nemotron-3.5-bolt:74b-a9b-iq2

2 19 minutes ago

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

vision tools thinking
... /
draft
080e1ea41c95 · 998MB
    Metadata
  • general.architecture
    dflash
  • general.file_type
    (!unknown_file_type 39!)
  • dflash.attention.causal
    false
  • dflash.attention.head_count
    32
  • dflash.attention.head_count_kv
    2
  • 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
    262144
  • dflash.conv_group_size
    16
  • dflash.conv_kernel_size
    2
  • dflash.embedding_length
    4096
  • dflash.feed_forward_length
    12288
  • dflash.rope.freq_base
    10000
  • dflash.selector_rank
    0
  • dflash.selector_top_k
    0
  • dflash.target_layers
    [2, 19, 36, 52, 69, ...]
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    11
  • tokenizer.ggml.mask_token_id
    990
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ t, e r, i n, Ġ ĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    11
  • tokenizer.ggml.pre
    pixtral
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, [INST], [/INST], ...]
  • tokenizer.ggml.unknown_token_id
    0
  • lilicorr.candidate_topk
    8
  • lilicorr.factor_dim
    1024
  • lilicorr.hidden_size
    1024
  • lilicorr.logit_scale
    8
  • lilicorr.mlp_hidden_size
    2048
  • lilicorr.num_heads
    8
  • lilicorr.num_layers
    2
  • lilicorr.vector_eps
    0.0001
    • Tensor
    blk.0
  • blk.0.attn_conv_base
    F32
    [4096, 2, 2]
  • blk.0.attn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.0.attn_k.weight
    BF16
    [4096, 256]
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [4096]
  • blk.0.attn_output.weight
    BF16
    [4096, 4096]
  • blk.0.attn_q.weight
    BF16
    [4096, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_sinks.weight
    F32
    [32]
  • blk.0.attn_v.weight
    BF16
    [4096, 256]
  • blk.0.ffn_conv_base
    F32
    [4096, 2, 2]
  • blk.0.ffn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.0.ffn_down.scale
    F32
    [1]
  • blk.0.ffn_down.weight
    (!unknown_type 40!)
    [12288, 4096]
  • blk.0.ffn_gate.scale
    F32
    [1]
  • blk.0.ffn_gate.weight
    (!unknown_type 40!)
    [4096, 12288]
  • blk.0.ffn_norm.weight
    F32
    [4096]
  • blk.0.ffn_up.scale
    F32
    [1]
  • blk.0.ffn_up.weight
    (!unknown_type 40!)
    [4096, 12288]
    • blk.1
  • blk.1.attn_conv_base
    F32
    [4096, 2, 2]
  • blk.1.attn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.1.attn_k.weight
    BF16
    [4096, 256]
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [4096]
  • blk.1.attn_output.weight
    BF16
    [4096, 4096]
  • blk.1.attn_q.weight
    BF16
    [4096, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_sinks.weight
    F32
    [32]
  • blk.1.attn_v.weight
    BF16
    [4096, 256]
  • blk.1.ffn_conv_base
    F32
    [4096, 2, 2]
  • blk.1.ffn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.1.ffn_down.scale
    F32
    [1]
  • blk.1.ffn_down.weight
    (!unknown_type 40!)
    [12288, 4096]
  • blk.1.ffn_gate.scale
    F32
    [1]
  • blk.1.ffn_gate.weight
    (!unknown_type 40!)
    [4096, 12288]
  • blk.1.ffn_norm.weight
    F32
    [4096]
  • blk.1.ffn_up.scale
    F32
    [1]
  • blk.1.ffn_up.weight
    (!unknown_type 40!)
    [4096, 12288]
    • blk.2
  • blk.2.attn_conv_base
    F32
    [4096, 2, 2]
  • blk.2.attn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.2.attn_k.weight
    BF16
    [4096, 256]
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [4096]
  • blk.2.attn_output.weight
    BF16
    [4096, 4096]
  • blk.2.attn_q.weight
    BF16
    [4096, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_sinks.weight
    F32
    [32]
  • blk.2.attn_v.weight
    BF16
    [4096, 256]
  • blk.2.ffn_conv_base
    F32
    [4096, 2, 2]
  • blk.2.ffn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.2.ffn_down.scale
    F32
    [1]
  • blk.2.ffn_down.weight
    (!unknown_type 40!)
    [12288, 4096]
  • blk.2.ffn_gate.scale
    F32
    [1]
  • blk.2.ffn_gate.weight
    (!unknown_type 40!)
    [4096, 12288]
  • blk.2.ffn_norm.weight
    F32
    [4096]
  • blk.2.ffn_up.scale
    F32
    [1]
  • blk.2.ffn_up.weight
    (!unknown_type 40!)
    [4096, 12288]
    • blk.3
  • blk.3.attn_conv_base
    F32
    [4096, 2, 2]
  • blk.3.attn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.3.attn_k.weight
    BF16
    [4096, 256]
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [4096]
  • blk.3.attn_output.weight
    BF16
    [4096, 4096]
  • blk.3.attn_q.weight
    BF16
