TCYZ/ cokertme:1m

71 2 months ago

Türkiye'nin yapay zeka hamlesinde stratejik bir boşluğu dolduran Çökertme serisi, devasa modellerin aksine "her cihazda çalışan zeka" mottosuyla geliştirilmiştir. TCYZ projesi kapsamında sunulan bu aile, en küçük donanımla çalışabilir.

1m 1.6m 6.7m 28m 57m
... /
model
25b3f70fc21c · 5.0MB
    Metadata
  • general.architecture
    llama
  • llama.attention.head_count
    4
  • llama.attention.head_count_kv
    4
  • llama.attention.layer_norm_rms_epsilon
    1e-06
  • llama.block_count
    6
  • llama.context_length
    64
  • llama.embedding_length
    128
  • llama.feed_forward_length
    256
  • llama.rope.dimension_count
    32
  • tokenizer.ggml.bos_token_id
    0
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.merges
    [Ä ±, o r, a n, Ġ b, e r, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    1
  • tokenizer.ggml.tokens
    [<s>, <pad>, </s>, <unk>, <mask>, ...]
    • Tensor
  • token_embd.weight
    F32
    [128, 1000]
    • blk.0
  • blk.0.attn_k.weight
    F32
    [128, 128]
  • blk.0.attn_norm.weight
    F32
    [128]
  • blk.0.attn_output.weight
    F32
    [128, 128]
  • blk.0.attn_q.weight
    F32
    [128, 128]
  • blk.0.attn_v.weight
    F32
    [128, 128]
  • blk.0.ffn_down.weight
    F32
    [256, 128]
  • blk.0.ffn_gate.weight
    F32
    [128, 256]
  • blk.0.ffn_norm.weight
    F32
    [128]
  • blk.0.ffn_up.weight
    F32
    [128, 256]
    • blk.1
  • blk.1.attn_k.weight
    F32
    [128, 128]
  • blk.1.attn_norm.weight
    F32
    [128]
  • blk.1.attn_output.weight
    F32
    [128, 128]
  • blk.1.attn_q.weight
    F32
    [128, 128]
  • blk.1.attn_v.weight
    F32
    [128, 128]
  • blk.1.ffn_down.weight
    F32
    [256, 128]
  • blk.1.ffn_gate.weight
    F32
    [128, 256]
  • blk.1.ffn_norm.weight
    F32
    [128]
  • blk.1.ffn_up.weight
    F32
    [128, 256]
    • blk.2
  • blk.2.attn_k.weight
    F32
    [128, 128]
  • blk.2.attn_norm.weight
    F32
    [128]
  • blk.2.attn_output.weight
    F32
    [128, 128]
  • blk.2.attn_q.weight
    F32
    [128, 128]
  • blk.2.attn_v.weight
    F32
    [128, 128]
  • blk.2.ffn_down.weight
    F32
    [256, 128]
  • blk.2.ffn_gate.weight
    F32
    [128, 256]
  • blk.2.ffn_norm.weight
    F32
    [128]
  • blk.2.ffn_up.weight
    F32
    [128, 256]
    • blk.3
  • blk.3.attn_k.weight
    F32
    [128, 128]
  • blk.3.attn_norm.weight
    F32
    [128]
  • blk.3.attn_output.weight
    F32
    [128, 128]
  • blk.3.attn_q.weight
    F32
    [128, 128]
  • blk.3.attn_v.weight
    F32
    [128, 128]
  • blk.3.ffn_down.weight
    F32
    [256, 128]
  • blk.3.ffn_gate.weight
    F32
    [128, 256]
  • blk.3.ffn_norm.weight
    F32
    [128]
  • blk.3.ffn_up.weight
    F32
    [128, 256]
    • blk.4
  • blk.4.attn_k.weight
    F32
    [128, 128]
  • blk.4.attn_norm.weight
    F32
    [128]
  • blk.4.attn_output.weight
    F32
    [128, 128]
  • blk.4.attn_q.weight
    F32
    [128, 128]
  • blk.4.attn_v.weight
    F32
    [128, 128]
  • blk.4.ffn_down.weight
    F32
    [256, 128]
  • blk.4.ffn_gate.weight
    F32
    [128, 256]
  • blk.4.ffn_norm.weight
    F32
    [128]
  • blk.4.ffn_up.weight
    F32
    [128, 256]
    • blk.5
  • blk.5.attn_k.weight
    F32
    [128, 128]
  • blk.5.attn_norm.weight
    F32
    [128]
  • blk.5.attn_output.weight
    F32
    [128, 128]
  • blk.5.attn_q.weight
    F32
    [128, 128]
  • blk.5.attn_v.weight
    F32
    [128, 128]
  • blk.5.ffn_down.weight
    F32
    [256, 128]
  • blk.5.ffn_gate.weight
    F32
    [128, 256]
  • blk.5.ffn_norm.weight
    F32
    [128]
  • blk.5.ffn_up.weight
    F32
    [128, 256]
  • output.weight
    F32
    [128, 1000]
  • output_norm.weight
    F32
    [128]