NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16

2026-10-15

Date the weights (or, if earlier, the technical report) were first made public, YYYY-MM-DD, as dated by the publisher. A Hugging Face repository creation date is used only when no dated publisher page exists, with a note.

other (openmdw-1.1)

License of the released weights as declared by the publisher (model card metadata or license file), e.g. apache-2.0, mit, or the name of a custom license.

text + image + video

Input modalities of the released checkpoint as the publisher states them, from: text, image, video, audio, joined with ' + ' in that order ("text", "text + image + video").

74B parameters

Total parameter count as reported by the publisher. See params.note for whether embeddings, MTP heads or vision towers are included.

9B parameters

Parameters used per token in the forward pass, as reported by the publisher.

Not disclosed

Active parameters excluding the input embedding and the output head (unembedding). Stated by the publisher, or computed from params.active by subtracting vocab × d_model for each of the embedding and output head that params.active is stated to include (basis inferred, note with the arithmetic). Not disclosed when the scope of params.active is not stated.

74.23B parameters

Calculated: embedding + output head + all experts + other weights, nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16@e09a4ce safetensors headers; reproduces stated total (backbone) and active (with output head)

Parameters of the text backbone counted from the released checkpoint's safetensors headers (embedding, output head, all experts and other weights; excluding MTP modules, vision/audio towers and n-gram lookup tables). Shown only when the counts reproduce the publisher's stated total and active parameters within 5% under some scope.

1.074B parameters

Calculated: input embedding + output head, nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16@e09a4ce safetensors headers

Input embedding plus output head (once if tied), counted from the checkpoint.

73.16B parameters

Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16@e09a4ce safetensors headers; reproduces stated total (backbone) and active (with output head)

Language backbone weights excluding input embeddings, output head, lookup/Engram tables, MTP modules and vision/audio towers. Equals params.counted_total minus params.counted_embedding. Same validation as params.counted_total.

2.655B parameters

Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16@e09a4ce safetensors headers

Largest group of language-backbone weights sharing a full layers.<index> tensor path, including every routed expert. Excludes lookup/Engram tables, embeddings, output head, MTP and other modalities. Hybrid checkpoints may store mixers and FFNs as separate layers. All-gather volume also depends on wrapping, sharding and dtype.

8.708B parameters

Calculated: non-expert weights + sum of each layer's routed weights × its top_k / experts, excluding embedding and output head, nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16@e09a4ce safetensors headers; reproduces stated total (backbone) and active (with output head)

Active parameters per token excluding the input embedding and output head, counted from the checkpoint as all non-expert weights plus routed-expert weights × top_k / experts. Same validation as params.counted_total.

Rounded publisher figures; scope (embeddings, vision encoder) not stated. No MTP module. The safetensors metadata lists 74.885B parameters including the C-RADIO v4-H vision encoder and projector.

What the publisher's counts include or exclude (MTP modules, embeddings, vision encoder, n-gram tables).

48

Inferred from the source; see the explanation below.

Number of transformer blocks in the main backbone (attention or token mixer plus FFN), excluding MTP modules.

4,096

Residual stream width.

131,072

Embedding table rows (vocab_size in config, may include padding).

No

Input embedding and output head share weights.

85.3333

Calculated: d_model / layers

Width-to-depth ratio, d_model divided by the number of blocks.

GQA, Mamba-2

Token-mixing layer types in the backbone, e.g. GQA, MLA, SWA, Gated DeltaNet, Mamba-2, DSA (DeepSeek sparse attention).

40 Mamba-2 + 8 GQA

Inferred from the source; see the explanation below.

Token-mixer order, summarized as a pattern or layer positions. The fact's optional sequence lists each mixer in order; a ratio alone does not establish the order. Auxiliary prediction modules and standalone FFNs are excluded.

8

Layers whose KV cache grows with full context (dense or sparse global attention, MLA). Excludes sliding-window and linear/recurrent layers.

5

Calculated: attn.linear_layers / attn.global_layers

attn.linear_layers divided by attn.global_layers (3 for '3 Gated DeltaNet : 1 GQA').

40

Layers with a fixed-size recurrent state (linear attention, DeltaNet variants, Mamba).

