odytrice/ muse:30b

46 2 months ago

Meta's Apache 2.0 agentic vision-language model, 30B at 4-bit with 131K context

vision tools 30b
... /
draft
27d9a805fa29 · 1.6GB
    Metadata
  • general.architecture
    dflash
  • general.file_type
    Q4_K_M
  • dflash.attention.head_count
    32
  • dflash.attention.head_count_kv
    8
  • dflash.attention.key_length
    128
  • dflash.attention.layer_norm_rms_epsilon
    1e-05
  • dflash.attention.sliding_window
    2048
  • dflash.attention.sliding_window_pattern
    [true, true, true, true, true]
  • dflash.attention.value_length
    128
  • dflash.block_count
    5
  • dflash.block_size
    16
  • dflash.context_length
    131072
  • dflash.embedding_length
    6656
  • dflash.feed_forward_length
    19968
  • dflash.rope.freq_base
    500000
  • dflash.target_layers
    [2, 14, 26, 38, 50]
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.bos_token_id
    200000
  • tokenizer.ggml.eos_token_id
    200001
  • tokenizer.ggml.eot_token_id
    200008
  • tokenizer.ggml.mask_token_id
    201818
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, e r, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    200018
  • tokenizer.ggml.pre
    llama4
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [À, Á, õ, ö, ÷, ...]
    • Tensor
    blk.0
  • blk.0.attn_k.weight
    Q4_K
    [6656, 1024]
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [6656]
  • blk.0.attn_output.weight
    Q4_K
    [4096, 6656]
  • blk.0.attn_q.weight
    Q4_K
    [6656, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_v.weight
    Q6_K
    [6656, 1024]
  • blk.0.ffn_down.weight
    Q6_K
    [19968, 6656]
  • blk.0.ffn_gate.weight
    Q4_K
    [6656, 19968]
  • blk.0.ffn_norm.weight
    F32
    [6656]
  • blk.0.ffn_up.weight
    Q4_K
    [6656, 19968]
    • blk.1
  • blk.1.attn_k.weight
    Q4_K
    [6656, 1024]
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [6656]
  • blk.1.attn_output.weight
    Q4_K
    [4096, 6656]
  • blk.1.attn_q.weight
    Q4_K
    [6656, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_v.weight
    Q6_K
    [6656, 1024]
  • blk.1.ffn_down.weight
    Q6_K
    [19968, 6656]
  • blk.1.ffn_gate.weight
    Q4_K
    [6656, 19968]
  • blk.1.ffn_norm.weight
    F32
    [6656]
  • blk.1.ffn_up.weight
    Q4_K
    [6656, 19968]
    • blk.2
  • blk.2.attn_k.weight
    Q4_K
    [6656, 1024]
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [6656]
  • blk.2.attn_output.weight
    Q4_K
    [4096, 6656]
  • blk.2.attn_q.weight
    Q4_K
    [6656, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_v.weight
    Q6_K
    [6656, 1024]
  • blk.2.ffn_down.weight
    Q6_K
    [19968, 6656]
  • blk.2.ffn_gate.weight
    Q4_K
    [6656, 19968]
  • blk.2.ffn_norm.weight
    F32
    [6656]
  • blk.2.ffn_up.weight
    Q4_K
    [6656, 19968]
    • blk.3
  • blk.3.attn_k.weight
    Q4_K
    [6656, 1024]
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [6656]
  • blk.3.attn_output.weight
    Q4_K
    [4096, 6656]
  • blk.3.attn_q.weight
    Q4_K
    [6656, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_v.weight
    Q6_K
    [6656, 1024]
  • blk.3.ffn_down.weight
    Q6_K
    [19968, 6656]
  • blk.3.ffn_gate.weight
    Q4_K
    [6656, 19968]
  • blk.3.ffn_norm.weight
    F32
    [6656]
  • blk.3.ffn_up.weight
    Q4_K
    [6656, 19968]
    • blk.4
  • blk.4.attn_k.weight
    Q4_K
    [6656, 1024]
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [6656]
  • blk.4.attn_output.weight
    Q4_K
    [4096, 6656]
  • blk.4.attn_q.weight
    Q4_K
    [6656, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_v.weight
    Q6_K
    [6656, 1024]
  • blk.4.ffn_down.weight
    Q6_K
    [19968, 6656]
  • blk.4.ffn_gate.weight
    Q4_K
    [6656, 19968]
  • blk.4.ffn_norm.weight
    F32
    [6656]
  • blk.4.ffn_up.weight
    Q4_K
    [6656, 19968]
  • enc.output_norm.weight
    F32
    [6656]
  • fc.weight
    Q4_K
    [33280, 6656]
  • output_norm.weight
    F32
    [6656]