MedAIBase/ AntAngelMed:100b

64 2 weeks ago

AntAngelMed是由浙江省卫生健康信息中心、蚂蚁健康、浙江省安诊儿医学人工智能科技有限公司共同联合开发的，迄今为止最大且最强大的开源医疗语言模型。

100b
... /
model
59059f0b4967 · 109GB
    Metadata
  • general.architecture
    bailingmoe2
  • general.file_type
    Q8_0
  • bailingmoe2.attention.head_count
    32
  • bailingmoe2.attention.head_count_kv
    4
  • bailingmoe2.attention.key_length
    128
  • bailingmoe2.attention.layer_norm_rms_epsilon
    1e-06
  • bailingmoe2.attention.value_length
    128
  • bailingmoe2.block_count
    32
  • bailingmoe2.context_length
    32768
  • bailingmoe2.embedding_length
    4096
  • bailingmoe2.expert_count
    256
  • bailingmoe2.expert_feed_forward_length
    1024
  • bailingmoe2.expert_gating_func
    2
  • bailingmoe2.expert_group_count
    8
  • bailingmoe2.expert_group_used_count
    4
  • bailingmoe2.expert_shared_count
    1
  • bailingmoe2.expert_shared_feed_forward_length
    1024
  • bailingmoe2.expert_used_count
    8
  • bailingmoe2.expert_weights_norm
    true
  • bailingmoe2.expert_weights_scale
    2.5
  • bailingmoe2.feed_forward_length
    9216
  • bailingmoe2.leading_dense_block_count
    1
  • bailingmoe2.rope.dimension_count
    64
  • bailingmoe2.rope.freq_base
    600000
  • bailingmoe2.vocab_size
    157184
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    156891
  • tokenizer.ggml.eos_token_id
    156892
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ t, i n, Ġ a, h e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    156892
  • tokenizer.ggml.pre
    bailingmoe2
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [4096, 157184]
    • blk.0
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [4096]
  • blk.0.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.0.ffn_down.weight
    Q8_0
    [9216, 4096]
  • blk.0.ffn_gate.weight
    Q8_0
    [4096, 9216]
  • blk.0.ffn_norm.weight
    F32
    [4096]
  • blk.0.ffn_up.weight
    Q8_0
    [4096, 9216]
    • blk.1
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [4096]
  • blk.1.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.1.exp_probs_b.bias
    F32
    [256]
  • blk.1.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.1.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.1.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.1.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.1.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.1.ffn_norm.weight
    F32
    [4096]
  • blk.1.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.1.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.2
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [4096]
  • blk.2.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.2.exp_probs_b.bias
    F32
    [256]
  • blk.2.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.2.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.2.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.2.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.2.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.2.ffn_norm.weight
    F32
    [4096]
  • blk.2.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.2.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.3
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [4096]
  • blk.3.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.3.exp_probs_b.bias
    F32
    [256]
  • blk.3.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.3.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.3.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.3.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.3.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.3.ffn_norm.weight
    F32
    [4096]
  • blk.3.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.3.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.4
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [4096]
  • blk.4.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.4.exp_probs_b.bias
    F32
    [256]
  • blk.4.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.4.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.4.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.4.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.4.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.4.ffn_norm.weight
    F32
    [4096]
  • blk.4.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.4.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.5
  • blk.5.attn_k_norm.weight
    F32
    [128]
  • blk.5.attn_norm.weight
    F32
    [4096]
  • blk.5.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.5.attn_q_norm.weight
    F32
    [128]
  • blk.5.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.5.exp_probs_b.bias
    F32
    [256]
  • blk.5.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.5.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.5.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.5.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.5.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.5.ffn_norm.weight
    F32
    [4096]
  • blk.5.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.5.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.6
  • blk.6.attn_k_norm.weight
    F32
    [128]
  • blk.6.attn_norm.weight
    F32
    [4096]
  • blk.6.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.6.attn_q_norm.weight
    F32
    [128]
  • blk.6.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.6.exp_probs_b.bias
    F32
    [256]
  • blk.6.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.6.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.6.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.6.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.6.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.6.ffn_norm.weight
    F32
    [4096]
  • blk.6.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.6.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.7
  • blk.7.attn_k_norm.weight
    F32
    [128]
  • blk.7.attn_norm.weight
    F32
    [4096]
  • blk.7.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.7.attn_q_norm.weight
    F32
    [128]
  • blk.7.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.7.exp_probs_b.bias
    F32
    [256]