32

Query heads in the global-attention layers.

2

Key/value heads in the global-attention layers. Not applicable to MLA, whose cache is a shared latent.

4,096

Calculated: heads × head_dim

attn.heads × attn.head_dim, the query projection width of global layers.

16

Calculated: heads / kv_heads

attn.heads divided by attn.kv_heads in global layers.

128

Query/key dimension per head in global-attention layers (for MLA, nope + rope dims).

NoPE

Positional scheme in global layers: RoPE, partial RoPE, NoPE, or mixed per layer type.

No

Queries and keys are normalized per head before the dot product.

No

Query/key/value projections have bias terms.

Mamba-2: 128 heads × head dim 64, state 128, 8 groups, conv kernel 4

Configuration of the linear / SSM layers (e.g. Gated DeltaNet heads and dims, Mamba-2 state size), as one string.

128

Value heads (or SSM heads) in the linear / SSM layers.

64

Head dimension of the linear / SSM layers (value head dim for DeltaNet variants, head_dim for Mamba-2).

4

Kernel size of the causal short convolution in linear / SSM layers or attention projections.

ReLU²

FFN activation (SwiGLU, GeGLU, ...).

No

FFN projections have bias terms.

496

Inferred from the source; see the explanation below.

Routed experts per MoE layer.

9

Routed experts selected per token.

1

Always-on experts per MoE layer.

2,560

Inferred from the source; see the explanation below.

Intermediate width of one routed expert.

5,376

Total intermediate width of the shared expert(s).

0

Number of initial blocks that use a dense FFN instead of MoE.

sigmoid

Gating function over expert logits (softmax, sigmoid).

Not disclosed

Load-balancing method, with coefficients where disclosed.

5

Multiplier applied to the combined routed-expert output (routed_scaling_factor).

Yes

Selected expert weights are renormalized to sum to 1.

1,024

Width of the projected space routed experts operate in (LatentMoE), when smaller than d_model.

4

Calculated: d_model / moe.latent_dim

shape.d_model divided by moe.latent_dim, how much LatentMoE compresses the hidden state before routed experts.

0.625

Calculated: moe.expert_dim / d_model

moe.expert_dim divided by shape.d_model.

55.1111

Calculated: experts / top_k

Routed experts divided by experts selected per token.

0.121622

Calculated: params.active / params.total

params.active divided by params.total.

0

Inferred from the source; see the explanation below.

Additional future tokens predicted during pretraining (0 when the model was trained without MTP).

RMSNorm

Normalization layer type in the blocks.

pre

Pre-norm, post-norm, or sandwich (pre and post).

1e-05

Epsilon of the block normalization layers.

8.192K tokens

Sequence length of the main pretraining stage.

131.1K tokens

Longest sequence length trained on after context extension.

262.1K tokens

Longest context the publisher says the model supports, possibly beyond trained length via RoPE scaling.

Distillation recovery continued at 128K for 48B tokens after 72B tokens at 8K; no RoPE to rescale (NoPE)

How context was extended (stages, RoPE base change, YaRN factor).

120B tokens

Tokens seen in pretraining including annealing and context extension, as reported.

Not disclosed

Optimizer used for the bulk of parameters.

Not disclosed

Peak learning rate of the main pretraining stage.

Not disclosed

Learning rate at the end of the decay.

Not disclosed

Learning-rate schedule shape with phase boundaries.

Not disclosed

Warmup length in steps or tokens.

2,048 × 8,192 at 8K; 128 × 131,072 at 128K (16.8M tokens in both)

Global batch size, including any batch-size ramp: sequences × sequence length (token total), or the token total alone when only that is stated.

16.78M tokens

Inferred from the source; see the explanation below.

Global batch size in tokens during the main pretraining stage, after any ramp (sequences × sequence length when stated in sequences, marked inferred).

Not disclosed

Decoupled weight decay coefficient.

Not disclosed

(beta1, beta2) of the Adam-family optimizer or its AdamW fallback.

Not disclosed

Global gradient-norm clipping threshold.

Not disclosed

Matmul precision of pretraining (BF16, FP8 recipe, NVFP4 recipe) with the scaling granularity where disclosed. Never inferred from the released checkpoint dtype.

