AZERDSQ/ g0-nano-base:latest

3 days ago

A 62M-parameter GPT trained from scratch on a single 8GB-RAM NVIDIA Jetson, offering a compact open-weights base model for raw text completion. Not instruction-tuned.

... /
model
6000868c66f7 · 292MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    F32
  • llama.attention.head_count
    10
  • llama.attention.head_count_kv
    2
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    12
  • llama.context_length
    1024
  • llama.embedding_length
    640
  • llama.feed_forward_length
    1728
  • llama.rope.dimension_count
    64
  • llama.rope.freq_base
    10000
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    3
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [0, 0, 0, 0, -0, ...]
  • tokenizer.ggml.token_type
    [3, 2, 3, 3, 1, ...]
  • tokenizer.ggml.tokens
    [<pad>, <unk>, <s>, </s>, ▁t, ...]
  • tokenizer.ggml.unknown_token_id
    1
    • Tensor
  • token_embd.weight
    F32
    [640, 16640]
    • blk.0
  • blk.0.attn_k.weight
    F32
    [640, 128]
  • blk.0.attn_norm.weight
    F32
    [640]
  • blk.0.attn_output.weight
    F32
    [640, 640]
  • blk.0.attn_q.weight
    F32
    [640, 640]
  • blk.0.attn_v.weight
    F32
    [640, 128]
  • blk.0.ffn_down.weight
    F32
    [1728, 640]
  • blk.0.ffn_gate.weight
    F32
    [640, 1728]
  • blk.0.ffn_norm.weight
    F32
    [640]
  • blk.0.ffn_up.weight
    F32
    [640, 1728]
    • blk.1
  • blk.1.attn_k.weight
    F32
    [640, 128]
  • blk.1.attn_norm.weight
    F32
    [640]
  • blk.1.attn_output.weight
    F32
    [640, 640]
  • blk.1.attn_q.weight
    F32
    [640, 640]
  • blk.1.attn_v.weight
    F32
    [640, 128]
  • blk.1.ffn_down.weight
    F32
    [1728, 640]
  • blk.1.ffn_gate.weight
    F32
    [640, 1728]
  • blk.1.ffn_norm.weight
    F32
    [640]
  • blk.1.ffn_up.weight
    F32
    [640, 1728]
    • blk.2
  • blk.2.attn_k.weight
    F32
    [640, 128]
  • blk.2.attn_norm.weight
    F32
    [640]
  • blk.2.attn_output.weight
    F32
    [640, 640]
  • blk.2.attn_q.weight
    F32
    [640, 640]
  • blk.2.attn_v.weight
    F32
    [640, 128]
  • blk.2.ffn_down.weight
    F32
    [1728, 640]
  • blk.2.ffn_gate.weight
    F32
    [640, 1728]
  • blk.2.ffn_norm.weight
    F32
    [640]
  • blk.2.ffn_up.weight
    F32
    [640, 1728]
    • blk.3
  • blk.3.attn_k.weight
    F32
    [640, 128]
  • blk.3.attn_norm.weight
    F32
    [640]
  • blk.3.attn_output.weight
    F32
    [640, 640]
  • blk.3.attn_q.weight
    F32
    [640, 640]
  • blk.3.attn_v.weight
    F32
    [640, 128]
  • blk.3.ffn_down.weight
    F32
    [1728, 640]
  • blk.3.ffn_gate.weight
    F32
    [640, 1728]
  • blk.3.ffn_norm.weight
    F32
    [640]
  • blk.3.ffn_up.weight
    F32
    [640, 1728]
    • blk.4
  • blk.4.attn_k.weight
    F32
    [640, 128]
  • blk.4.attn_norm.weight
    F32
    [640]
  • blk.4.attn_output.weight
    F32
    [640, 640]
  • blk.4.attn_q.weight
    F32
    [640, 640]
  • blk.4.attn_v.weight
    F32
    [640, 128]
  • blk.4.ffn_down.weight
    F32
    [1728, 640]
  • blk.4.ffn_gate.weight
    F32
    [640, 1728]
  • blk.4.ffn_norm.weight
    F32
    [640]
  • blk.4.ffn_up.weight
    F32
    [640, 1728]