  • blk.7.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.7.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.7.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.7.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.7.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.7.ffn_norm.weight
    F32
    [4096]
  • blk.7.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.7.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.8
  • blk.8.attn_k_norm.weight
    F32
    [128]
  • blk.8.attn_norm.weight
    F32
    [4096]
  • blk.8.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.8.attn_q_norm.weight
    F32
    [128]
  • blk.8.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.8.exp_probs_b.bias
    F32
    [256]
  • blk.8.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.8.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.8.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.8.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.8.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.8.ffn_norm.weight
    F32
    [4096]
  • blk.8.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.8.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.9
  • blk.9.attn_k_norm.weight
    F32
    [128]
  • blk.9.attn_norm.weight
    F32
    [4096]
  • blk.9.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.9.attn_q_norm.weight
    F32
    [128]
  • blk.9.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.9.exp_probs_b.bias
    F32
    [256]
  • blk.9.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.9.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.9.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.9.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.9.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.9.ffn_norm.weight
    F32
    [4096]
  • blk.9.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.9.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.10
  • blk.10.attn_k_norm.weight
    F32
    [128]
  • blk.10.attn_norm.weight
    F32
    [4096]
  • blk.10.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.10.attn_q_norm.weight
    F32
    [128]
  • blk.10.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.10.exp_probs_b.bias
    F32
    [256]
  • blk.10.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.10.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.10.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.10.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.10.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.10.ffn_norm.weight
    F32
    [4096]
  • blk.10.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.10.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.11
  • blk.11.attn_k_norm.weight
    F32
    [128]
  • blk.11.attn_norm.weight
    F32
    [4096]
  • blk.11.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.11.attn_q_norm.weight
    F32
    [128]
  • blk.11.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.11.exp_probs_b.bias
    F32
    [256]
  • blk.11.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.11.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.11.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.11.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.11.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.11.ffn_norm.weight
    F32
    [4096]
  • blk.11.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.11.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.12
  • blk.12.attn_k_norm.weight
    F32
    [128]
  • blk.12.attn_norm.weight
    F32
    [4096]
  • blk.12.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.12.attn_q_norm.weight
    F32
    [128]
  • blk.12.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.12.exp_probs_b.bias
    F32
    [256]
  • blk.12.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.12.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.12.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.12.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.12.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.12.ffn_norm.weight
    F32
    [4096]
  • blk.12.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.12.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.13
  • blk.13.attn_k_norm.weight
    F32
    [128]
  • blk.13.attn_norm.weight
    F32
    [4096]
  • blk.13.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.13.attn_q_norm.weight
    F32
    [128]
  • blk.13.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.13.exp_probs_b.bias
    F32
    [256]
  • blk.13.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.13.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.13.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.13.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.13.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.13.ffn_norm.weight
    F32
    [4096]
  • blk.13.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.13.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.14
  • blk.14.attn_k_norm.weight
    F32
    [128]
  • blk.14.attn_norm.weight
    F32
    [4096]
  • blk.14.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.14.attn_q_norm.weight
    F32
    [128]
  • blk.14.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.14.exp_probs_b.bias
    F32
    [256]
  • blk.14.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.14.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.14.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.14.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.14.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.14.ffn_norm.weight
    F32
    [4096]
  • blk.14.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.14.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.15
  • blk.15.attn_k_norm.weight
    F32
    [128]
  • blk.15.attn_norm.weight
    F32
    [4096]
  • blk.15.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.15.attn_q_norm.weight
    F32
    [128]
  • blk.15.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.15.exp_probs_b.bias
    F32
    [256]
  • blk.15.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.15.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.15.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.15.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.15.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.15.ffn_norm.weight
