msn/
artl:latest

18 9 months ago

shorter access to arnir-0's tinyllm

... /
model
8620f9087a68 · 11MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    IQ4_XS
  • llama.attention.head_count
    2
  • llama.attention.head_count_kv
    1
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    1
  • llama.context_length
    1024
  • llama.embedding_length
    192
  • llama.feed_forward_length
    1024
  • llama.rope.dimension_count
    96
  • llama.vocab_size
    32000
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_space_prefix
    true
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [-1000, -1000, -1000, 0, 0, ...]
  • tokenizer.ggml.token_type
    [3, 3, 3, 6, 6, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, <0x00>, <0x01>, ...]
  • tokenizer.ggml.unknown_token_id
    0
  • mradermacher.convert_type
    hf
  • mradermacher.quantize_version
    2
  • mradermacher.quantized_at
    2025-02-10T07:25:17+01:00
  • mradermacher.quantized_by
    mradermacher
  • mradermacher.quantized_on
    rich1
    • Tensor
  • token_embd.weight
    IQ4_NL
    [192, 32000]
    • blk.0
  • blk.0.attn_k.weight
    IQ4_NL
    [192, 96]
  • blk.0.attn_norm.weight
    F32
    [192]
  • blk.0.attn_output.weight
    IQ4_NL
    [192, 192]
  • blk.0.attn_q.weight
    IQ4_NL
    [192, 192]
  • blk.0.attn_v.weight
    IQ4_NL
    [192, 96]
  • blk.0.ffn_down.weight
    IQ4_XS
    [1024, 192]
  • blk.0.ffn_gate.weight
    IQ4_NL
    [192, 1024]
  • blk.0.ffn_norm.weight
    F32
    [192]
  • blk.0.ffn_up.weight
    IQ4_NL
    [192, 1024]
  • output.weight
    Q8_0
    [192, 32000]
  • output_norm.weight
    F32
    [192]