falcon3

A family of efficient AI models under 10B parameters performant in science, math, and coding through innovative training techniques.

1b 3b 7b 10b

3,914 3 days ago

17 Tags
... /
model
89056211c7c7 · 1.8GB
    Metadata
  • general.architecture
    llama
  • general.base_model.0.name
    Falcon3 1B Base
  • general.base_model.0.organization
    Tiiuae
  • general.base_model.0.repo_url
    https://huggingface.co/tiiuae/Falcon3-1B-Base
  • general.base_model.count
    1
  • general.basename
    Falcon3
  • general.file_type
    7
  • general.finetune
    Instruct
  • general.languages
    [en, fr, es, pt]
  • general.license
    other
  • general.license.link
    https://falconllm.tii.ae/falcon-terms-and-conditions.html
  • general.license.name
    falcon-llm-license
  • general.name
    Falcon3 1B Instruct
  • general.quantization_version
    2
  • general.size_label
    1B
  • general.tags
    [falcon3]
  • general.type
    model
  • llama.attention.head_count
    8
  • llama.attention.head_count_kv
    4
  • llama.attention.key_length
    256
  • llama.attention.layer_norm_rms_epsilon
    1e-06
  • llama.attention.value_length
    256
  • llama.block_count
    18
  • llama.context_length
    8192
  • llama.embedding_length
    2048
  • llama.feed_forward_length
    8192
  • llama.rope.dimension_count
    256
  • llama.rope.freq_base
    1.000042e+06
  • llama.vocab_size
    131072
  • tokenizer.ggml.add_space_prefix
    false
  • tokenizer.ggml.eos_token_id
    11
  • tokenizer.ggml.merges
    [N E, Ġ Ġ, Ġ t, Ġ a, > >, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    2023
  • tokenizer.ggml.pre
    falcon3
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [>>TITLE<<, >>ABSTRACT<<, >>INTRODUCTION<<, >>SUMMARY<<, >>COMMENT<<, ...]
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q8_0
    [2048, 131072]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.0.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.0.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.0.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.0.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.1.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.1.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.1.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.1.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.2.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.2.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.3.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.3.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.3.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.3.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.4.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.4.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.4.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.4.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.5.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.5.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.5.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.5.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.6.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.6.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.7.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.7.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.7.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.7.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.8.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.8.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.8.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.8.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.9
  • blk.9.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.9.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.9.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.9.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.9.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.10.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.10.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.11.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.11.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.11.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.11.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.12
  • blk.12.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.12.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.12.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.12.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.12.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.13
  • blk.13.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.13.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.13.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.13.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.13.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.14
  • blk.14.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.14.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.14.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.15
  • blk.15.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.15.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.15.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.15.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.15.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.16
  • blk.16.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.16.attn_norm.weight
    F32
    [2048]
  • blk.16.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.16.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.16.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.16.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.16.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.16.ffn_norm.weight
    F32
    [2048]
  • blk.16.ffn_up.weight
    Q8_0
    [2048, 8192]
    • blk.17
  • blk.17.attn_k.weight
    Q8_0
    [2048, 1024]
  • blk.17.attn_norm.weight
    F32
    [2048]
  • blk.17.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.17.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.17.attn_v.weight
    Q8_0
    [2048, 1024]
  • blk.17.ffn_down.weight
    Q8_0
    [8192, 2048]
  • blk.17.ffn_gate.weight
    Q8_0
    [2048, 8192]
  • blk.17.ffn_norm.weight
    F32
    [2048]
  • blk.17.ffn_up.weight
    Q8_0
    [2048, 8192]
  • output.weight
    Q8_0
    [2048, 131072]
  • output_norm.weight
    F32
    [2048]