    • blk.5
  • blk.5.attn_k.weight
    F32
    [640, 128]
  • blk.5.attn_norm.weight
    F32
    [640]
  • blk.5.attn_output.weight
    F32
    [640, 640]
  • blk.5.attn_q.weight
    F32
    [640, 640]
  • blk.5.attn_v.weight
    F32
    [640, 128]
  • blk.5.ffn_down.weight
    F32
    [1728, 640]
  • blk.5.ffn_gate.weight
    F32
    [640, 1728]
  • blk.5.ffn_norm.weight
    F32
    [640]
  • blk.5.ffn_up.weight
    F32
    [640, 1728]
    • blk.6
  • blk.6.attn_k.weight
    F32
    [640, 128]
  • blk.6.attn_norm.weight
    F32
    [640]
  • blk.6.attn_output.weight
    F32
    [640, 640]
  • blk.6.attn_q.weight
    F32
    [640, 640]
  • blk.6.attn_v.weight
    F32
    [640, 128]
  • blk.6.ffn_down.weight
    F32
    [1728, 640]
  • blk.6.ffn_gate.weight
    F32
    [640, 1728]
  • blk.6.ffn_norm.weight
    F32
    [640]
  • blk.6.ffn_up.weight
    F32
    [640, 1728]
    • blk.7
  • blk.7.attn_k.weight
    F32
    [640, 128]
  • blk.7.attn_norm.weight
    F32
    [640]
  • blk.7.attn_output.weight
    F32
    [640, 640]
  • blk.7.attn_q.weight
    F32
    [640, 640]
  • blk.7.attn_v.weight
    F32
    [640, 128]
  • blk.7.ffn_down.weight
    F32
    [1728, 640]
  • blk.7.ffn_gate.weight
    F32
    [640, 1728]
  • blk.7.ffn_norm.weight
    F32
    [640]
  • blk.7.ffn_up.weight
    F32
    [640, 1728]
    • blk.8
  • blk.8.attn_k.weight
    F32
    [640, 128]
  • blk.8.attn_norm.weight
    F32
    [640]
  • blk.8.attn_output.weight
    F32
    [640, 640]
  • blk.8.attn_q.weight
    F32
    [640, 640]
  • blk.8.attn_v.weight
    F32
    [640, 128]
  • blk.8.ffn_down.weight
    F32
    [1728, 640]
  • blk.8.ffn_gate.weight
    F32
    [640, 1728]
  • blk.8.ffn_norm.weight
    F32
    [640]
  • blk.8.ffn_up.weight
    F32
    [640, 1728]
    • blk.9
  • blk.9.attn_k.weight
    F32
    [640, 128]
  • blk.9.attn_norm.weight
    F32
    [640]
  • blk.9.attn_output.weight
    F32
    [640, 640]
  • blk.9.attn_q.weight
    F32
    [640, 640]
  • blk.9.attn_v.weight
    F32
    [640, 128]
  • blk.9.ffn_down.weight
    F32
    [1728, 640]
  • blk.9.ffn_gate.weight
    F32
    [640, 1728]
  • blk.9.ffn_norm.weight
    F32
    [640]
  • blk.9.ffn_up.weight
    F32
    [640, 1728]
    • blk.10
  • blk.10.attn_k.weight
    F32
    [640, 128]
  • blk.10.attn_norm.weight
    F32
    [640]
  • blk.10.attn_output.weight
    F32
    [640, 640]
  • blk.10.attn_q.weight
    F32
    [640, 640]
  • blk.10.attn_v.weight
    F32
    [640, 128]
  • blk.10.ffn_down.weight
    F32
    [1728, 640]
  • blk.10.ffn_gate.weight
    F32
    [640, 1728]
  • blk.10.ffn_norm.weight
    F32
    [640]
  • blk.10.ffn_up.weight
    F32
    [640, 1728]
    • blk.11
  • blk.11.attn_k.weight
    F32
    [640, 128]
  • blk.11.attn_norm.weight
    F32
    [640]
  • blk.11.attn_output.weight
    F32
    [640, 640]
  • blk.11.attn_q.weight
    F32
    [640, 640]
  • blk.11.attn_v.weight
    F32
    [640, 128]
  • blk.11.ffn_down.weight
    F32
    [1728, 640]
  • blk.11.ffn_gate.weight
    F32
    [640, 1728]
  • blk.11.ffn_norm.weight
    F32
    [640]
  • blk.11.ffn_up.weight
    F32
    [640, 1728]
  • output.weight
    F32
    [640, 16640]
  • output_norm.weight
    F32
    [640]