Hy4-preview

2026-08-28

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.

apache-2.0

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

Inferred from the source; see the explanation below.

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").

770B parameters

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

49B 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.

769.9B parameters

Calculated: embedding + output head + all experts + other weights, tencent/Hy4-preview@705d81e safetensors headers; reproduces stated total (backbone) and active (with embedding and 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.

1.485B parameters

Calculated: input embedding + output head, tencent/Hy4-preview@705d81e safetensors headers

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

768.4B parameters

Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, tencent/Hy4-preview@705d81e safetensors headers; reproduces stated total (backbone) and active (with embedding and 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.

9.978B parameters

Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, tencent/Hy4-preview@705d81e 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.

47.57B parameters

Calculated: non-expert weights + routed experts × 8/256, excluding embedding and output head, tencent/Hy4-preview@705d81e safetensors headers; reproduces stated total (backbone) and active (with embedding and 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.

Backbone only, excluding the MTP layer (10B total, 0.7B active).

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

78

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

6,144

Residual stream width.

120,832

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

No

Input embedding and output head share weights.

78.7692

Calculated: d_model / layers

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

MLA, DSA

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

78 DSA

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.

78

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

64

Query heads in the global-attention layers.

16,384

Calculated: heads × head_dim

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

256

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

256

Value dimension per head in global layers (MLA v_head_dim; stated for MLA even when equal to the QK head dim).

512

Width of the compressed KV latent that MLA caches per token per layer.

2,048

Width of the low-rank query projection in MLA; 0 when queries use a full-rank projection.

64

Width of the shared positional key that MLA caches next to the latent.

partial RoPE (MLA decoupled, 64 of 256 QK dims)

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

10,000,000

RoPE base frequency of the global layers at release.

No

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

elementwise sigmoid gate computed from the block input, before o_proj

Gate applied to the attention output before the output projection, with its placement.

DSA indexer (32 heads × 128, top-2048); IndexCache: 21 layers run the indexer, 57 reuse the previous indexer layer's top-k

Token-selection mechanism inside global layers, e.g. DSA lightning indexer with top-k.

learned per-head sink logit, all 78 layers

Learned sink logits or sink bias in the softmax, with the layer types that use them.

No

Query/key/value projections have bias terms.

128

Width of the sparse-attention indexer key cached per token in each global layer (DSA lightning indexer), counted in the KV cache.

• MLA latent + RoPE key

  Layers

  78

  Elements

  576

  Tokens per entry

  1

  Source: code · modeling_hy_v4.py; HYV4Attention.__init__, HYV4Attention.forward

  k_pass, k_rot = past_key_values.update(k_pass, k_rot, self.layer_idx)

  kv_lora_rank 512 + qk_rope_head_dim 64.

• DSA indexer key

  Layers

  21

  Elements

  128

  Tokens per entry

  1

  Source: code · modeling_hy_v4.py; HYV4Attention.__init__, HYV4Attention.forward

  self.indexer = None if self.skip_topk else HYV4Indexer(config, layer_idx)

  Only layers with indexer_types 'full' own an indexer; 'shared' layers reuse the previous top-k.

Cache owners and entry shapes for compressed or shared attention. Each item specifies layers owning the cache, scalar elements per entry (K and V together), tokens_per_entry (compression ratio, default 1), and optionally window (maximum entries retained). Completed compressed entries are floor(context / tokens_per_entry); incomplete compressor state is excluded. KV sizes use these groups when present.

SwiGLU

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

18,432

Intermediate width of the dense FFN layers that are actually executed.

SwiGLU limit 10 (gate ≤ 10, up in [-10, 10]), routed experts only

Clamping of GLU gate / up activations (e.g. SwiGLU limit 7), with the layers it applies to.

No

FFN projections have bias terms.

3

Calculated: ffn.dim / d_model

ffn.dim divided by shape.d_model.

256

Routed experts per MoE layer.

8

Routed experts selected per token.

1

Always-on experts per MoE layer.

2,048

Intermediate width of one routed expert.

2,048

Inferred from the source; see the explanation below.

Total intermediate width of the shared expert(s).

1

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

sigmoid

Gating function over expert logits (softmax, sigmoid).

expert bias added to selection scores (e_score_correction_bias); update rule and any aux loss not disclosed

Load-balancing method, with coefficients where disclosed.

2.827

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

Yes

Selected expert weights are renormalized to sum to 1.

0.333333

Calculated: moe.expert_dim / d_model

moe.expert_dim divided by shape.d_model.

32

Calculated: experts / top_k

Routed experts divided by experts selected per token.

0.0636364

Calculated: params.active / params.total

params.active divided by params.total.

Not disclosed

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

Yes

The released checkpoint contains the MTP module weights.

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.

iHC (identity Hyper-Connections), 4 streams: per sublayer, input-dependent sigmoid read weights and 2·sigmoid write weights, identity stream mixing (no Sinkhorn); streams merged by a sigmoid-weighted sum before the final norm

Residual design when it is not a plain sum, e.g. hyper-connections with n streams, gated residuals, depth-scaled residuals, FP32 residual accumulation.

Not disclosed

Sequence length of the main pretraining stage.

Not disclosed

Longest sequence length trained on after context extension.

1.049M tokens

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

Not disclosed

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.

Not disclosed

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

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.

95,232 bytes

Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. MLA latent + RoPE key: 78 × (576 × 16/8) B/entry; DSA indexer key: 21 × (128 × 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.

95,232 bytes

Calculated: 89856 B / 1 tokens (MLA latent + RoPE key) + 5376 B / 1 tokens (DSA indexer key); uniform BF16, payload only; quantization overhead excluded. MLA latent + RoPE key: 78 × (576 × 16/8) B/entry; DSA indexer key: 21 × (128 × 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.

1,944 bytes

Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. MLA latent + RoPE key: 78 × (576 × 16/8) B/entry; DSA indexer key: 21 × (128 × 16/8) B/entry

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

254,739,769 bytes

Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. MLA latent + RoPE key: 78 × (576 × 16/8) B/entry; DSA indexer key: 21 × (128 × 16/8) B/entry

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

12,482,248,704 bytes

Calculated: 89856 B × 131072 (MLA latent + RoPE key) + 5376 B × 131072 (DSA indexer key); uniform BF16, payload only; quantization overhead excluded. MLA latent + RoPE key: 78 × (576 × 16/8) B/entry; DSA indexer key: 21 × (128 × 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.