Hunyuan · 2026-05 · Dense · Multimodal
Hunyuan-TurboS 7B
Hunyuan-TurboS 7B is a grouped-query attention (GQA) transformer released by Hunyuan in 2026-05, with 32 layers, hidden size 4096 and a context window of 262,144 tokens.
Layer stack
Attention ×32 · Layers 32
Key facts
| Family | Hunyuan |
|---|---|
| Released | 2026-05 |
| Params | 7.8 B |
| Context | 262,144 tokens |
| Attention | GQA (32:8) |
| Layers | 32 |
| Hidden | 4,096 |
| Heads | 32 |
| Vocab | 128,167 |
| Position | RoPE |
| Norm | RMSNorm |
| Activation | silu |
| Dtype | bf16 |
| Architecture class | HunYuanDenseV1ForCausalLM |
Architecture overview
Attention
grouped-query attention (GQA) — 32 q-heads / 8 kv-heads. RoPE θ=10,000.
Feed-forward / MoE
It is a dense model: all 7.8B parameters participate in computing every token.
Field-level comparison
Compared against the previous model of the same lab; where none exists, against the structurally closest model. The ratio column is this model divided by the comparison model.
Full-field comparison vs predecessor Hunyuan-TurboS 30B A3B
| Model | Hunyuan-TurboS 7B | Hunyuan-TurboS 30B A3B | Ratio |
|---|---|---|---|
| model_type | hunyuan_v1_dense | hy_v3 | ≠ |
| architectures | HunYuanDenseV1ForCausalLM | HYV3ForCausalLM | ≠ |
| hidden_size | 4096 | 2048 | ×2.00 |
| num_hidden_layers | 32 | 48 | ×0.667 |
| num_attention_heads | 32 | 32 | ≈1 |
| num_key_value_heads | 8 | 4 | ×2.00 |
| head_dim | 128 | 128 | ≈1 |
| intermediate_size | 14336 | 6912 | ×2.07 |
| hidden_act | silu | silu | ＝ |
| num_experts_per_tok | — | 8 | |
| moe_intermediate_size | — | 768 | |
| rope_theta | 10000 | 11158840 | ×0.001 |
| max_position_embeddings | 262144 | 262144 | ≈1 |
| rms_norm_eps | 0.00001 | 0.00001 | ≈1 |
| vocab_size | 128167 | 120832 | ×1.06 |
| torch_dtype | bfloat16 | — | |
| add_classification_head | false | — | |
| attention_bias | false | — | |
| attention_dropout | 0.1 | — | |
| attention_head_dim | 128 | — | |
| bos_token_id | 127958 | 120000 | ×1.07 |
| cla_share_factor | 2 | — | |
| class_num | 0 | — | |
| dense_list | 4096×1 + 0×1 | — | |
| eos_token_id | 127960 | 120025 | ×1.07 |
| im_end_id | 5 | — |
Most similar architectures
- Llama 3.1 8B (Meta, 2024-07) 1.000
- Qwen3 8B (Qwen, 2025-04) 0.998
- Mistral 7B (Mistral, 2023-09) 0.998
- Hunyuan 7B (Hunyuan, 2025-08) 0.998
- Qwen2.5 7B (Qwen, 2024-09) 0.998
- Qwen3.6 27B (Qwen, 2026-04) 0.992
Similarity values range from 0 (no shared categorical features) to 1 (identical profiles).
Raw config fields
44 fields
| add_classification_head | false |
|---|---|
| architectures | HunYuanDenseV1ForCausalLM |
| attention_bias | false |
| attention_dropout | 0.1 |
| attention_head_dim | 128 |
| bos_token_id | 127958 |
| cla_share_factor | 2 |
| class_num | 0 |
| dense_list | 4096×1 + 0×1 |
| torch_dtype | bfloat16 |
| eos_token_id | 127960 |
| head_dim | 128 |
| hidden_act | silu |
| hidden_size | 4096 |
| im_end_id | 5 |
| im_newline_id | 11 |
| im_start_id | 4 |
| initializer_range | 0.02 |
| intermediate_size | 14336 |
| mask_init_id | 12 |
| max_position_embeddings | 262144 |
| mlp_bias | false |
| model_type | hunyuan_v1_dense |
| norm_type | rms |
| num_attention_heads | 32 |
| num_hidden_layers | 32 |
| num_key_value_heads | 8 |
| org_vocab_size | 128167 |
| pad_id | 127961 |
| pad_token_id | 127961 |
| pool_type | last |
| pretraining_tp | 1 |
| rms_norm_eps | 0.00001 |
| rope_theta | 10000 |
| sep_token_id | 127962 |
| text_end_id | 7 |
| text_start_id | 6 |
| tie_word_embeddings | true |
| transformers_version | 4.57.6 |
| use_cache | true |
| use_cla | false |
| use_qk_norm | true |
| use_rotary_pos_emb | true |
| vocab_size | 128167 |