Meta · 2024-07 · Dense
Llama 3.1 70B
Llama 3.1 70B is a grouped-query attention (GQA) transformer released by Meta in 2024-07, with 80 layers, hidden size 8192 and a context window of 131,072 tokens.
Layer stack
Attention ×80 · Layers 80
Key facts
| Family | Meta |
|---|---|
| Released | 2024-07 |
| Params | 70.55 B |
| Context | 131,072 tokens |
| Attention | GQA (64:8) |
| Layers | 80 |
| Hidden | 8,192 |
| Heads | 64 |
| Vocab | 128,256 |
| Position | RoPE |
| Norm | RMSNorm |
| Activation | silu |
| Dtype | bf16 |
| Architecture class | LlamaForCausalLM |
Architecture overview
Attention
grouped-query attention (GQA) — 64 q-heads / 8 kv-heads. RoPE θ=500,000.
Feed-forward / MoE
It is a dense model: all 70.55B 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 Llama 2 70B
| Model | Llama 3.1 70B | Llama 2 70B | Ratio |
|---|---|---|---|
| model_type | llama | llama | ＝ |
| architectures | LlamaForCausalLM | LlamaForCausalLM | ＝ |
| hidden_size | 8192 | 8192 | ≈1 |
| num_hidden_layers | 80 | 80 | ≈1 |
| num_attention_heads | 64 | 64 | ≈1 |
| num_key_value_heads | 8 | 8 | ≈1 |
| intermediate_size | 28672 | 28672 | ≈1 |
| hidden_act | silu | silu | ＝ |
| rope_theta | 500000 | — | |
| max_position_embeddings | 131072 | 4096 | ×32.00 |
| rms_norm_eps | 0.00001 | 0.00001 | ≈1 |
| vocab_size | 128256 | 32000 | ×4.01 |
| torch_dtype | bfloat16 | float16 | ≠ |
| attention_bias | false | — | |
| attention_dropout | 0 | — | |
| bos_token_id | 128000 | 1 | ×128000.00 |
| eos_token_id | 128001×1 + 128008×1 + 128009×1 | 2 | ≠ |
| initializer_range | 0.02 | 0.02 | ≈1 |
| mlp_bias | false | — | |
| pretraining_tp | 1 | — | |
| tie_word_embeddings | false | false | ＝ |
| transformers_version | 4.42.3 | 4.31.0.dev0 | ≠ |
| use_cache | true | true | ＝ |
Most similar architectures
- Qwen2 72B (Qwen, 2024-06) 0.999
- Qwen2.5 72B (Qwen, 2024-09) 0.999
- Qwen3 32B (Qwen, 2025-04) 0.995
- Qwen3.6 27B (Qwen, 2026-04) 0.994
- Qwen3.8 27B (Qwen, 2026-08) 0.994
- Qwen3 8B (Qwen, 2025-04) 0.978
Similarity values range from 0 (no shared categorical features) to 1 (identical profiles).
Raw config fields
23 fields
| architectures | LlamaForCausalLM |
|---|---|
| attention_bias | false |
| attention_dropout | 0 |
| bos_token_id | 128000 |
| eos_token_id | 128001×1 + 128008×1 + 128009×1 |
| hidden_act | silu |
| hidden_size | 8192 |
| initializer_range | 0.02 |
| intermediate_size | 28672 |
| max_position_embeddings | 131072 |
| mlp_bias | false |
| model_type | llama |
| num_attention_heads | 64 |
| num_hidden_layers | 80 |
| num_key_value_heads | 8 |
| pretraining_tp | 1 |
| rms_norm_eps | 0.00001 |
| rope_theta | 500000 |
| tie_word_embeddings | false |
| torch_dtype | bfloat16 |
| transformers_version | 4.42.3 |
| use_cache | true |
| vocab_size | 128256 |