Laguna-XS-2.1
Poolside · reasoning checkpoint
No base checkpoint is public; facts describe the released post-trained checkpoint.
Open in the comparison table Hugging Face ↗ Blog ↗
KV calculations use one sequence at 131,072 tokens and 16-bit cache elements. They include bounded sliding-window caches. This is retained cache payload, not peak prefill memory. The comparison table also offers sourced mixed-precision layouts, including quantization scales.
Release
- Released
2026-07-02
Source: blog · Introducing Laguna XS 2.1; post date
category:"announcements",date:"2026-07-02"
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.
- License
openmdw-1.1
Source: model-card · README.md; YAML front matter
license: 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.
- Input modalities
text
Source: model-card · Model overview
Modality: text-to-text
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").
Parameters
- Total params
33B parameters
Source: model-card · README.md; Model overview
Number of parameters: 33B total with 3B activated per token
Total parameter count as reported by the publisher. See params.note for whether embeddings, MTP heads or vision towers are included.
- Active params
3B parameters
Source: blog · Introducing Laguna XS 2.1; intro
Laguna XS 2.1 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token
Parameters used per token in the forward pass, as reported by the publisher.
- Active non-embedding (stated)
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.
- Counted total params
33.44B parameters
Calculated: embedding + output head + all experts + other weights, poolside/Laguna-XS-2.1@c5f3626 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.
- Counted embedding + head
411M parameters
Calculated: input embedding + output head, poolside/Laguna-XS-2.1@c5f3626 safetensors headers
Input embedding plus output head (once if tied), counted from the checkpoint.
- Non-emb. params (counted)
33.03B parameters
Calculated: dense + routed backbone weights; excludes input embedding, output head, lookup tables, MTP and other modalities, poolside/Laguna-XS-2.1@c5f3626 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.
- Largest layer (counted)
846.9M parameters
Calculated: max(sum of dense and routed weights sharing a full layers.<index> tensor path), all experts included, poolside/Laguna-XS-2.1@c5f3626 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.
- Active non-emb. (counted)
2.606B parameters
Calculated: non-expert weights + routed experts × 8/256, excluding embedding and output head, poolside/Laguna-XS-2.1@c5f3626 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.
Shape
- Layers
40
Source: config · /num_hidden_layers
Number of transformer blocks in the main backbone (attention or token mixer plus FFN), excluding MTP modules.
- d_model
2,048
Source: config · /hidden_size
Residual stream width.
- Vocab
100,352
Source: config · /vocab_size
Embedding table rows (vocab_size in config, may include padding).
- Tied embeddings
No
Source: config · /tie_word_embeddings
Input embedding and output head share weights.
- d_model / layers
51.2
Calculated: d_model / layers
Width-to-depth ratio, d_model divided by the number of blocks.
Attention
- Token mixers
GQA, SWA
Source: model-card · README.md; Highlights
enabling mixed SWA (Sliding Window Attention) and global attention layers in a 3:1 ratio
GQA: 8 KV heads for 48/64 Q-heads (config).
Token-mixing layer types in the backbone, e.g. GQA, MLA, SWA, Gated DeltaNet, Mamba-2, DSA (DeepSeek sparse attention).
- Layer layout
1 global : 3 SWA
Inferred from the source; see the explanation below.
Source: config · config.json; /layer_types
"layer_types": [ "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention" ]
The pinned /layer_types array repeats this four-layer pattern 10 times, beginning with the global layer. Ordered 40-layer backbone from /layer_types; auxiliary modules are excluded.
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.
- Global-attention layers
10
Source: config · /layer_types | count full_attention
Layers whose KV cache grows with full context (dense or sparse global attention, MLA). Excludes sliding-window and linear/recurrent layers.
- SWA : global
3
Calculated: attn.local_layers / attn.global_layers
attn.local_layers divided by attn.global_layers, the sliding-window to global interleave ratio (5 for '5 SWA : 1 global').
- Sliding-window layers
30
Source: config · /layer_types | count sliding_attention
Layers with a fixed attention window.
- Query heads
48
Source: config · /num_attention_heads
Global layers; SWA layers use 64 (attn.local_heads).
Query heads in the global-attention layers.
- KV heads
8
Source: config · /num_key_value_heads
Key/value heads in the global-attention layers. Not applicable to MLA, whose cache is a shared latent.
- Heads × head dim
6,144
Calculated: heads × head_dim
attn.heads × attn.head_dim, the query projection width of global layers.
- Q heads / KV head
6
Calculated: heads / kv_heads
attn.heads divided by attn.kv_heads in global layers.
