TCYZ/ cokertme2:ultra

45 1 month ago

Çökertme 2, Türkçe doğal dil işleme görevleri için optimize edilmiş, ultra-nano ölçekli bir dil modeli ailesidir.

0.8m 1.6m 5m 6m 14m 100m
... /
model
c2284d209bb4 · 21MB
    Metadata
  • general.architecture
    llama
  • llama.attention.head_count
    3
  • llama.attention.head_count_kv
    3
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    2
  • llama.context_length
    128
  • llama.embedding_length
    48
  • llama.feed_forward_length
    128
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • token_embd.weight
    F32
    [48, 50257]
    • blk.0
  • blk.0.attn_k.weight
    F32
    [48, 48]
  • blk.0.attn_norm.weight
    F32
    [48]
  • blk.0.attn_output.weight
    F32
    [48, 48]
  • blk.0.attn_q.weight
    F32
    [48, 48]
  • blk.0.attn_v.weight
    F32
    [48, 48]
  • blk.0.ffn_down.weight
    F32
    [128, 48]
  • blk.0.ffn_gate.weight
    F32
    [48, 128]
  • blk.0.ffn_norm.weight
    F32
    [48]
  • blk.0.ffn_up.weight
    F32
    [48, 128]
    • blk.1
  • blk.1.attn_k.weight
    F32
    [48, 48]
  • blk.1.attn_norm.weight
    F32
    [48]
  • blk.1.attn_output.weight
    F32
    [48, 48]
  • blk.1.attn_q.weight
    F32
    [48, 48]
  • blk.1.attn_v.weight
    F32
    [48, 48]
  • blk.1.ffn_down.weight
    F32
    [128, 48]
  • blk.1.ffn_gate.weight
    F32
    [48, 128]
  • blk.1.ffn_norm.weight
    F32
    [48]
  • blk.1.ffn_up.weight
    F32
    [48, 128]
  • output.weight
    F32
    [48, 50257]
  • output_norm.weight
    F32
    [48]