SKILLEMALL.ai
GitHub ↗ RU Sign in

CB nemotron-super35

Reference desk for NVIDIA Nemotron 3.5 Super (120B-A12B), covering vision-language architecture, MTP, distillation pretraining from Ultra 3 Base, Indic CPT, SFT, RLVR, specialist teachers, two-phase MOPD, MTP boosting, NVFP4 quantization, and inference. Use when the user asks facts about Super 3.5 rather than building a pipeline.

NVIDIA-NeMo/Nemotron Agent Skills author: NVIDIA-NeMo 27 files body ≈ 2 347 tokens Open the sourcegithub.com↗ analyzed 5 d ago

Reference desk for NVIDIA Nemotron 3.5 Super (120B-A12B), covering vision-language architecture, MTP, distillation pretraining from Ultra 3 Base, Indic CPT…

As a process B 68/100 · Nearly there — 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
B
68/100
Nearly there
Inputs and preconditions w 11
0
Progress reporting w 2
0
Running it twice w 4
30
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/index.toml:52
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      description = "Nemotron 3.5 Super Pretraining Data: Nemo…-v1, Indic CPT Mixture, Agentic Mid-training Blend"
      quoted
    • low Secrets in code secret-high-entropy-token context/quick-reference.md:40
      High-entropy token-like string (may be an id, hash or a credential)
      - Nemo…v1: ~2.83T synthetic tokens, 22 languages, 34 language-script configurations, 4-stage NeMo Data Designer pipeline (Gemm…-it).
    • low Secrets in code secret-high-entropy-token INDEX.md:30
      High-entropy token-like string (may be an id, hash or a credential)
      | Indic data / mid-training mix | `paper/data.md` | Nemo…-v1, Figure 2 pipeline, Tables 1–5 |
    • low Secrets in code secret-high-entropy-token model-card.md:52
      High-entropy token-like string (may be an id, hash or a credential)
      | Nemo…-v1 (~2.83T synthetic Indic tokens, 22 languages, 34 language-script configurations) | Announced |
    • low Secrets in code secret-high-entropy-token model-card.md:55
      High-entropy token-like string (may be an id, hash or a credential)
      | Nemo…-v2, Nemotron-RL-Super-Training-Blends | Public; used as seed prompts for MTP boosting (§3.7) |
    • low Secrets in code secret-high-entropy-token paper/_overview.md:96
      High-entropy token-like string (may be an id, hash or a credential)
      Indic-specific release commitments (§2.3, future tense in the report): the broader synthetic Indic corpus (Nemo…-v1), open-sourced generation pipelines, a companion technical repo
    • low Secrets in code secret-high-entropy-token paper/data.md:6
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      title: "Nemotron 3.5 Super Pretraining Data: Nemo…-v1, Indic CPT Mixture, Agentic Mid-training Blend"
      quoted
    • low Secrets in code secret-high-entropy-token paper/data.md:8
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      Covers the two data contributions of Super 3.5's base-model training. First, Nemo…v1:
      quoted
    • low Secrets in code secret-high-entropy-token paper/data.md:16
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "Nemo…v1: ~2.83T (2,825.6B) generated output tokens, 22 Indian languages, 34 language-script configurations, counted with a Gemma-4 tokenizer (Table 1)."
      quoted
    • low Secrets in code secret-high-entropy-token paper/data.md:33
      High-entropy token-like string (may be an id, hash or a credential)
      - What is Nemo…-v1 (languages, scripts, token counts, sources)?
    • low Secrets in code secret-high-entropy-token paper/data.md:44
      High-entropy token-like string (may be an id, hash or a credential)
      - **Nemo…-v1** total: ~**2.83T** generated tokens (2,825.6B, Gemma-4 tokenizer). Priority languages ~2.50T (incl. 224.6B romanized). Additional languages ~327B native-script.
    • low Secrets in code secret-high-entropy-token paper/evaluation.md:105
      High-entropy token-like string (may be an id, hash or a credential)
      | | tau2-Bench Retail / Airline / Telecom | avg over 8 trials; user simulator Qwen…507 |
    • low Secrets in code secret-high-entropy-token paper/rl/mtp-boosting.md:19
      High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)
      - "Data (§3.7): on-policy rollouts from the corresponding backbone checkpoint; seed prompts from Nemo…-v2 (general) and Nemotron-RL-Super-Training-Blends (agentic); temp = 1;
      quoted
    • low Secrets in code secret-high-entropy-token paper/rl/mtp-boosting.md:59
      High-entropy token-like string (may be an id, hash or a credential)
      | Seed prompts | Nemo…-v2 (general), Nemotron-RL-Super-Training-Blends (agentic) |
    • low Secrets in code secret-high-entropy-token SKILL.md:83
      High-entropy token-like string (may be an id, hash or a credential)
      | Indic data / Nemo…-v1 / mid-training mix | `paper/data.md` |

    Files scanned: 27. 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 68/100

    • 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
    • 40Result and completion. Does not say what the result is
    • 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. 58 steps
    • 100Consistency. Name and required fields are in place
    • 100Execution cost. Instruction body is 2347 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 331: enough signal without eating the budget
    • +4Structure: 17 headings
    • +3Step-by-step instructions: 58 items

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