    F32
    [4096]
  • blk.15.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.15.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.16
  • blk.16.attn_k_norm.weight
    F32
    [128]
  • blk.16.attn_norm.weight
    F32
    [4096]
  • blk.16.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.16.attn_q_norm.weight
    F32
    [128]
  • blk.16.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.16.exp_probs_b.bias
    F32
    [256]
  • blk.16.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.16.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.16.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.16.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.16.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.16.ffn_norm.weight
    F32
    [4096]
  • blk.16.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.16.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.17
  • blk.17.attn_k_norm.weight
    F32
    [128]
  • blk.17.attn_norm.weight
    F32
    [4096]
  • blk.17.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.17.attn_q_norm.weight
    F32
    [128]
  • blk.17.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.17.exp_probs_b.bias
    F32
    [256]
  • blk.17.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.17.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.17.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.17.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.17.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.17.ffn_norm.weight
    F32
    [4096]
  • blk.17.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.17.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.18
  • blk.18.attn_k_norm.weight
    F32
    [128]
  • blk.18.attn_norm.weight
    F32
    [4096]
  • blk.18.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.18.attn_q_norm.weight
    F32
    [128]
  • blk.18.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.18.exp_probs_b.bias
    F32
    [256]
  • blk.18.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.18.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.18.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.18.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.18.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.18.ffn_norm.weight
    F32
    [4096]
  • blk.18.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.18.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.19
  • blk.19.attn_k_norm.weight
    F32
    [128]
  • blk.19.attn_norm.weight
    F32
    [4096]
  • blk.19.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.19.attn_q_norm.weight
    F32
    [128]
  • blk.19.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.19.exp_probs_b.bias
    F32
    [256]
  • blk.19.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.19.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.19.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.19.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.19.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.19.ffn_norm.weight
    F32
    [4096]
  • blk.19.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.19.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.20
  • blk.20.attn_k_norm.weight
    F32
    [128]
  • blk.20.attn_norm.weight
    F32
    [4096]
  • blk.20.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.20.attn_q_norm.weight
    F32
    [128]
  • blk.20.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.20.exp_probs_b.bias
    F32
    [256]
  • blk.20.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.20.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.20.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.20.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.20.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.20.ffn_norm.weight
    F32
    [4096]
  • blk.20.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.20.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.21
  • blk.21.attn_k_norm.weight
    F32
    [128]
  • blk.21.attn_norm.weight
    F32
    [4096]
  • blk.21.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.21.attn_q_norm.weight
    F32
    [128]
  • blk.21.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.21.exp_probs_b.bias
    F32
    [256]
  • blk.21.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.21.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.21.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.21.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.21.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.21.ffn_norm.weight
    F32
    [4096]
  • blk.21.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.21.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.22
  • blk.22.attn_k_norm.weight
    F32
    [128]
  • blk.22.attn_norm.weight
    F32
    [4096]
  • blk.22.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.22.attn_q_norm.weight
    F32
    [128]
  • blk.22.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.22.exp_probs_b.bias
    F32
    [256]
  • blk.22.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.22.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.22.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.22.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.22.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.22.ffn_norm.weight
    F32
    [4096]
  • blk.22.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.22.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.23
  • blk.23.attn_k_norm.weight
    F32
    [128]
  • blk.23.attn_norm.weight
    F32
    [4096]
  • blk.23.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.23.attn_q_norm.weight
    F32
    [128]
  • blk.23.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.23.exp_probs_b.bias
    F32
    [256]
  • blk.23.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.23.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.23.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.23.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.23.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.23.ffn_norm.weight
    F32
    [4096]
  • blk.23.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.23.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.24
