mapler / lite-mistral-150m-instruct

This is a Lite series model based on the Mistral architecture, comprising approximately 157 million parameters.

100 4 months ago

1 Tag
... /
model
0fd75f692a01 · 167MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    7
  • general.name
    Lite-Mistral-150M-Instruct
  • general.quantization_version
    2
  • llama.attention.head_count
    16
  • llama.attention.head_count_kv
    8
  • llama.attention.layer_norm_rms_epsilon
    1e-06
  • llama.block_count
    12
  • llama.context_length
    2048
  • llama.embedding_length
    768
  • llama.feed_forward_length
    3072
  • llama.rope.dimension_count
    48
  • llama.rope.freq_base
    10000
  • llama.vocab_size
    32768
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [2, 3, 3, 6, 6, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, <0x00>, <0x01>, ...]
  • tokenizer.ggml.unknown_token_id
    0
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q8_0
    [768, 32768]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [768, 384]
  • blk.0.attn_norm.weight
    F32
    [768]
  • blk.0.attn_output.weight
    Q8_0
    [768, 768]
  • blk.0.attn_q.weight
    Q8_0
    [768, 768]
  • blk.0.attn_v.weight
    Q8_0
    [768, 384]
  • blk.0.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.0.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.0.ffn_norm.weight
    F32
    [768]
  • blk.0.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [768, 384]
  • blk.1.attn_norm.weight
    F32
    [768]
  • blk.1.attn_output.weight
    Q8_0
    [768, 768]
  • blk.1.attn_q.weight
    Q8_0
    [768, 768]
  • blk.1.attn_v.weight
    Q8_0
    [768, 384]
  • blk.1.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.1.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.1.ffn_norm.weight
    F32
    [768]
  • blk.1.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [768, 384]
  • blk.2.attn_norm.weight
    F32
    [768]
  • blk.2.attn_output.weight
    Q8_0
    [768, 768]
  • blk.2.attn_q.weight
    Q8_0
    [768, 768]
  • blk.2.attn_v.weight
    Q8_0
    [768, 384]
  • blk.2.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.2.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.2.ffn_norm.weight
    F32
    [768]
  • blk.2.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [768, 384]
  • blk.3.attn_norm.weight
    F32
    [768]
  • blk.3.attn_output.weight
    Q8_0
    [768, 768]
  • blk.3.attn_q.weight
    Q8_0
    [768, 768]
  • blk.3.attn_v.weight
    Q8_0
    [768, 384]
  • blk.3.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.3.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.3.ffn_norm.weight
    F32
    [768]
  • blk.3.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [768, 384]
  • blk.4.attn_norm.weight
    F32
    [768]
  • blk.4.attn_output.weight
    Q8_0
    [768, 768]
  • blk.4.attn_q.weight
    Q8_0
    [768, 768]
  • blk.4.attn_v.weight
    Q8_0
    [768, 384]
  • blk.4.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.4.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.4.ffn_norm.weight
    F32
    [768]
  • blk.4.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [768, 384]
  • blk.5.attn_norm.weight
    F32
    [768]
  • blk.5.attn_output.weight
    Q8_0
    [768, 768]
  • blk.5.attn_q.weight
    Q8_0
    [768, 768]
  • blk.5.attn_v.weight
    Q8_0
    [768, 384]
  • blk.5.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.5.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.5.ffn_norm.weight
    F32
    [768]
  • blk.5.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [768, 384]
  • blk.6.attn_norm.weight
    F32
    [768]
  • blk.6.attn_output.weight
    Q8_0
    [768, 768]
  • blk.6.attn_q.weight
    Q8_0
    [768, 768]
  • blk.6.attn_v.weight
    Q8_0
    [768, 384]
  • blk.6.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.6.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.6.ffn_norm.weight
    F32
    [768]
  • blk.6.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [768, 384]
  • blk.7.attn_norm.weight
    F32
    [768]
  • blk.7.attn_output.weight
    Q8_0
    [768, 768]
  • blk.7.attn_q.weight
    Q8_0
    [768, 768]
  • blk.7.attn_v.weight
    Q8_0
    [768, 384]
  • blk.7.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.7.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.7.ffn_norm.weight
    F32
    [768]
  • blk.7.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [768, 384]
  • blk.8.attn_norm.weight
    F32
    [768]
  • blk.8.attn_output.weight
    Q8_0
    [768, 768]
  • blk.8.attn_q.weight
    Q8_0
    [768, 768]
  • blk.8.attn_v.weight
    Q8_0
    [768, 384]
  • blk.8.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.8.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.8.ffn_norm.weight
    F32
    [768]
  • blk.8.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.9
  • blk.9.attn_k.weight
    Q8_0
    [768, 384]
  • blk.9.attn_norm.weight
    F32
    [768]
  • blk.9.attn_output.weight
    Q8_0
    [768, 768]
  • blk.9.attn_q.weight
    Q8_0
    [768, 768]
  • blk.9.attn_v.weight
    Q8_0
    [768, 384]
  • blk.9.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.9.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.9.ffn_norm.weight
    F32
    [768]
  • blk.9.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [768, 384]
  • blk.10.attn_norm.weight
    F32
    [768]
  • blk.10.attn_output.weight
    Q8_0
    [768, 768]
  • blk.10.attn_q.weight
    Q8_0
    [768, 768]
  • blk.10.attn_v.weight
    Q8_0
    [768, 384]
  • blk.10.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.10.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.10.ffn_norm.weight
    F32
    [768]
  • blk.10.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [768, 384]
  • blk.11.attn_norm.weight
    F32
    [768]
  • blk.11.attn_output.weight
    Q8_0
    [768, 768]
  • blk.11.attn_q.weight
    Q8_0
    [768, 768]
  • blk.11.attn_v.weight
    Q8_0
    [768, 384]
  • blk.11.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.11.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.11.ffn_norm.weight
    F32
    [768]
  • blk.11.ffn_up.weight
    Q8_0
    [768, 3072]
  • output.weight
    Q8_0
    [768, 32768]
  • output_norm.weight
    F32
    [768]