sam860/
olmoe-1b-7b-0924:latest

127 10 months ago

An MoE model developed by allenai competitive with llama2

... /
model
e99875673941 · 3.0GB
    Metadata
  • general.architecture
    olmoe
  • general.file_type
    Q3_K_S
  • olmoe.attention.head_count
    16
  • olmoe.attention.head_count_kv
    16
  • olmoe.attention.layer_norm_rms_epsilon
    1e-05
  • olmoe.block_count
    16
  • olmoe.context_length
    4096
  • olmoe.embedding_length
    2048
  • olmoe.expert_count
    64
  • olmoe.expert_used_count
    8
  • olmoe.feed_forward_length
    1024
  • olmoe.rope.freq_base
    10000
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    50279
  • tokenizer.ggml.eos_token_id
    50279
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ t, Ġ a, h e, i n, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    50280
  • tokenizer.ggml.pre
    olmo
  • tokenizer.ggml.token_type
    [4, 3, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [|||IP_ADDRESS|||, <|padding|>, !, ", #, ...]
    • Tensor
  • token_embd.weight
    Q3_K
    [2048, 50304]
    • blk.0
  • blk.0.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.0.attn_k_norm.weight
    F32
    [2048]
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.0.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.0.attn_q_norm.weight
    F32
    [2048]
  • blk.0.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.0.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.0.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.0.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.1
  • blk.1.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.1.attn_k_norm.weight
    F32
    [2048]
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.1.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.1.attn_q_norm.weight
    F32
    [2048]
  • blk.1.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.1.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.1.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.1.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.2
  • blk.2.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.2.attn_k_norm.weight
    F32
    [2048]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.2.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.2.attn_q_norm.weight
    F32
    [2048]
  • blk.2.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.2.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.2.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.2.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.3
  • blk.3.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.3.attn_k_norm.weight
    F32
    [2048]
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.3.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.3.attn_q_norm.weight
    F32
    [2048]
  • blk.3.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.3.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.3.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.3.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.4
  • blk.4.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.4.attn_k_norm.weight
    F32
    [2048]
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.4.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.4.attn_q_norm.weight
    F32
    [2048]
  • blk.4.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.4.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.4.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.4.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.5
  • blk.5.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.5.attn_k_norm.weight
    F32
    [2048]
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.5.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.5.attn_q_norm.weight
    F32
    [2048]
  • blk.5.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.5.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.5.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.5.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.6
  • blk.6.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.6.attn_k_norm.weight
    F32
    [2048]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.6.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.6.attn_q_norm.weight
    F32
    [2048]
  • blk.6.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.6.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.6.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.6.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.7
  • blk.7.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.7.attn_k_norm.weight
    F32
    [2048]
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.7.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.7.attn_q_norm.weight
    F32
    [2048]
  • blk.7.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.7.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.7.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.7.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.8
  • blk.8.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.8.attn_k_norm.weight
    F32
    [2048]
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.8.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.8.attn_q_norm.weight
    F32
    [2048]
  • blk.8.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.8.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.8.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.8.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.9
  • blk.9.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.9.attn_k_norm.weight
    F32
    [2048]
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.9.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.9.attn_q_norm.weight
    F32
    [2048]
  • blk.9.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.9.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.9.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.9.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.10
  • blk.10.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.10.attn_k_norm.weight
    F32
    [2048]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.10.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.10.attn_q_norm.weight
    F32
    [2048]
  • blk.10.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.10.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.10.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.10.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.11
  • blk.11.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.11.attn_k_norm.weight
    F32
    [2048]
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.11.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.11.attn_q_norm.weight
    F32
    [2048]
  • blk.11.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.11.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.11.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.11.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.12
  • blk.12.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.12.attn_k_norm.weight
    F32
    [2048]
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.12.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.12.attn_q_norm.weight
    F32
    [2048]
  • blk.12.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.12.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.12.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.12.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.13
  • blk.13.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.13.attn_k_norm.weight
    F32
    [2048]
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.13.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.13.attn_q_norm.weight
    F32
    [2048]
  • blk.13.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.13.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.13.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.13.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.14
  • blk.14.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.14.attn_k_norm.weight
    F32
    [2048]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.14.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.14.attn_q_norm.weight
    F32
    [2048]
  • blk.14.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.14.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.14.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.14.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
    • blk.15
  • blk.15.attn_k.weight
    Q3_K
    [2048, 2048]
  • blk.15.attn_k_norm.weight
    F32
    [2048]
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.attn_output.weight
    Q3_K
    [2048, 2048]
  • blk.15.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.15.attn_q_norm.weight
    F32
    [2048]
  • blk.15.attn_v.weight
    Q3_K
    [2048, 2048]
  • blk.15.ffn_down_exps.weight
    Q3_K
    [1024, 2048, 64]
  • blk.15.ffn_gate_exps.weight
    Q3_K
    [2048, 1024, 64]
  • blk.15.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up_exps.weight
    Q3_K
    [2048, 1024, 64]
  • output.weight
    Q6_K
    [2048, 50304]
  • output_norm.weight
    F32
    [2048]