Tromset/ flycoder0.3:router

17 1 week ago

vision tools thinking
... /
draft
b0108ebdcf5a · 303MB
    Metadata
  • general.architecture
    qwen35
  • general.file_type
    Q8_0
  • qwen35.attention.head_count
    8
  • qwen35.attention.head_count_kv
    2
  • qwen35.attention.key_length
    256
  • qwen35.attention.layer_norm_rms_epsilon
    1e-06
  • qwen35.attention.recurrent_layers
    [true, true, true, false, true, ...]
  • qwen35.attention.value_length
    256
  • qwen35.block_count
    25
  • qwen35.context_length
    262144
  • qwen35.embedding_length
    1024
  • qwen35.feed_forward_length
    3584
  • qwen35.full_attention_interval
    4
  • qwen35.nextn_predict_layers
    1
  • qwen35.rope.dimension_count
    64
  • qwen35.rope.dimension_sections
    [11, 11, 10, 0]
  • qwen35.rope.freq_base
    1e+07
  • qwen35.ssm.conv_kernel
    4
  • qwen35.ssm.group_count
    16
  • qwen35.ssm.inner_size
    2048
  • qwen35.ssm.state_size
    128
  • qwen35.ssm.time_step_rank
    16
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.eos_token_id
    248046
  • tokenizer.ggml.merges
    [Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    248044
  • tokenizer.ggml.pre
    qwen35
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [1024, 248320]
    • blk.24
  • blk.24.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.24.attn_k_norm.weight
    F32
    [256]
  • blk.24.attn_norm.weight
    F32
    [1024]
  • blk.24.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.24.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.24.attn_q_norm.weight
    F32
    [256]
  • blk.24.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.24.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.24.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.24.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.24.nextn.eh_proj.weight
    Q8_0
    [2048, 1024]
  • blk.24.nextn.enorm.weight
    F32
    [1024]
  • blk.24.nextn.hnorm.weight
    F32
    [1024]
  • blk.24.nextn.shared_head_norm.weight
    F32
    [1024]
  • blk.24.post_attention_norm.weight
    F32
    [1024]
  • output_norm.weight
    F32
    [1024]