeed/runtime/config.py: Catch (RuntimeError, AssertionError, AttributeError) when querying torch distributed rank and world size.
• NVMe & Memory/Storage:
  • deepspeed/nvme/parse_nvme_stats.py: Catch (ValueError, IndexError) during string parsing.
• Module Injection & Policy Containers:
  • deepspeed/module_inject/auto_tp.py: Catch (RuntimeError, ValueError, TypeError) during tensor operations.
  • deepspeed/module_inject/load_checkpoint.py & policy containers (gpt2, gptneo, gptj, llama, llama2, opt, distil_bert, clip, bert): Catch (ImportError, AttributeError) or ImportError on imports of transformers, fairscale, and llama.
• Mixture of Experts (MoE) & Auto-EP:
  • deepspeed/moe/sharded_moe.py: Catch ImportError when importing tutel.
  • deepspeed/module_inject/auto_ep_presets/base.py: Catch ImportError when importing transformers.utils.output_capturing.
• Launcher & Autotuner:
  • deepspeed/launcher/launch.py: Catch OSError when killing the subprocess tree.
  • deepspeed/autotuning/autotuner.py: Catch OSError when creating autotuning experiments/results directories.
• Inference & Compilation:
  • deepspeed/inference/v2/inference_parameter.py: Catch (RuntimeError, TypeError, IndexError) when casting device/parameters.
  • deepspeed/compile/input_storage.py: Catch (RuntimeError, AttributeError) when retrieving tensor memory format.

Verification

• Verified code changes maintain neighboring formatting rules (YAPF/flake8).
• Commits are signed off (--signoff).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: be8a4a1a57

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".