h2oGPT clone of Meta's Llama 2 70B Chat.
Try it live on our h2oGPT demo with side-by-side LLM comparisons and private document chat!
See how it compares to other models on our LLM Leaderboard!
See more at H2O.ai
Model Architecture
LlamaForCausalLM(
(model): LlamaModel(
(embed_tokens): Embedding(32000, 8192, padding_idx=0)
(layers): ModuleList(
(0-79): 80 x LlamaDecoderLayer(
(self_attn): LlamaAttention(
(q_proj): Linear4bit(in_features=8192, out_features=8192, bias=False)
(k_proj): Linear4bit(in_features=8192, out_features=1024, bias=False)
(v_proj): Linear4bit(in_features=8192, out_features=1024, bias=False)
(o_proj): Linear4bit(in_features=8192, out_features=8192, bias=False)
(rotary_emb): LlamaRotaryEmbedding()
)
(mlp): LlamaMLP(
(gate_proj): Linear4bit(in_features=8192, out_features=28672, bias=False)
(up_proj): Linear4bit(in_features=8192, out_features=28672, bias=False)
(down_proj): Linear4bit(in_features=28672, out_features=8192, bias=False)
(act_fn): SiLUActivation()
)
(input_layernorm): LlamaRMSNorm()
(post_attention_layernorm): LlamaRMSNorm()
)
)
(norm): LlamaRMSNorm()
)
(lm_head): Linear(in_features=8192, out_features=32000, bias=False)
)