- QK head dim
128
Source: config · /head_dim
Query/key dimension per head in global-attention layers (for MLA, nope + rope dims).
- V head dim
128
Source: config · /head_dim
Value dimension per head in global layers (MLA v_head_dim; stated for MLA even when equal to the QK head dim).
- Window
512 tokens
Source: config · /sliding_window
Sliding-window size of the local layers.
- SWA KV heads
8
Source: config · /num_key_value_heads
LagunaAttention builds k_proj/v_proj from config.num_key_value_heads and config.head_dim for both layer types.
KV heads in sliding-window layers (stated even when equal to attn.kv_heads).
- SWA query heads
64
Source: config · /num_attention_heads_per_layer/1
Layer 1 is sliding_attention; every sliding_attention entry is 64, every full_attention entry 48. modeling_laguna.py LagunaDecoderLayer.__init__ passes the per-layer count to LagunaAttention.
Query heads in sliding-window layers (stated even when equal to attn.heads).
- SWA QK head dim
128
Source: config · /head_dim
LagunaAttention uses config.head_dim for both layer types.
Query/key head dimension in sliding-window layers. Value head dimensions may differ; see attn.local_v_head_dim.
- Positional encoding
partial RoPE (global, 50%), RoPE (SWA)
Source: config · config.json; rope_parameters.full_attention
"partial_rotary_factor": 0.5
rope_parameters.sliding_attention has partial_rotary_factor 1.0.
Positional scheme in global layers: RoPE, partial RoPE, NoPE, or mixed per layer type.
- RoPE base
500,000
Source: config · /rope_parameters/full_attention/rope_theta
Global layers; SWA layers use 10,000.
RoPE base frequency of the global layers at release.
- SWA RoPE base
10,000
Source: config · /rope_parameters/sliding_attention/rope_theta
configuration_laguna.py derives swa_rope_parameters from rope_parameters["sliding_attention"]; LagunaModel builds swa_rotary_emb from it.
RoPE base frequency of the sliding-window layers (stated even when equal to attn.rope_theta).
- Rotary fraction
0.5
Source: config · /rope_parameters/full_attention/partial_rotary_factor
Global layers; SWA layers rotate all dims (/rope_parameters/sliding_attention/partial_rotary_factor = 1.0).
Fraction of each head's QK dimensions that are rotated.
- QK norm
Yes
Source: code · modeling_laguna.py; LagunaAttention.__init__
self.q_norm = LagunaRMSNorm(config.head_dim, eps=config.rms_norm_eps)
Queries and keys are normalized per head before the dot product.
- Output gate
softplus, per head, on attention output before o_proj
Source: code · modeling_laguna.py; LagunaAttention.forward
gate = F.softplus(self.g_proj(hidden_states).float()).to(attn_output.dtype)
Per head: config /gating = "per-head". The card says "sigmoid gating"; the code uses softplus.
Gate applied to the attention output before the output projection, with its placement.
- QKV bias
No
Source: code · modeling_laguna.py; LagunaAttention.__init__
self.q_proj = nn.Linear(config.hidden_size, self.num_heads * config.head_dim, bias=False)
Query/key/value projections have bias terms.
Feed-forward
- Activation
SwiGLU
Source: code · configuration_laguna.py; LagunaConfig.__init__
hidden_act: str = "silu"
config.json sets no hidden_act; LagunaMLP computes down_proj(act_fn(gate_proj(x)) * up_proj(x)).
FFN activation (SwiGLU, GeGLU, ...).
- Dense FFN dim
8,192
Source: config · /intermediate_size
Intermediate width of the dense FFN layers that are actually executed.
- FFN bias
No
Source: code · modeling_laguna.py; LagunaMLP.__init__
self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
FFN projections have bias terms.
- Dense FFN / d_model
4
Calculated: ffn.dim / d_model
ffn.dim divided by shape.d_model.
Mixture of experts
- Routed experts
256
Source: config · /num_experts
Routed experts per MoE layer.
- Experts per token
8
Source: config · /num_experts_per_tok
Routed experts selected per token.
- Expert dim
512
Source: config · /moe_intermediate_size
Intermediate width of one routed expert.
- Leading dense layers
1
Source: config · /mlp_only_layers | len
Layer 0; decoder_sparse_step 1 makes every other layer MoE (LagunaDecoderLayer.__init__).
Number of initial blocks that use a dense FFN instead of MoE.
- Router score
sigmoid
Source: code · modeling_laguna.py; LagunaTopKRouter
Laguna MoE router using sigmoid scoring (not softmax).
