10 3 days ago

tools 8b 30b
93f4491fecea ยท 16GB
    Metadata
  • split.count
    2
  • split.no
    1
  • split.tensors.count
    1346
  • Tensor
    blk.46
  • blk.46.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.46.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.46.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.47
  • blk.47.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.47.attn_norm.weight
    F32
    [4096]
  • blk.47.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.47.attn_output.weight
    BF16
    [4096, 4096]
  • blk.47.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.47.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.47.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.47.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.47.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.47.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.47.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.47.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.47.ffn_norm.weight
    F32
    [4096]
  • blk.47.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.47.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.47.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.48
  • blk.48.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.48.attn_norm.weight
    F32
    [4096]
  • blk.48.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.48.attn_output.weight
    BF16
    [4096, 4096]
  • blk.48.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.48.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.48.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.48.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.48.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.48.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.48.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.48.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.48.ffn_norm.weight
    F32
    [4096]
  • blk.48.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.48.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.48.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.49
  • blk.49.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.49.attn_norm.weight
    F32
    [4096]
  • blk.49.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.49.attn_output.weight
    BF16
    [4096, 4096]
  • blk.49.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.49.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.49.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.49.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.49.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.49.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.49.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.49.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.49.ffn_norm.weight
    F32
    [4096]
  • blk.49.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.49.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.49.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.50
  • blk.50.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.50.attn_norm.weight
    F32
    [4096]
  • blk.50.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.50.attn_output.weight
    BF16
    [4096, 4096]
  • blk.50.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.50.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.50.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.50.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.50.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.50.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.50.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.50.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.50.ffn_norm.weight
    F32
    [4096]
  • blk.50.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.50.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.50.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.51
  • blk.51.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.51.attn_norm.weight
    F32
    [4096]
  • blk.51.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.51.attn_output.weight
    BF16
    [4096, 4096]
  • blk.51.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.51.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.51.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.51.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.51.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.51.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.51.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.51.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.51.ffn_norm.weight
    F32
    [4096]
  • blk.51.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.51.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.51.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.52
  • blk.52.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.52.attn_norm.weight
    F32
    [4096]
  • blk.52.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.52.attn_output.weight
    BF16
    [4096, 4096]
  • blk.52.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.52.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.52.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.52.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.52.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.52.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.52.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.52.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.52.ffn_norm.weight
    F32
    [4096]
  • blk.52.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.52.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.52.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.53
  • blk.53.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.53.attn_norm.weight
    F32
    [4096]
  • blk.53.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.53.attn_output.weight
    BF16
    [4096, 4096]
  • blk.53.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.53.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.53.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.53.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.53.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.53.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.53.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.53.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.53.ffn_norm.weight
    F32
    [4096]
  • blk.53.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.53.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.53.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.54
  • blk.54.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.54.attn_norm.weight
    F32
    [4096]
  • blk.54.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.54.attn_output.weight
    BF16
    [4096, 4096]
  • blk.54.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.54.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.54.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.54.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.54.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.54.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.54.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.54.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.54.ffn_norm.weight
    F32
    [4096]
  • blk.54.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.54.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.54.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.55
  • blk.55.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.55.attn_norm.weight
    F32
    [4096]
  • blk.55.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.55.attn_output.weight
    BF16
    [4096, 4096]
  • blk.55.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.55.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.55.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.55.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.55.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.55.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.55.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.55.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.55.ffn_norm.weight
    F32
    [4096]
  • blk.55.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.55.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.55.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.56
  • blk.56.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.56.attn_norm.weight
    F32
    [4096]
  • blk.56.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.56.attn_output.weight
    BF16
    [4096, 4096]
  • blk.56.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.56.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.56.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.56.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.56.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.56.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.56.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.56.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.56.ffn_norm.weight
    F32
    [4096]
  • blk.56.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.56.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.56.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.57
  • blk.57.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.57.attn_norm.weight
    F32
    [4096]
  • blk.57.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.57.attn_output.weight
    BF16
    [4096, 4096]
  • blk.57.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.57.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.57.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.57.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.57.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.57.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.57.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.57.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.57.ffn_norm.weight
    F32
    [4096]
  • blk.57.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.57.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.57.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.58
  • blk.58.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.58.attn_norm.weight
    F32
    [4096]
  • blk.58.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.58.attn_output.weight
    BF16
    [4096, 4096]
  • blk.58.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.58.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.58.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.58.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.58.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.58.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.58.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.58.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.58.ffn_norm.weight
    F32
    [4096]
  • blk.58.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.58.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.58.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.59
  • blk.59.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.59.attn_norm.weight
    F32
    [4096]
  • blk.59.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.59.attn_output.weight
    BF16
    [4096, 4096]
  • blk.59.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.59.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.59.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.59.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.59.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.59.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.59.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.59.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.59.ffn_norm.weight
    F32
    [4096]
  • blk.59.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.59.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.59.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.60
  • blk.60.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.60.attn_norm.weight
    F32
    [4096]
  • blk.60.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.60.attn_output.weight
    BF16
    [4096, 4096]
  • blk.60.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.60.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.60.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.60.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.60.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.60.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.60.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.60.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.60.ffn_norm.weight
    F32
    [4096]
  • blk.60.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.60.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.60.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.61
  • blk.61.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.61.attn_norm.weight
    F32
    [4096]
  • blk.61.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.61.attn_output.weight
    BF16
    [4096, 4096]
  • blk.61.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.61.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.61.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.61.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.61.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.61.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.61.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.61.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.61.ffn_norm.weight
    F32
    [4096]
  • blk.61.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.61.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.61.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.62
  • blk.62.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.62.attn_norm.weight
    F32
    [4096]
  • blk.62.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.62.attn_output.weight
    BF16
    [4096, 4096]
  • blk.62.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.62.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.62.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.62.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.62.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.62.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.62.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.62.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.62.ffn_norm.weight
    F32
    [4096]
  • blk.62.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.62.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.62.ffn_up.weight
    BF16
    [4096, 32768]
  • blk.63
  • blk.63.attn_k.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.attn_k.lora_b
    BF16
    [64, 1024, 3]
  • blk.63.attn_norm.weight
    F32
    [4096]
  • blk.63.attn_output.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.attn_output.lora_b
    BF16
    [64, 4096, 3]
  • blk.63.attn_output.weight
    BF16
    [4096, 4096]
  • blk.63.attn_q.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.attn_q.lora_b
    BF16
    [64, 4096, 3]
  • blk.63.attn_qkv.weight
    BF16
    [4096, 6144]
  • blk.63.attn_v.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.attn_v.lora_b
    BF16
    [64, 1024, 3]
  • blk.63.ffn_down.lora_a
    BF16
    [32768, 64, 3]
  • blk.63.ffn_down.lora_b
    BF16
    [64, 4096, 3]
  • blk.63.ffn_down.weight
    BF16
    [32768, 4096]
  • blk.63.ffn_gate.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.ffn_gate.lora_b
    BF16
    [64, 32768, 3]
  • blk.63.ffn_gate.weight
    BF16
    [4096, 32768]
  • blk.63.ffn_norm.weight
    F32
    [4096]
  • blk.63.ffn_up.lora_a
    BF16
    [4096, 64, 3]
  • blk.63.ffn_up.lora_b
    BF16
    [64, 32768, 3]
  • blk.63.ffn_up.weight
    BF16
    [4096, 32768]
  • output_norm.weight
    F32
    [4096]