  • blk.24.attn_k_norm.weight
    F32
    [128]
  • blk.24.attn_norm.weight
    F32
    [4096]
  • blk.24.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.24.attn_q_norm.weight
    F32
    [128]
  • blk.24.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.24.exp_probs_b.bias
    F32
    [256]
  • blk.24.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.24.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.24.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.24.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.24.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.24.ffn_norm.weight
    F32
    [4096]
  • blk.24.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.24.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.25
  • blk.25.attn_k_norm.weight
    F32
    [128]
  • blk.25.attn_norm.weight
    F32
    [4096]
  • blk.25.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.25.attn_q_norm.weight
    F32
    [128]
  • blk.25.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.25.exp_probs_b.bias
    F32
    [256]
  • blk.25.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.25.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.25.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.25.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.25.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.25.ffn_norm.weight
    F32
    [4096]
  • blk.25.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.25.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.26
  • blk.26.attn_k_norm.weight
    F32
    [128]
  • blk.26.attn_norm.weight
    F32
    [4096]
  • blk.26.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.26.attn_q_norm.weight
    F32
    [128]
  • blk.26.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.26.exp_probs_b.bias
    F32
    [256]
  • blk.26.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.26.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.26.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.26.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.26.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.26.ffn_norm.weight
    F32
    [4096]
  • blk.26.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.26.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.27
  • blk.27.attn_k_norm.weight
    F32
    [128]
  • blk.27.attn_norm.weight
    F32
    [4096]
  • blk.27.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.27.attn_q_norm.weight
    F32
    [128]
  • blk.27.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.27.exp_probs_b.bias
    F32
    [256]
  • blk.27.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.27.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.27.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.27.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.27.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.27.ffn_norm.weight
    F32
    [4096]
  • blk.27.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.27.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.28
  • blk.28.attn_k_norm.weight
    F32
    [128]
  • blk.28.attn_norm.weight
    F32
    [4096]
  • blk.28.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.28.attn_q_norm.weight
    F32
    [128]
  • blk.28.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.28.exp_probs_b.bias
    F32
    [256]
  • blk.28.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.28.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.28.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.28.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.28.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.28.ffn_norm.weight
    F32
    [4096]
  • blk.28.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.28.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.29
  • blk.29.attn_k_norm.weight
    F32
    [128]
  • blk.29.attn_norm.weight
    F32
    [4096]
  • blk.29.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.29.attn_q_norm.weight
    F32
    [128]
  • blk.29.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.29.exp_probs_b.bias
    F32
    [256]
  • blk.29.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.29.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.29.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.29.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.29.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.29.ffn_norm.weight
    F32
    [4096]
  • blk.29.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.29.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.30
  • blk.30.attn_k_norm.weight
    F32
    [128]
  • blk.30.attn_norm.weight
    F32
    [4096]
  • blk.30.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.30.attn_q_norm.weight
    F32
    [128]
  • blk.30.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.30.exp_probs_b.bias
    F32
    [256]
  • blk.30.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.30.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.30.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.30.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.30.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.30.ffn_norm.weight
    F32
    [4096]
  • blk.30.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.30.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
    • blk.31
  • blk.31.attn_k_norm.weight
    F32
    [128]
  • blk.31.attn_norm.weight
    F32
    [4096]
  • blk.31.attn_output.weight
    Q8_0
    [4096, 4096]
  • blk.31.attn_q_norm.weight
    F32
    [128]
  • blk.31.attn_qkv.weight
    Q8_0
    [4096, 5120]
  • blk.31.exp_probs_b.bias
    F32
    [256]
  • blk.31.ffn_down_exps.weight
    Q8_0
    [1024, 4096, 256]
  • blk.31.ffn_down_shexp.weight
    Q8_0
    [1024, 4096]
  • blk.31.ffn_gate_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.31.ffn_gate_inp.weight
    F32
    [4096, 256]
  • blk.31.ffn_gate_shexp.weight
    Q8_0
    [4096, 1024]
  • blk.31.ffn_norm.weight
    F32
    [4096]
  • blk.31.ffn_up_exps.weight
    Q8_0
    [4096, 1024, 256]
  • blk.31.ffn_up_shexp.weight
    Q8_0
    [4096, 1024]
  • output.weight
    Q8_0
    [4096, 157184]
  • output_norm.weight
    F32
    [4096]