Skip to content

feat(ci): add Devstral 2 nightly benchmark #1022

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
nishanthp wants to merge 3 commits into from
Closed

feat(ci): add Devstral 2 nightly benchmark #1022

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3732dc7
feat(ci): add Devstral 2 nightly benchmark
nishanthp Apr 2, 2026
81176e9
Merge branch 'main' into add-devstral2-nightly-benchmark
nishanthp Apr 2, 2026
8fccc6d
Merge branch 'main' into add-devstral2-nightly-benchmark
nishanthp Apr 2, 2026
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
3 changes: 3 additions & 0 deletions .github/workflows/nightly-benchmark.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -348,6 +348,9 @@ jobs:
matrix:
model:
- { id: meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8, slug: meta-llama-Llama-4-Maverick-17B-128E-Instruct-FP8, test_class: TestNightlyLlama4MaverickSingle }
- id: mistralai/Devstral-2-123B-Instruct-2512
slug: mistralai-Devstral-2-123B-Instruct-2512
test_class: TestNightlyDevstral123bSingle

@smfirmin smfirmin Apr 2, 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.

nit: lets collapse to one line for parity

# Pausing MiniMax M2 nightly benchmark
# - { id: minimaxai/minimax-m2, slug: minimaxai-minimax-m2, test_class: TestNightlyMinimaxM2Single }
variant:
Expand Down
1 change: 1 addition & 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 +102,7 @@ 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", "Devstral123b", 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 @@ -80,6 +80,15 @@ def _resolve_model_path(hf_path: str) -> str:
'{"disable_any_whitespace": true, "backend": "xgrammar"}',
],
},
# Devstral 2 - nightly benchmarks
"mistralai/Devstral-2-123B-Instruct-2512": {
"model": _resolve_model_path("mistralai/Devstral-2-123B-Instruct-2512"),
"tp": 8,
"features": ["chat", "streaming", "function_calling"],
"worker_args": ["--tool-call-parser", "mistral"],

@chatgpt-codex-connector chatgpt-codex-connector Bot Apr 2, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 Badge Use recognized SGLang spec key for Devstral args

worker_args is not consumed by the worker launcher, so this new Devstral setting is currently a no-op. In e2e_test/infra/worker.py, SGLang startup only reads spec.get("sglang_args", []) (and vLLM reads vllm_args), which means the --tool-call-parser mistral flag is never applied for the sglang nightly job and those benchmark runs execute with unintended runtime settings.

Useful? React with 👍 / 👎.

"vllm_args": ["--tool-call-parser", "mistral", "--enable-auto-tool-choice"],
Comment on lines +88 to +89

@gemini-code-assist gemini-code-assist Bot Apr 2, 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

There's an inconsistency in the configured arguments for SGLang (worker_args) and vLLM (vllm_args). The vllm_args enable auto tool choice, while worker_args do not. For a fair benchmark comparison, both runtimes should be configured with the same features. Please either add the corresponding flag to worker_args if SGLang supports it, or remove --enable-auto-tool-choice from vllm_args if it does not.

@smfirmin smfirmin Apr 2, 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.

lets remove enable-auto-tool-choice

@nishanthp nishanthp Apr 2, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

@slin1237 @CatherineSue Should I just remove --enable-auto-tool-choice from vllm_args?

"startup_timeout": 1200,
},
Comment on lines +83 to +91

@coderabbitai coderabbitai Bot Apr 2, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

⚠️ Potential issue | 🟠 Major

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Description: Verify which spec keys are accessed in worker.py

echo "=== Keys accessed from spec dict in worker.py ==="
rg -n 'spec\.get\(' e2e_test/infra/worker.py

echo ""
echo "=== Any reference to worker_args? ==="
rg -n 'worker_args' e2e_test/

Repository: lightseekorg/smg

Length of output: 624

worker_args is never read; must be sglang_args for tool-call-parser to be configured.

The spec keys accessed in e2e_test/infra/worker.py are: tp, features, sglang_args, vllm_args, trtllm_extra_config, trtllm_args, and startup_timeout. The worker_args key on line 88 is never consumed and has no effect. Since this model spec declares "function_calling" in features and includes tool-call-parser configuration, rename worker_args to sglang_args so the setting is actually applied to the SGLang engine.

🔧 Proposed fix 
     # Devstral 2 - nightly benchmarks
     "mistralai/Devstral-2-123B-Instruct-2512": {
         "model": _resolve_model_path("mistralai/Devstral-2-123B-Instruct-2512"),
         "tp": 8,
         "features": ["chat", "streaming", "function_calling"],
-        "worker_args": ["--tool-call-parser", "mistral"],
+        "sglang_args": ["--tool-call-parser", "mistral"],
         "vllm_args": ["--tool-call-parser", "mistral", "--enable-auto-tool-choice"],
         "startup_timeout": 1200,
     },
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
# Devstral 2 - nightly benchmarks
"mistralai/Devstral-2-123B-Instruct-2512": {
"model": _resolve_model_path("mistralai/Devstral-2-123B-Instruct-2512"),
"tp": 8,
"features": ["chat", "streaming", "function_calling"],
"worker_args": ["--tool-call-parser", "mistral"],
"vllm_args": ["--tool-call-parser", "mistral", "--enable-auto-tool-choice"],
"startup_timeout": 1200,
},
# Devstral 2 - nightly benchmarks
"mistralai/Devstral-2-123B-Instruct-2512": {
"model": _resolve_model_path("mistralai/Devstral-2-123B-Instruct-2512"),
"tp": 8,
"features": ["chat", "streaming", "function_calling"],
"sglang_args": ["--tool-call-parser", "mistral"],
"vllm_args": ["--tool-call-parser", "mistral", "--enable-auto-tool-choice"],
"startup_timeout": 1200,
},
🤖 Prompt for AI Agents 
Verify each finding against the current code and only fix it if needed.

In `@e2e_test/infra/model_specs.py` around lines 83 - 91, The model spec for
"mistralai/Devstral-2-123B-Instruct-2512" mistakenly uses the unused key
worker_args instead of sglang_args, so the tool-call-parser config never reaches
the SGLang engine; change the key name from "worker_args" to "sglang_args" in
that dict so the value ["--tool-call-parser", "mistral"] is consumed (the worker
reads tp, features, sglang_args, vllm_args, trtllm_extra_config, trtllm_args,
startup_timeout in e2e_test/infra/worker.py), ensuring function_calling
tool-parser is applied.

# Embedding model
"intfloat/e5-mistral-7b-instruct": {
"model": _resolve_model_path("intfloat/e5-mistral-7b-instruct"),
Expand Down