alibayram/
kumru:latest

19 2 weeks ago

Kumru-2B is the lightweight, open-source version of Kumru LLM, developed for Turkish from scratch by VNGRS.

... /
model
ddd7e27e50d9 · 1.5GB
    Metadata
  • general.architecture
    llama
  • general.file_type
    Q4_K_M
  • llama.attention.head_count
    16
  • llama.attention.head_count_kv
    4
  • llama.attention.key_length
    128
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.attention.value_length
    128
  • llama.block_count
    18
  • llama.context_length
    8192
  • llama.embedding_length
    3072
  • llama.feed_forward_length
    10752
  • llama.rope.dimension_count
    128
  • llama.rope.freq_base
    500000
  • llama.vocab_size
    50176
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    3
  • tokenizer.ggml.merges
    [Ä ±, n Ġ, e Ġ, l a, e r, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<PAD>, <UNK>, <BOS>, <EOS>, <INST>, ...]
  • tokenizer.ggml.unknown_token_id
    1
    • Tensor
  • token_embd.weight
    Q4_K
    [3072, 50176]
    • blk.0
  • blk.0.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.0.attn_norm.weight
    F32
    [3072]
  • blk.0.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.0.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.0.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.0.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.0.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.0.ffn_norm.weight
    F32
    [3072]
  • blk.0.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.1
  • blk.1.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.1.attn_norm.weight
    F32
    [3072]
  • blk.1.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.1.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.1.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.1.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.1.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.1.ffn_norm.weight
    F32
    [3072]
  • blk.1.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.2
  • blk.2.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.2.attn_norm.weight
    F32
    [3072]
  • blk.2.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.2.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.2.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.2.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.2.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.2.ffn_norm.weight
    F32
    [3072]
  • blk.2.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.3
  • blk.3.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.3.attn_norm.weight
    F32
    [3072]
  • blk.3.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.3.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.3.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.3.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.3.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.3.ffn_norm.weight
    F32
    [3072]
  • blk.3.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.4
  • blk.4.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.4.attn_norm.weight
    F32
    [3072]
  • blk.4.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.4.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.4.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.4.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.4.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.4.ffn_norm.weight
    F32
    [3072]
  • blk.4.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.5
  • blk.5.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.5.attn_norm.weight
    F32
    [3072]
  • blk.5.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.5.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.5.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.5.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.5.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.5.ffn_norm.weight
    F32
    [3072]
  • blk.5.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.6
  • blk.6.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.6.attn_norm.weight
    F32
    [3072]
  • blk.6.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.6.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.6.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.6.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.6.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.6.ffn_norm.weight
    F32
    [3072]
  • blk.6.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.7
  • blk.7.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.7.attn_norm.weight
    F32
    [3072]
  • blk.7.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.7.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.7.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.7.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.7.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.7.ffn_norm.weight
    F32
    [3072]
  • blk.7.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.8
  • blk.8.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.8.attn_norm.weight
    F32
    [3072]
  • blk.8.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.8.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.8.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.8.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.8.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.8.ffn_norm.weight
    F32
    [3072]
  • blk.8.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.9
  • blk.9.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.9.attn_norm.weight
    F32
    [3072]
  • blk.9.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.9.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.9.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.9.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.9.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.9.ffn_norm.weight
    F32
    [3072]
  • blk.9.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.10
  • blk.10.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.10.attn_norm.weight
    F32
    [3072]
  • blk.10.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.10.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.10.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.10.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.10.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.10.ffn_norm.weight
    F32
    [3072]
  • blk.10.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.11
  • blk.11.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.11.attn_norm.weight
    F32
    [3072]
  • blk.11.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.11.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.11.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.11.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.11.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.11.ffn_norm.weight
    F32
    [3072]
  • blk.11.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.12
  • blk.12.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.12.attn_norm.weight
    F32
    [3072]
  • blk.12.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.12.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.12.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.12.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.12.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.12.ffn_norm.weight
    F32
    [3072]
  • blk.12.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.13
  • blk.13.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.13.attn_norm.weight
    F32
    [3072]
  • blk.13.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.13.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.13.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.13.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.13.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.13.ffn_norm.weight
    F32
    [3072]
  • blk.13.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.14
  • blk.14.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.14.attn_norm.weight
    F32
    [3072]
  • blk.14.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.14.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.14.attn_v.weight
    Q4_K
    [3072, 512]
  • blk.14.ffn_down.weight
    Q4_K
    [10752, 3072]
  • blk.14.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.14.ffn_norm.weight
    F32
    [3072]
  • blk.14.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.15
  • blk.15.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.15.attn_norm.weight
    F32
    [3072]
  • blk.15.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.15.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.15.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.15.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.15.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.15.ffn_norm.weight
    F32
    [3072]
  • blk.15.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.16
  • blk.16.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.16.attn_norm.weight
    F32
    [3072]
  • blk.16.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.16.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.16.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.16.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.16.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.16.ffn_norm.weight
    F32
    [3072]
  • blk.16.ffn_up.weight
    Q4_K
    [3072, 10752]
    • blk.17
  • blk.17.attn_k.weight
    Q4_K
    [3072, 512]
  • blk.17.attn_norm.weight
    F32
    [3072]
  • blk.17.attn_output.weight
    Q4_K
    [2048, 3072]
  • blk.17.attn_q.weight
    Q4_K
    [3072, 2048]
  • blk.17.attn_v.weight
    Q6_K
    [3072, 512]
  • blk.17.ffn_down.weight
    Q6_K
    [10752, 3072]
  • blk.17.ffn_gate.weight
    Q4_K
    [3072, 10752]
  • blk.17.ffn_norm.weight
    F32
    [3072]
  • blk.17.ffn_up.weight
    Q4_K
    [3072, 10752]
  • output.weight
    Q6_K
    [3072, 50176]
  • output_norm.weight
    F32
    [3072]