Nemotron 3.5 Super VL 120B-A12B checkpoint with routed experts sorted by REAP scores and MoE intermediate channels by Star Elastic importance (estimated on 8M tokens), then pruned with Gumbel-softmax-weighted nested prefix masks chosen per layer by a budget-conditioned router

Weight initialization (std or scheme) as stated by the publisher; config initializer_range alone is not evidence.

Knowledge distillation from the frozen parent (token-level KL on next-token distributions) for the student weights, plus a router budget loss (λ = 1, tolerance 0.05) on the architecture router only

Objectives beyond next-token prediction and MTP (fill-in-the-middle rate, distillation, multimodal tokens from the start).

70% Nemotron 3.5 Super VL SFT data, 30% Nemotron Ultra-v3 pretraining data

Pretraining data composition as disclosed (domain shares, synthetic fraction, repetition).

February 24, 2026

Pretraining data cutoff date.

32 GB200 nodes (128 GPUs), about 64 hours as 16 consecutive four-hour jobs

Accelerators used for pretraining, count and duration where stated.

128 GB200

Accelerator count and type for the main pretraining run, as the count with thousands separators and the chip's short name ("2,048 H800", "2,048 GH200"). A stated range or bound stays as stated ("2,048–4,096 GH200").

• architecture search

  Tokens

  8000000000

  Seq len

  8192

  Batch sequences

  2048

  Source: paper · PDF page 6; §2.3

  Architecture search runs at 8K context, with the router converging during the first 8B tokens.

  Router and student trained jointly with KD plus budget loss until the router converges.

• short-context recovery

  Tokens

  64000000000

  Seq len

  8192

  Batch sequences

  2048

  Source: paper · PDF page 6; §2.3

  We then fix the architecture and continue distillation for another 64B tokens at 8K

  Architecture fixed; KD continues.

• long-context recovery

  Tokens

  48000000000

  Seq len

  131072

  Batch sequences

  128

  Source: paper · PDF page 6; §2.3

  A further 48B tokens at 128K brings the total to 120B tokens.

Ordered pretraining stages. Each item: name, tokens, seq_len, lr, batch_tokens, batch_sequences, steps, note, quote, and source when it differs from the fact's source.

1.62162

Calculated: train.tokens / params.total

train.tokens divided by params.total.

13.7812

Calculated: train.tokens / params.counted_active_non_embedding

train.tokens divided by params.counted_active_non_embedding, or by the stated params.active_non_embedding when there is no validated checkpoint count.

8,192 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. Global KV: 8 × (512 × 16/8) B/entry

Retained KV cache divided by context length, including bounded SWA caches and indexer keys. Exports use uniform BF16 at 131,072 tokens. The website can select uniform payload precision (no quantization overhead) or sourced component formats (including scales and documented padding). This is cache payload, not peak prefill memory or persistent prefix storage.

8,192 bytes

Calculated: 8192 B / 1 tokens (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 8 × (512 × 16/8) B/entry

Long-context global KV bytes per token, including indexer keys; excludes bounded SWA caches. Compression rates and cross-layer cache reuse are included. Exports use uniform BF16; the website's storage selector also applies here. This is the field to compare with global-KV figures in reports, using the matching storage recipe.

910 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 8 × (512 × 16/8) B/entry

kv.bytes_per_token divided by active parameters in billions (131,072 tokens in exports).

119,304,647 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 8 × (512 × 16/8) B/entry

kv.bytes_at_context divided by active parameters in billions (131,072 tokens in exports).

1,073,741,824 bytes

Calculated: 8192 B × 131072 (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 8 × (512 × 16/8) B/entry

Retained KV cache for one sequence at the selected context, including SWA windows and indexer keys. Exports use uniform BF16 at 131,072 tokens. Excludes recurrent state, compressor tail state, draft/MTP caches, temporary prefill tensors, allocator metadata and persistent prefix snapshots. Does not estimate peak prefill or total serving memory.

21

Artificial Analysis Intelligence Index v4.3.2 of the post-trained model, from a dated Wayback snapshot of its Artificial Analysis page. The note names the variant (reasoning setting).