    [4096, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_sinks.weight
    F32
    [32]
  • blk.3.attn_v.weight
    BF16
    [4096, 256]
  • blk.3.ffn_conv_base
    F32
    [4096, 2, 2]
  • blk.3.ffn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.3.ffn_down.scale
    F32
    [1]
  • blk.3.ffn_down.weight
    (!unknown_type 40!)
    [12288, 4096]
  • blk.3.ffn_gate.scale
    F32
    [1]
  • blk.3.ffn_gate.weight
    (!unknown_type 40!)
    [4096, 12288]
  • blk.3.ffn_norm.weight
    F32
    [4096]
  • blk.3.ffn_up.scale
    F32
    [1]
  • blk.3.ffn_up.weight
    (!unknown_type 40!)
    [4096, 12288]
    • blk.4
  • blk.4.attn_conv_base
    F32
    [4096, 2, 2]
  • blk.4.attn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.4.attn_k.weight
    BF16
    [4096, 256]
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [4096]
  • blk.4.attn_output.weight
    BF16
    [4096, 4096]
  • blk.4.attn_q.weight
    BF16
    [4096, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_sinks.weight
    F32
    [32]
  • blk.4.attn_v.weight
    BF16
    [4096, 256]
  • blk.4.ffn_conv_base
    F32
    [4096, 2, 2]
  • blk.4.ffn_conv_proj.weight
    BF16
    [4096, 1024]
  • blk.4.ffn_down.scale
    F32
    [1]
  • blk.4.ffn_down.weight
    (!unknown_type 40!)
    [12288, 4096]
  • blk.4.ffn_gate.scale
    F32
    [1]
  • blk.4.ffn_gate.weight
    (!unknown_type 40!)
    [4096, 12288]
  • blk.4.ffn_norm.weight
    F32
    [4096]
  • blk.4.ffn_up.scale
    F32
    [1]
  • blk.4.ffn_up.weight
    (!unknown_type 40!)
    [4096, 12288]
    • blk.5
  • blk.5.attn_norm.weight
    F32
    [1024]
  • blk.5.attn_output.bias
    F32
    [1024]
  • blk.5.attn_output.scale
    F32
    [1]
  • blk.5.attn_output.weight
    (!unknown_type 40!)
    [1024, 1024]
  • blk.5.ffn_down.bias
    F32
    [1024]
  • blk.5.ffn_down.scale
    F32
    [1]
  • blk.5.ffn_down.weight
    (!unknown_type 40!)
    [2048, 1024]
  • blk.5.ffn_norm.weight
    F32
    [1024]
  • blk.5.ffn_up.bias
    F32
    [2048]
  • blk.5.ffn_up.scale
    F32
    [1]
  • blk.5.ffn_up.weight
    (!unknown_type 40!)
    [1024, 2048]
    • blk.6
  • blk.6.attn_norm.weight
    F32
    [1024]
  • blk.6.attn_output.bias
    F32
    [1024]
  • blk.6.attn_output.scale
    F32
    [1]
  • blk.6.attn_output.weight
    (!unknown_type 40!)
    [1024, 1024]
  • blk.6.ffn_down.bias
    F32
    [1024]
  • blk.6.ffn_down.scale
    F32
    [1]
  • blk.6.ffn_down.weight
    (!unknown_type 40!)
    [2048, 1024]
  • blk.6.ffn_norm.weight
    F32
    [1024]
  • blk.6.ffn_up.bias
    F32
    [2048]
  • blk.6.ffn_up.scale
    F32
    [1]
  • blk.6.ffn_up.weight
    (!unknown_type 40!)
    [1024, 2048]
    • blk.7
  • blk.7.ffn_down.bias
    F32
    [1024]
  • blk.7.ffn_down.weight
    F32
    [1024, 1024]
  • blk.7.ffn_norm.bias
    F32
    [5]
  • blk.7.ffn_norm.weight
    F32
    [5]
  • blk.7.ffn_up.bias
    F32
    [1024]
  • blk.7.ffn_up.weight
    F32
    [5, 1024]
  • enc.output_norm.weight
    F32
    [4096]
  • fc.scale
    F32
    [1]
  • fc.weight
    (!unknown_type 40!)
    [24576, 4096]
  • lilicorr.anchor_norm.weight
    F32
    [1024]
  • lilicorr.anchor_out_head.bias
    F32
    [1024]
  • lilicorr.anchor_out_head.weight
    F32
    [1024, 1024]
  • lilicorr.attention_bias.bias
    F32
    [56, 56, 8]
  • lilicorr.context_proj.bias
    F32
    [1024]
  • lilicorr.context_proj.scale
    F32
    [1]
  • lilicorr.context_proj.weight
    (!unknown_type 40!)
    [4096, 1024]
  • lilicorr.factor_input_proj.bias
    F32
    [1024]
  • lilicorr.factor_input_proj.weight
    F32
    [3072, 1024]
  • lilicorr.in_head.bias
    F32
    [1024]
  • lilicorr.in_head.weight
    F32
    [1024, 1024]
  • lilicorr.layers.0.attn.in_proj.bias
    F32
    [3072]
  • lilicorr.layers.0.attn.in_proj.weight
    F32
    [1024, 3072]
  • lilicorr.layers.1.attn.in_proj.bias
    F32
    [3072]
  • lilicorr.layers.1.attn.in_proj.weight
    F32
    [1024, 3072]
  • lilicorr.out_head.bias
    F32
    [1024]
  • lilicorr.out_head.weight
    F32
    [1024, 1024]
  • lilicorr.output_norm.weight
    F32
    [1024]
  • lilicorr.pass_hidden_proj.bias
    F32
    [1024]
  • lilicorr.pass_hidden_proj.scale
    F32
    [1]
  • lilicorr.pass_hidden_proj.weight
    (!unknown_type 40!)
    [4096, 1024]
  • lilicorr.rank_embedding.weight
    F32
    [1024, 8]
  • lilicorr.slot_embedding.weight
    F32
    [1024, 7]
  • lilicorr.token_proj.bias
    F32
    [1024]
  • lilicorr.token_proj.scale
    F32
    [1]
  • lilicorr.token_proj.weight
    (!unknown_type 40!)
    [4096, 1024]
  • output_norm.weight
    F32
    [4096]