[None][feat] Add DeepSeek-V4-Pro curated configs

[lfr-0531]

Description

This PR adds curated inference configs for DeepSeek-V4-Pro (DeepseekV4ForCausalLM) under examples/configs/curated/, following the same convention as the existing DeepSeek-R1 curated configs (see #11653).

Two scenarios are provided, both tuned on B200 with TP=8 / EP=8 and MTP speculative decoding:

• deepseek-v4-pro-latency.yaml — Min Latency (max_batch_size=128, MTP max_draft_len=3, attention DP disabled, wide CUDA graph batch-size sweep).
• deepseek-v4-pro-throughput.yaml — Max Throughput (max_batch_size=32, MTP max_draft_len=1, attention DP + balance enabled, LM-head TP in ADP).

Both are registered in examples/configs/curated/lookup.yaml with arch: DeepseekV4ForCausalLM, model deepseek-ai/DeepSeek-V4-Pro, and gpu_compatibility: "B200".

Test Coverage

Covered by the existing curated-config validation suite in tests/unittest/llmapi/test_config_database.py, which loads every entry in curated/lookup.yaml, validates it against LlmArgs, runs a trtllm-serve sanity check, and asserts no unnecessary default values are present. The two new entries are picked up automatically via the arch field.

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (config validation runs automatically over curated/lookup.yaml).

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. (No API changes — configs only.)

• Any new dependencies have been scanned for license and vulnerabilities. (None.)

• CODEOWNERS updated if ownership changes. (N/A.)

• Documentation updated as needed.

• Update tava architecture diagram if there is a significant design change in PR. (N/A.)

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds two new curated YAML configuration files for DeepSeek V4 Flash (latency-optimized and throughput-optimized variants) defining CUDA graph, KV cache, MoE, parallelism, and speculative decoding settings, and registers both in lookup.yaml with corresponding model mapping entries.

Changes

DeepSeek V4 Flash curated configs and lookup registration

Layer / File(s)	Summary
Latency and throughput config files examples/configs/curated/deepseek-v4-flash-latency.yaml, examples/configs/curated/deepseek-v4-flash-throughput.yaml	New YAML configs define CUDA graph batch sizes, KV cache dtype/memory settings, MoE/TRTLLM backend parameters, parallelism (TP/PP/EP), and speculative decoding options tuned for latency vs. throughput scenarios.
Lookup registration for new configs examples/configs/curated/lookup.yaml	Adds two deepseek-ai/DeepSeek-V4-Flash entries mapping to the new latency and throughput configs, both using DeepseekV4ForCausalLM and gpu_compatibility: "B200, GB200".

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Title check	⚠️ Warning	The title names DeepSeek-V4-Pro, but the changes add DeepSeek-V4-Flash curated configs, so it is misleading.	Rename the title to match the actual DeepSeek-V4-Flash curated configs and keep the [None][feat] format.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description check	✅ Passed	The description includes the required Description, Test Coverage, and PR Checklist sections and is mostly complete.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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