Meta · 2023-07 · Dense
Llama 2 70B
Llama 2 70B is a grouped-query attention (GQA) transformer released by Meta in 2023-07, with 80 layers, hidden size 8192 and a context window of 4,096 tokens.
Layer stack
Attention ×80 · Layers 80
Key facts
| Family | Meta |
|---|---|
| Released | 2023-07 |
| Params | 68.98 B |
| Context | 4,096 tokens |
| Attention | GQA (64:8) |
| Layers | 80 |
| Hidden | 8,192 |
| Heads | 64 |
| Vocab | 32,000 |
| Position | learned/absolute |
| Norm | RMSNorm |
| Activation | silu |
| Dtype | — |
| Architecture class | LlamaForCausalLM |
Architecture overview
Attention
grouped-query attention (GQA) — 64 q-heads / 8 kv-heads.
Feed-forward / MoE
It is a dense model: all 68.98B 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 1 7B
| Model | Llama 2 70B | LLaMA 1 7B | Ratio |
|---|---|---|---|
| model_type | llama | llama | ＝ |
| architectures | LlamaForCausalLM | LlamaForCausalLM | ＝ |
| hidden_size | 8192 | 4096 | ×2.00 |
| num_hidden_layers | 80 | 32 | ×2.50 |
| num_attention_heads | 64 | 32 | ×2.00 |
| num_key_value_heads | 8 | — | |
| intermediate_size | 28672 | 11008 | ×2.60 |
| hidden_act | silu | silu | ＝ |
| max_position_embeddings | 4096 | 2048 | ×2.00 |
| rms_norm_eps | 0.00001 | 0.000001 | ×10.00 |
| vocab_size | 32000 | 32000 | ≈1 |
| torch_dtype | float16 | float16 | ＝ |
| bos_token_id | 1 | 1 | ≈1 |
| eos_token_id | 2 | 2 | ≈1 |
| initializer_range | 0.02 | 0.02 | ≈1 |
| pad_token_id | 0 | 0 | ＝ |
| tie_word_embeddings | false | false | ＝ |
| transformers_version | 4.31.0.dev0 | 4.28.0.dev0 | ≠ |
| use_cache | true | true | ＝ |
Most similar architectures
- Qwen3.5 27B (Qwen, 2026-02) 0.982
- Qwen3.5 9B (Qwen, 2026-02) 0.956
- Qwen3.5 9B Base (Qwen, 2026-02) 0.956
- Qwen2 72B (Qwen, 2024-06) 0.902
- Qwen2.5 72B (Qwen, 2024-09) 0.902
- Llama 3.1 70B (Meta, 2024-07) 0.901
Similarity values range from 0 (no shared categorical features) to 1 (identical profiles).
Raw config fields
19 fields
| architectures | LlamaForCausalLM |
|---|---|
| bos_token_id | 1 |
| eos_token_id | 2 |
| hidden_act | silu |
| hidden_size | 8192 |
| initializer_range | 0.02 |
| intermediate_size | 28672 |
| max_position_embeddings | 4096 |
| model_type | llama |
| num_attention_heads | 64 |
| num_hidden_layers | 80 |
| num_key_value_heads | 8 |
| pad_token_id | 0 |
| rms_norm_eps | 0.00001 |
| tie_word_embeddings | false |
| torch_dtype | float16 |
| transformers_version | 4.31.0.dev0 |
| use_cache | true |
| vocab_size | 32000 |