muhammadrafikov/
Bootstrap:latest

1 2 weeks ago

This is a Qwen 2 based LLM that I trained on my own Laptop. Pre-trained with C4 and fine-tuned with dolly 15k

... /
model
b12541cd50d9 · 270MB
    Metadata
  • general.architecture
    qwen2
  • general.file_type
    F16
  • qwen2.attention.head_count
    6
  • qwen2.attention.head_count_kv
    6
  • qwen2.attention.layer_norm_rms_epsilon
    1e-06
  • qwen2.block_count
    6
  • qwen2.context_length
    1024
  • qwen2.embedding_length
    768
  • qwen2.feed_forward_length
    3072
  • qwen2.rope.freq_base
    10000
  • tokenizer.ggml.bos_token_id
    50257
  • tokenizer.ggml.eos_token_id
    50256
  • tokenizer.ggml.merges
    [Ġ t, Ġ a, h e, i n, r e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    50258
  • tokenizer.ggml.pre
    gpt-2
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
  • tokenizer.ggml.unknown_token_id
    50259
    • Tensor
  • token_embd.weight
    F16
    [768, 50262]
    • blk.0
  • blk.0.attn_k.bias
    F32
    [768]
  • blk.0.attn_k.weight
    F16
    [768, 768]
  • blk.0.attn_norm.weight
    F32
    [768]
  • blk.0.attn_output.weight
    F16
    [768, 768]
  • blk.0.attn_q.bias
    F32
    [768]
  • blk.0.attn_q.weight
    F16
    [768, 768]
  • blk.0.attn_v.bias
    F32
    [768]
  • blk.0.attn_v.weight
    F16
    [768, 768]
  • blk.0.ffn_down.weight
    F16
    [3072, 768]
  • blk.0.ffn_gate.weight
    F16
    [768, 3072]
  • blk.0.ffn_norm.weight
    F32
    [768]
  • blk.0.ffn_up.weight
    F16
    [768, 3072]
    • blk.1
  • blk.1.attn_k.bias
    F32
    [768]
  • blk.1.attn_k.weight
    F16
    [768, 768]
  • blk.1.attn_norm.weight
    F32
    [768]
  • blk.1.attn_output.weight
    F16
    [768, 768]
  • blk.1.attn_q.bias
    F32
    [768]
  • blk.1.attn_q.weight
    F16
    [768, 768]
  • blk.1.attn_v.bias
    F32
    [768]
  • blk.1.attn_v.weight
    F16
    [768, 768]
  • blk.1.ffn_down.weight
    F16
    [3072, 768]
  • blk.1.ffn_gate.weight
    F16
    [768, 3072]
  • blk.1.ffn_norm.weight
    F32
    [768]
  • blk.1.ffn_up.weight
    F16
    [768, 3072]
    • blk.2
  • blk.2.attn_k.bias
    F32
    [768]
  • blk.2.attn_k.weight
    F16
    [768, 768]
  • blk.2.attn_norm.weight
    F32
    [768]
  • blk.2.attn_output.weight
    F16
    [768, 768]
  • blk.2.attn_q.bias
    F32
    [768]
  • blk.2.attn_q.weight
    F16
    [768, 768]
  • blk.2.attn_v.bias
    F32
    [768]
  • blk.2.attn_v.weight
    F16
    [768, 768]
  • blk.2.ffn_down.weight
    F16
    [3072, 768]
  • blk.2.ffn_gate.weight
    F16
    [768, 3072]
  • blk.2.ffn_norm.weight
    F32
    [768]
  • blk.2.ffn_up.weight
    F16
    [768, 3072]
    • blk.3
  • blk.3.attn_k.bias
    F32
    [768]
  • blk.3.attn_k.weight
    F16
    [768, 768]
  • blk.3.attn_norm.weight
    F32
    [768]
  • blk.3.attn_output.weight
    F16
    [768, 768]
  • blk.3.attn_q.bias
    F32
    [768]
  • blk.3.attn_q.weight
    F16
    [768, 768]
  • blk.3.attn_v.bias
    F32
    [768]
  • blk.3.attn_v.weight
    F16
    [768, 768]
  • blk.3.ffn_down.weight
    F16
    [3072, 768]
  • blk.3.ffn_gate.weight
    F16
    [768, 3072]
  • blk.3.ffn_norm.weight
    F32
    [768]
  • blk.3.ffn_up.weight
    F16
    [768, 3072]
    • blk.4
  • blk.4.attn_k.bias
    F32
    [768]
  • blk.4.attn_k.weight
    F16
    [768, 768]
  • blk.4.attn_norm.weight
    F32
    [768]
  • blk.4.attn_output.weight
    F16
    [768, 768]
  • blk.4.attn_q.bias
    F32
    [768]
  • blk.4.attn_q.weight
    F16
    [768, 768]
  • blk.4.attn_v.bias
    F32
    [768]
  • blk.4.attn_v.weight
    F16
    [768, 768]
  • blk.4.ffn_down.weight
    F16
    [3072, 768]
  • blk.4.ffn_gate.weight
    F16
    [768, 3072]
  • blk.4.ffn_norm.weight
    F32
    [768]
  • blk.4.ffn_up.weight
    F16
    [768, 3072]
    • blk.5
  • blk.5.attn_k.bias
    F32
    [768]
  • blk.5.attn_k.weight
    F16
    [768, 768]
  • blk.5.attn_norm.weight
    F32
    [768]
  • blk.5.attn_output.weight
    F16
    [768, 768]
  • blk.5.attn_q.bias
    F32
    [768]
  • blk.5.attn_q.weight
    F16
    [768, 768]
  • blk.5.attn_v.bias
    F32
    [768]
  • blk.5.attn_v.weight
    F16
    [768, 768]
  • blk.5.ffn_down.weight
    F16
    [3072, 768]
  • blk.5.ffn_gate.weight
    F16
    [768, 3072]
  • blk.5.ffn_norm.weight
    F32
    [768]
  • blk.5.ffn_up.weight
    F16
    [768, 3072]
  • output.weight
    F16
    [768, 50262]
  • output_norm.weight
    F32
    [768]