Skip to content

fix: Updated default reasoning model for nvidia #568

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

Merged
andreatgretel merged 5 commits into from
Apr 22, 2026
+23 −14
Merged

fix: Updated default reasoning model for nvidia #568

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
8d94e97
Updated default reasoning model for nvidia
kirit93 Apr 22, 2026
b7e3c81
Updated inference params for super
kirit93 Apr 22, 2026
5033446
Add reasoning_effort to Nemotron Super params, update stale docs
andreatgretel Apr 22, 2026
5156c9c
Merge branch 'main' into kirit93/update-default-models
andreatgretel Apr 22, 2026
f42bd97
Remove build-time README accidentally tracked
andreatgretel Apr 22, 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
2 changes: 1 addition & 1 deletion docs/concepts/models/default-model-settings.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ The following model configurations are automatically available when `NVIDIA_API_
| Alias | Model | Use Case | Inference Parameters |
|-------|-------|----------|---------------------|
| `nvidia-text` | `nvidia/nemotron-3-nano-30b-a3b` | General text generation | `temperature=1.0, top_p=1.0` |
| `nvidia-reasoning` | `openai/gpt-oss-20b` | Reasoning and analysis tasks | `temperature=0.35, top_p=0.95` |
| `nvidia-reasoning` | `nvidia/nemotron-3-super-120b-a12b` | Reasoning and analysis tasks | `temperature=1.0, top_p=0.95` |
| `nvidia-vision` | `nvidia/nemotron-nano-12b-v2-vl` | Vision and image understanding | `temperature=0.85, top_p=0.95` |
| `nvidia-embedding` | `nvidia/llama-3.2-nv-embedqa-1b-v2` | Text embeddings | `encoding_format="float", extra_body={"input_type": "query"}` |

Expand Down
6 changes: 5 additions & 1 deletion packages/data-designer-config/src/data_designer/config/utils/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -336,6 +336,7 @@ class NordColor(Enum):
DEFAULT_VISION_INFERENCE_PARAMS = {"temperature": 0.85, "top_p": 0.95}
DEFAULT_EMBEDDING_INFERENCE_PARAMS = {"encoding_format": "float"}
NEMOTRON_3_NANO_30B_A3B_INFERENCE_PARAMS = {"temperature": 1.0, "top_p": 1.0}
NEMOTRON_3_SUPER_120B_A12B_INFERENCE_PARAMS = {"temperature": 1.0, "top_p": 0.95}
Copy link
Copy Markdown
Contributor

@andreatgretel andreatgretel Apr 22, 2026

Choose a reason for hiding this comment

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

Nemotron Super supports reasoning_effort as a request param (same as gpt-oss did). Might be worth setting a default here - without it the model runs full chain-of-thought on every call, which can add up for bulk generation. We don't set it for gpt-oss either so not blocking, but since we're touching this anyway:

NEMOTRON_3_SUPER_120B_A12B_INFERENCE_PARAMS = {"temperature": 1.0, "top_p": 0.95, "extra_body": {"reasoning_effort": "medium"}}

Mirrors what we do for GPT-5 on line 340.

GPT5_INFERENCE_PARAMS = {"extra_body": {"reasoning_effort": "medium"}}

PREDEFINED_PROVIDERS_MODEL_MAP = {
Expand All @@ -344,7 +345,10 @@ class NordColor(Enum):
"model": "nvidia/nemotron-3-nano-30b-a3b",
"inference_parameters": NEMOTRON_3_NANO_30B_A3B_INFERENCE_PARAMS,
},
"reasoning": {"model": "openai/gpt-oss-20b", "inference_parameters": DEFAULT_REASONING_INFERENCE_PARAMS},
"reasoning": {
"model": "nvidia/nemotron-3-super-120b-a12b",
"inference_parameters": NEMOTRON_3_SUPER_120B_A12B_INFERENCE_PARAMS,
},
"vision": {"model": "nvidia/nemotron-nano-12b-v2-vl", "inference_parameters": DEFAULT_VISION_INFERENCE_PARAMS},
"embedding": {
"model": "nvidia/llama-3.2-nv-embedqa-1b-v2",
Expand Down
6 changes: 3 additions & 3 deletions packages/data-designer-config/tests/config/test_default_model_settings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,9 +30,9 @@ def test_get_default_inference_parameters():
top_p=0.95,
)
assert get_default_inference_parameters(
"reasoning", {"temperature": 0.35, "top_p": 0.95}
"reasoning", {"temperature": 1.0, "top_p": 0.95}
) == ChatCompletionInferenceParams(
temperature=0.35,
temperature=1.0,
top_p=0.95,
)
assert get_default_inference_parameters(
Expand All @@ -59,7 +59,7 @@ def test_get_builtin_model_configs():
assert builtin_model_configs[0].model == "nvidia/nemotron-3-nano-30b-a3b"
assert builtin_model_configs[0].provider == "nvidia"
assert builtin_model_configs[1].alias == "nvidia-reasoning"
assert builtin_model_configs[1].model == "openai/gpt-oss-20b"
assert builtin_model_configs[1].model == "nvidia/nemotron-3-super-120b-a12b"
assert builtin_model_configs[1].provider == "nvidia"
assert builtin_model_configs[2].alias == "nvidia-vision"
assert builtin_model_configs[2].model == "nvidia/nemotron-nano-12b-v2-vl"
Expand Down
Loading