⚡ Bolt: Remove HashMap allocation in repetition penalty

[EffortlessSteven]

💡 What

Replaced the HashMap<usize, usize> used for tracking repetition counts in SamplingStrategy with a pre-allocated chronological Vec<u32>. Changed apply_repetition_penalty to use this history slice directly instead of dynamically generating a new vector using flat_map and repeat_n on every token.

🎯 Why

Token generation is the tightest hot loop in LLM inference. Allocating new vectors and performing HashMap lookups/updates for every sampled token introduces significant overhead. The underlying mathematics of bitnet_logits::apply_repetition_penalty apply the exponential penalty for every token present in the history slice, so passing a chronologically accurate Vec<u32> has identical mathematical behavior without requiring HashMap accounting or per-token vector materialization.

📊 Impact

• Eliminates one HashMap insertion per generated token.
• Eliminates one vocabulary-sized Vec<u32> allocation and expansion loop per generated token.
• Amortizes cleanup cost by using a 2000-token threshold sliding window (drain), improving semantic stability over the previous hard clear() approach.

🔬 Measurement

Verified the logic using existing tests, which passed immediately after applying changes.
Run: cargo test -p bitnet-inference --lib generation::sampling

---

PR created automatically by Jules for task 9867502972988120625 started by @EffortlessSteven

[google-labs-jules]

👋 Jules, reporting for duty! I'm here to lend a hand with this pull request.

When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down.

I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job!

For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with @jules. You can find this option in the Pull Request section of your global Jules UI settings. You can always switch back!

New to Jules? Learn more at jules.google/docs.

---

For security, I will only act on instructions from the user who triggered this task.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 19b50933-ed8b-4d8b-8a9a-043ef7573cca

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch bolt/remove-hashmap-allocation-9867502972988120625

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}