SKILLEMALL.ai
GitHub ↗ RU Sign in

CC nemotron-ultra

Reference desk for NVIDIA Nemotron 3 Ultra (550B-A55B) — architecture, NVFP4 pretraining, SFT, MOPD (multi-teacher on-policy distillation), MTP boosting, quantization, inference. Use when the user asks facts about Ultra rather than building a pipeline.

The skillemall take

Reference desk for NVIDIA's Nemotron 3 Ultra (550B): covers NVFP4 pretraining, SFT, MOPD multi-teacher distillation, MTP boosting, quantization. Use it to answer questions about the model's internals, not to build production pipelines.

Grade C with quality 80 but process score 63 signals gaps in how the material is organized. Twenty files, zero scripts—it's a lookup tool, not a runnable toolkit. No critical findings; fifteen medium and low-level issues scattered through. Safety at 85, so the facts about the model check out. Install if you regularly dig into Nemotron docs; if you need working code, keep looking.

NVIDIA-NeMo/Nemotron Agent Skills author: NVIDIA-NeMo 20 files body ≈ 1 773 tokens Open the sourcegithub.com↗ analyzed 5 d ago

Reference desk for NVIDIA Nemotron 3 Ultra (550B-A55B) — architecture, NVFP4 pretraining, SFT, MOPD (multi-teacher on-policy distillation), MTP boosting…

As a process C 63/100 · Has gaps — weak spots: result and completion, inputs and preconditions, running it twice

ReferenceInfrastructuretype and topics are labelled automatically from the skill text
Runs in: Claude / Claude Code ChatGPT / Codex Gemini CLI Cursor GitHub Copilot / VS Code Qwen Code Kimi Code CLI Z.ai (GLM) DeepSeek (Deep Code) Mistral Vibe LangChain (deepagents) OpenClaw Goose OpenCode
JSON
Technical rating
C
83/100
safety, quality, tests
Safety 60%
85
Quality 40%
80
Run on models
none yet
Process rating
C
63/100
Has gaps
Result and completion w 14
0
Inputs and preconditions w 11
0
Progress reporting w 2
0
the three weakest of ten parameters · all ten

How to improve

    For the model run — optional
    • Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
    • A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.

    Guard findings · 15

    ✓ No critical or high findings

    Medium and low: 15
    • low Secrets in code secret-high-entropy-token context/quick-reference.md:90
      High-entropy token-like string (may be an id, hash or a credential)
      - **Nemo…-v3**: 173B new GitHub tokens, cutoff Sept 30 2025
    • low Secrets in code secret-high-entropy-token context/quick-reference.md:93
      High-entropy token-like string (may be an id, hash or a credential)
      - **Moral-Scenarios**: Moral Stories + Social Chemistry; CoT via Qwen…507
    • low Secrets in code secret-high-entropy-token paper/_overview.md:24
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "Released datasets named in intro: Nemo…-v3 (173B tokens, GitHub through Sept 30 2025), Nemo…-v1, Nemo….2."
      quoted
    • low Secrets in code secret-high-entropy-token paper/_overview.md:74
      High-entropy token-like string (may be an id, hash or a credential)
      - Nemo…v3: 173B tokens of fresh GitHub code, cut-off Sept 30, 2025.
    • low Secrets in code secret-high-entropy-token paper/_overview.md:75
      High-entropy token-like string (may be an id, hash or a credential)
      - Nemo…v1: synthetic legal datasets; one ablation boosted a proxy LegalBench average from 64.6 to 74.7 (on Nemotron 3 Nano pretraining).
    • low Secrets in code secret-high-entropy-token paper/data.md:13
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "Nemo…v3: 173B new tokens of GitHub source code, cut-off date September 30, 2025."
      quoted
    • low Secrets in code secret-high-entropy-token paper/data.md:37
      High-entropy token-like string (may be an id, hash or a credential)
      | Nemo…-v3 (§2.3.1) | 173B new tokens of GitHub source code, cut-off Sept 30, 2025 | (refresh; no separate ablation reported) |
    • low Secrets in code secret-high-entropy-token paper/data.md:40
      High-entropy token-like string (may be an id, hash or a credential)
      | Nemotron-Pretraining-Moral-Scenarios (§2.3.4) | MC moral-scenario Q from Moral Stories (situations/norms) + Social Chemistry (actions); CoT version via Qwen…507 | (no separate m
    • low Secrets in code secret-high-entropy-token paper/data.md:57
      High-entropy token-like string (may be an id, hash or a credential)
      - Synthetic (all via Qwen…507 unless noted): Legal-Canadian-Case-Law-Outcome (CHRT/RPD/RAD/RLLR subsets), Legal-Definition-Classification (from Caselaw), Legal-Diversity-Jurisdict
    • low Secrets in code secret-high-entropy-token paper/evaluation.md:48
      High-entropy token-like string (may be an id, hash or a credential)
      # §2.6 Base Model Evaluations (Nemo…ase)
    • low Secrets in code secret-high-entropy-token paper/evaluation.md:52
      High-entropy token-like string (may be an id, hash or a credential)
      | Task | Metric | N3-U…ase | Deep…ase | Mist…512 | Kimi…ase | GLM-…ase |
    • low Secrets in code secret-high-entropy-token paper/mopd/mtp-reasoning.md:22
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "MTP Boosting data: on-policy rollouts from the MOPD checkpoint, seed prompts from Nemo…-v2 (general) and Nemotron-RL-Super-Training-Blends (agentic); rollouts sampled with
      quoted
    • low Secrets in code secret-high-entropy-token paper/mopd/mtp-reasoning.md:55
      High-entropy token-like string (may be an id, hash or a credential)
      | Seed prompts (general) | Nemo…-v2 |
    • low Secrets in code secret-high-entropy-token paper/sft.md:30
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "Software issue resolution: synthetic GitHub-issue trajectories from Mini…2.5 (thinking) and Qwen…uct (non-thinking), captured with OpenHands, SWE-agent, Mini-SWE-agent, Op
      quoted
    • low Secrets in code secret-high-entropy-token paper/sft.md:87
      High-entropy token-like string (may be an id, hash or a credential)
      | Software Issue Resolution | Synthetic GitHub-issue trajectories; Mini…2.5 (thinking) + Qwen…uct (non-thinking); heuristic-analyzer filtered |

    Files scanned: 20. Evidence is masked. Grey chips explain why severity was lowered.

    Against the Agent Skills spec

    ✓ No remarks against the Agent Skills spec

    Process rating: all ten parameters 63/100

    • 0Result and completion. Does not say what the result is
    • 0Inputs and preconditions. Does not say what the process needs to start
    • 0Progress reporting. Says nothing while it works
    • 30Running it twice. 1 mutating operations with no state check
    • 60Failures and branches. 2 branches
    • 70When it triggers. States when to use, but not when not to
    • 100Tools and files. No external tools needed
    • 100Steps. 50 steps
    • 100Consistency. Name and required fields are in place
    • 100Execution cost. Instruction body is 1773 tokens

    Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.

    Quality signals

    • +5Description has no quoted example phrases that should trigger the skill
    • +4Description does not say when NOT to use the skill (false activations)
    • +3Output format is not stated: the model decides each time
    • +4No input/output examples
    • +1No license
    • +2Single-language instructions
    • +3Description length 252: enough signal without eating the budget
    • +4Structure: 16 headings
    • +3Step-by-step instructions: 50 items

    Quality base 70; lint remarks subtract, signals add up to 100. Result: 80.