Gemma is a family of lightweight, state-of-the-art open models built by Google DeepMind. Updated to version 1.1

2b 7b

4.2M 8 months ago

4f76842861e7 · 1.5GB
    Metadata
  • gemma.attention.head_count
    8
  • gemma.attention.head_count_kv
    1
  • gemma.attention.key_length
    256
  • gemma.attention.layer_norm_rms_epsilon
    1e-06
  • gemma.attention.value_length
    256
  • gemma.block_count
    18
  • gemma.context_length
    8192
  • gemma.embedding_length
    2048
  • gemma.feed_forward_length
    16384
  • general.architecture
    gemma
  • general.file_type
    12
  • general.name
    gemma-2b-it
  • general.quantization_version
    2
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [3, 3, 3, 2, 1, ...]
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
  • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q8_0
    [2048, 256128]
  • blk.0
  • blk.0.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.0.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.0.attn_v.weight
    Q5_K
    [2048, 256]
  • blk.0.ffn_down.weight
    Q5_K
    [16384, 2048]
  • blk.0.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.1
  • blk.1.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.1.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.1.attn_v.weight
    Q5_K
    [2048, 256]
  • blk.1.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.1.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.2
  • blk.2.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.2.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.2.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.2.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.2.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.3
  • blk.3.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.3.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.3.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.3.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.3.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.4
  • blk.4.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.4.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.4.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.4.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.4.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.5
  • blk.5.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.5.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.5.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.5.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.5.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.6
  • blk.6.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.6.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.6.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.6.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.6.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.7
  • blk.7.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.7.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.7.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.7.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.7.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.8
  • blk.8.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.8.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.8.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.8.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.8.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.9
  • blk.9.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.9.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.9.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.9.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.9.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.10
  • blk.10.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.10.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.10.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.10.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.10.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.11
  • blk.11.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.11.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.11.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.11.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.11.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.12
  • blk.12.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.12.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.12.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.12.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.12.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.13
  • blk.13.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.13.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.13.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.13.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.13.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.14
  • blk.14.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.14.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.14.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.14.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.14.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.15
  • blk.15.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.15.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.15.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.15.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.15.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.16
  • blk.16.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.16.attn_norm.weight
    F32
    [2048]
  • blk.16.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.16.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.16.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.16.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.16.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.16.ffn_norm.weight
    F32
    [2048]
  • blk.16.ffn_up.weight
    Q3_K
    [2048, 16384]
  • blk.17
  • blk.17.attn_k.weight
    Q3_K
    [2048, 256]
  • blk.17.attn_norm.weight
    F32
    [2048]
  • blk.17.attn_output.weight
    Q4_K
    [2048, 2048]
  • blk.17.attn_q.weight
    Q3_K
    [2048, 2048]
  • blk.17.attn_v.weight
    Q4_K
    [2048, 256]
  • blk.17.ffn_down.weight
    Q4_K
    [16384, 2048]
  • blk.17.ffn_gate.weight
    Q3_K
    [2048, 16384]
  • blk.17.ffn_norm.weight
    F32
    [2048]
  • blk.17.ffn_up.weight
    Q3_K
    [2048, 16384]
  • output_norm.weight
    F32
    [2048]