Gating function over expert logits (softmax, sigmoid).
- Load balancing
Not disclosed
Load-balancing method, with coefficients where disclosed.
- Routed scaling factor
2.5
Source: config · /moe_routed_scaling_factor
Multiplier applied to the combined routed-expert output (routed_scaling_factor).
- Renormalize top-k
Yes
Source: config · /norm_topk_prob
Selected expert weights are renormalized to sum to 1.
- Expert dim / d_model
0.25
Calculated: moe.expert_dim / d_model
moe.expert_dim divided by shape.d_model.
- Sparsity (experts / top-k)
32
Calculated: experts / top_k
Routed experts divided by experts selected per token.
- Active / total
0.0909091
Calculated: params.active / params.total
params.active divided by params.total.
Multi-token prediction
- MTP depth
Not disclosed
Additional future tokens predicted during pretraining (0 when the model was trained without MTP).
Normalization and stability
- Norm
RMSNorm
Source: code · modeling_laguna.py; LagunaDecoderLayer.__init__
self.input_layernorm = LagunaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
Normalization layer type in the blocks.
- Norm placement
pre
Source: code · modeling_laguna.py; LagunaDecoderLayer.forward
residual = hidden_states hidden_states = self.input_layernorm(hidden_states)
Pre-norm, post-norm, or sandwich (pre and post).
- Norm epsilon
1e-06
Source: config · /rms_norm_eps
Epsilon of the block normalization layers.
Context length
- Pretraining context
Not disclosed
Sequence length of the main pretraining stage.
- Trained context
Not disclosed
Longest sequence length trained on after context extension.
- Advertised context
262.1K tokens
Source: model-card · README.md; Model overview
Context window: 262,144 tokens
Longest context the publisher says the model supports, possibly beyond trained length via RoPE scaling.
- Extension method
YaRN factor 32 over original 8,192 on global layers (release config)
Source: config · config.json; rope_parameters.full_attention
"rope_type": "yarn", "factor": 32.0, "original_max_position_embeddings": 8192,
Training-time extension stages are not disclosed. Same dict: beta_fast 64, beta_slow 1, attention_factor 1.347; max_position_embeddings 262,144.
How context was extended (stages, RoPE base change, YaRN factor).
Pretraining recipe
- Pretraining tokens
Not disclosed
Tokens seen in pretraining including annealing and context extension, as reported.
- Optimizer
Muon
Source: model-card · README.md; Model overview
Optimizer: Muon
Optimizer used for the bulk of parameters.
- Peak LR
Not disclosed
Peak learning rate of the main pretraining stage.
- Final LR
Not disclosed
Learning rate at the end of the decay.
- LR schedule
Not disclosed
Learning-rate schedule shape with phase boundaries.
- Warmup
Not disclosed
Warmup length in steps or tokens.
- Batch size
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.
- Weight decay
Not disclosed
Decoupled weight decay coefficient.
- Adam betas
Not disclosed
(beta1, beta2) of the Adam-family optimizer or its AdamW fallback.
- Grad clip
Not disclosed
Global gradient-norm clipping threshold.
- Training precision
Not disclosed
Matmul precision of pretraining (BF16, FP8 recipe, NVFP4 recipe) with the scaling granularity where disclosed. Never inferred from the released checkpoint dtype.
KV cache
- KV bytes / token
41,440 bytes
Calculated: kv.bytes_at_context / 131072 tokens; uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (2048 × 16/8) B/entry; SWA KV: 30 × (2048 × 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.
- Global KV bytes / token
40,960 bytes
Calculated: 40960 B / 1 tokens (Global KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (2048 × 16/8) B/entry; SWA KV: 30 × (2048 × 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.
- KV bytes / token / B active
13,813 bytes
Calculated: kv.bytes_per_token (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (2048 × 16/8) B/entry; SWA KV: 30 × (2048 × 16/8) B/entry
kv.bytes_per_token divided by active parameters in billions (131,072 tokens in exports).
- KV at context / B active
1,810,541,227 bytes
Calculated: kv.bytes_at_context (131072 tokens) / (params.active / 1e9); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (2048 × 16/8) B/entry; SWA KV: 30 × (2048 × 16/8) B/entry
kv.bytes_at_context divided by active parameters in billions (131,072 tokens in exports).
- KV at context
5,431,623,680 bytes
Calculated: 40960 B × 131072 (Global KV) + 122880 B × min(131072, 512) (SWA KV); uniform BF16, payload only; quantization overhead excluded. Global KV: 10 × (2048 × 16/8) B/entry; SWA KV: 30 × (2048 × 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.