nemotron-3.5-bolt:e4b

9,522 21 minutes ago

Nemotron 3.5 Bolt is NVIDIA’s compact vision-language model for local coding, reasoning, and tool use.

vision tools thinking audio e4b
... /
draft
8103aa42fd31 · 737MB
    Metadata
  • general.architecture
    dflash
  • general.file_type
    BF16
  • dflash.attention.causal
    false
  • dflash.attention.head_count
    32
  • dflash.attention.head_count_kv
    8
  • dflash.attention.key_length
    128
  • dflash.attention.layer_norm_rms_epsilon
    1e-06
  • dflash.attention.sliding_window
    1024
  • dflash.attention.sliding_window_pattern
    [true, true, true, true, true]
  • dflash.attention.value_length
    128
  • dflash.block_count
    5
  • dflash.block_size
    8
  • dflash.context_length
    131072
  • dflash.embedding_length
    2560
  • dflash.embedding_scale
    50.596443
  • dflash.feed_forward_length
    5120
  • dflash.rope.freq_base
    1e+06
  • dflash.target_layers
    [6, 12, 18, 24, 36]
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_space_prefix
    false
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.mask_token_id
    4
  • tokenizer.ggml.merges
    [ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
  • tokenizer.ggml.model
    gemma4
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.scores
    [-1000, -1000, -1000, -1000, -1000, ...]
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
    • Tensor
    blk.0
  • blk.0.attn_k.weight
    BF16
    [2560, 1024]
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [2560]
  • blk.0.attn_output.weight
    BF16
    [4096, 2560]
  • blk.0.attn_q.weight
    BF16
    [2560, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_v.weight
    BF16
    [2560, 1024]
  • blk.0.ffn_down.weight
    BF16
    [5120, 2560]
  • blk.0.ffn_gate.weight
    BF16
    [2560, 5120]
  • blk.0.ffn_norm.weight
    F32
    [2560]
  • blk.0.ffn_up.weight
    BF16
    [2560, 5120]
    • blk.1
  • blk.1.attn_k.weight
    BF16
    [2560, 1024]
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [2560]
  • blk.1.attn_output.weight
    BF16
    [4096, 2560]
  • blk.1.attn_q.weight
    BF16
    [2560, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_v.weight
    BF16
    [2560, 1024]
  • blk.1.ffn_down.weight
    BF16
    [5120, 2560]
  • blk.1.ffn_gate.weight
    BF16
    [2560, 5120]
  • blk.1.ffn_norm.weight
    F32
    [2560]
  • blk.1.ffn_up.weight
    BF16
    [2560, 5120]
    • blk.2
  • blk.2.attn_k.weight
    BF16
    [2560, 1024]
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [2560]
  • blk.2.attn_output.weight
    BF16
    [4096, 2560]
  • blk.2.attn_q.weight
    BF16
    [2560, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_v.weight
    BF16
    [2560, 1024]
  • blk.2.ffn_down.weight
    BF16
    [5120, 2560]
  • blk.2.ffn_gate.weight
    BF16
    [2560, 5120]
  • blk.2.ffn_norm.weight
    F32
    [2560]
  • blk.2.ffn_up.weight
    BF16
    [2560, 5120]
    • blk.3
  • blk.3.attn_k.weight
    BF16
    [2560, 1024]
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [2560]
  • blk.3.attn_output.weight
    BF16
    [4096, 2560]
  • blk.3.attn_q.weight
    BF16
    [2560, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_v.weight
    BF16
    [2560, 1024]
  • blk.3.ffn_down.weight
    BF16
    [5120, 2560]
  • blk.3.ffn_gate.weight
    BF16
    [2560, 5120]
  • blk.3.ffn_norm.weight
    F32
    [2560]
  • blk.3.ffn_up.weight
    BF16
    [2560, 5120]
    • blk.4
  • blk.4.attn_k.weight
    BF16
    [2560, 1024]
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [2560]
  • blk.4.attn_output.weight
    BF16
    [4096, 2560]
  • blk.4.attn_q.weight
    BF16
    [2560, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_v.weight
    BF16
    [2560, 1024]
  • blk.4.ffn_down.weight
    BF16
    [5120, 2560]
  • blk.4.ffn_gate.weight
    BF16
    [2560, 5120]
  • blk.4.ffn_norm.weight
    F32
    [2560]
  • blk.4.ffn_up.weight
    BF16
    [2560, 5120]
  • enc.output_norm.weight
    F32
    [2560]
  • fc.weight
    BF16
    [12800, 2560]
  • output_norm.weight
    F32
    [2560]