Skip to content

ci(benchmarks): add Devstral 2 to H200 nightly benchmarks #1068

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

Closed
smfirmin wants to merge 1 commit into from
Closed

ci(benchmarks): add Devstral 2 to H200 nightly benchmarks #1068

Show file tree
Hide file tree
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
1 change: 1 addition & 0 deletions .github/workflows/nightly-benchmark.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -350,6 +350,7 @@ jobs:
- { id: meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8, slug: meta-llama-Llama-4-Maverick-17B-128E-Instruct-FP8, test_class: TestNightlyLlama4MaverickSingle }
# Pausing MiniMax M2 nightly benchmark
# - { id: minimaxai/minimax-m2, slug: minimaxai-minimax-m2, test_class: TestNightlyMinimaxM2Single }
- { id: mistralai/Devstral-2-123B-Instruct-2512, slug: mistralai-Devstral-2-123B-Instruct-2512, test_class: TestNightlyDevstral2Single }
variant:
- { id: sglang, runtime: sglang, grpc_only: "false", setup_vllm: false, setup_trtllm: false, extra_deps: "genai-bench" }
- { id: vllm, runtime: vllm, grpc_only: "false", setup_vllm: true, setup_trtllm: false, extra_deps: "genai-bench" }
Expand Down
7 changes: 7 additions & 0 deletions e2e_test/benchmarks/test_nightly_perf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,13 @@ def _run_nightly(setup_backend, genai_bench_runner, model_id, worker_count=1, **
("Qwen/Qwen3-30B-A3B", "Qwen30b", 4, ["http", "grpc"], {}),
("openai/gpt-oss-20b", "GptOss20b", 1, ["http", "grpc"], {}),
("minimaxai/minimax-m2", "MinimaxM2", 1, ["http", "grpc"], {}),
(
"mistralai/Devstral-2-123B-Instruct-2512",

@gemini-code-assist gemini-code-assist Bot Apr 8, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

high

The model ID mistralai/Devstral-2-123B-Instruct-2512 appears to contain typos. "Devstral" is not a known Mistral model series (likely "Mistral" or "Codestral"), and the date suffix "2512" (December 2025) is likely a typo for "2412" or "2411". An incorrect model ID will cause the benchmark to fail during model download. Additionally, the PR description mentions that minimax-m2 (line 104) should be commented out, but it remains active in the code.

"Devstral2",
1,
["http", "grpc"],
{},
),
(
"meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
"Llama4Maverick",
Expand Down
9 changes: 9 additions & 0 deletions e2e_test/infra/model_specs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,6 +114,15 @@ def _resolve_model_path(hf_path: str) -> str:
"sglang_args": ["--trust-remote-code"],
"vllm_args": ["--trust-remote-code"],
},
# Devstral 2 123B - Nightly benchmarks
"mistralai/Devstral-2-123B-Instruct-2512": {
"model": _resolve_model_path("mistralai/Devstral-2-123B-Instruct-2512"),
"tp": 4,
"features": ["chat", "streaming", "function_calling", "reasoning"],

@gemini-code-assist gemini-code-assist Bot Apr 8, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

medium

The reasoning feature is included here but is omitted for other similar high-capability models like Llama-3.3-70B-Instruct (line 171). In this codebase, the reasoning feature typically refers to models that support a specific reasoning output field (e.g., DeepSeek-R1 or Harmony-compatible models). Unless this model specifically supports such a field in its API response, this feature should be removed to maintain consistency and avoid triggering incompatible tests.

Suggested change
"features": ["chat", "streaming", "function_calling", "reasoning"],
"features": ["chat", "streaming", "function_calling"],

"startup_timeout": 1200,
"sglang_args": ["--trust-remote-code"],
"vllm_args": ["--trust-remote-code"],
},
# Vision-language model for multimodal benchmarks (MMMU)
"Qwen/Qwen3-VL-8B-Instruct": {
"model": _resolve_model_path("Qwen/Qwen3-VL-8B-Instruct"),
Expand Down