maternion/ nemotron-3.5-embed:1b-q6_k

317 1 week ago

Nemotron-3.5-Embed 1B and 8B: multilingual text + image embedding models for retrieval and RAG. 32,768 tokens context, 34 languages, bundled vision projector.

vision embedding 1b 8b
... /
model
1377e15f4649 · 944MB
    Metadata
  • general.architecture
    mistral3
  • general.file_type
    Q6_K
  • mistral3.attention.causal
    false
  • mistral3.attention.head_count
    24
  • mistral3.attention.head_count_kv
    8
  • mistral3.attention.key_length
    128
  • mistral3.attention.layer_norm_rms_epsilon
    1e-05
  • mistral3.attention.temperature_scale
    0.1
  • mistral3.attention.value_length
    128
  • mistral3.block_count
    16
  • mistral3.context_length
    262144
  • mistral3.embedding_length
    2048
  • mistral3.feed_forward_length
    6144
  • mistral3.pooling_type
    Mean
  • mistral3.rope.dimension_count
    128
  • mistral3.rope.freq_base
    1e+06
  • mistral3.rope.scaling.factor
    16
  • mistral3.rope.scaling.original_context_length
    16384
  • mistral3.rope.scaling.type
    yarn
  • mistral3.rope.scaling.yarn_beta_fast
    32
  • mistral3.rope.scaling.yarn_beta_slow
    1
  • mistral3.rope.scaling.yarn_log_multiplier
    1
  • mistral3.vocab_size
    131072
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ t, e r, i n, Ġ ĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    11
  • tokenizer.ggml.pre
    pixtral
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, [INST], [/INST], ...]
  • tokenizer.ggml.unknown_token_id
    0
    • Tensor
  • token_embd.weight
    Q6_K
    [2048, 131072]
    • blk.0
  • blk.0.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.0.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.0.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.0.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.0.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.1
  • blk.1.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.1.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.1.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.1.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.1.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.2
  • blk.2.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.2.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.2.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.2.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.2.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.3
  • blk.3.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.3.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.3.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.3.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.3.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.4
  • blk.4.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.4.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.4.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.4.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.4.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.5
  • blk.5.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.5.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.5.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.5.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.5.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.6
  • blk.6.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.6.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.6.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.6.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.6.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.7
  • blk.7.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.7.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.7.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.7.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.7.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.8
  • blk.8.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.8.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.8.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.8.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.8.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.9
  • blk.9.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.9.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.9.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.9.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.9.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.10
  • blk.10.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.10.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.10.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.10.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.10.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.11
  • blk.11.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.11.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.11.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.11.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.11.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.12
  • blk.12.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.12.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.12.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.12.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.12.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.13
  • blk.13.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.13.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.13.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.13.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.13.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.14
  • blk.14.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.14.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.14.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.14.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.14.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up.weight
    Q6_K
    [2048, 6144]
    • blk.15
  • blk.15.attn_k.weight
    Q6_K
    [2048, 1024]
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.attn_output.weight
    Q6_K
    [3072, 2048]
  • blk.15.attn_q.weight
    Q6_K
    [2048, 3072]
  • blk.15.attn_v.weight
    Q6_K
    [2048, 1024]
  • blk.15.ffn_down.weight
    Q6_K
    [6144, 2048]
  • blk.15.ffn_gate.weight
    Q6_K
    [2048, 6144]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up.weight
    Q6_K
    [2048, 6144]
  • output_norm.weight
    F32
    [2048]