guoxuter/ ov_intent_analysis_sft:v4_q8

44 yesterday

A lightweight local intent-analysis model built for the OpenViking search API. It detects whether a conversation turn needs context retrieval, skips chitchat to reduce unnecessary memory injection and token usage, and emits structured retrieval queries

tools thinking
... /
model
186ea65a1e40 · 812MB
    Metadata
  • general.architecture
    qwen35
  • general.file_type
    Q8_0
  • qwen35.attention.head_count
    8
  • qwen35.attention.head_count_kv
    2
  • qwen35.attention.key_length
    256
  • qwen35.attention.layer_norm_rms_epsilon
    1e-06
  • qwen35.attention.value_length
    256
  • qwen35.block_count
    24
  • qwen35.context_length
    262144
  • qwen35.embedding_length
    1024
  • qwen35.feed_forward_length
    3584
  • qwen35.full_attention_interval
    4
  • qwen35.rope.dimension_count
    64
  • qwen35.rope.dimension_sections
    [11, 11, 10, 0]
  • qwen35.rope.freq_base
    1e+07
  • qwen35.ssm.conv_kernel
    4
  • qwen35.ssm.group_count
    16
  • qwen35.ssm.inner_size
    2048
  • qwen35.ssm.state_size
    128
  • qwen35.ssm.time_step_rank
    16
  • tokenizer.ggml.eos_token_id
    248046
  • tokenizer.ggml.merges
    [Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    248044
  • tokenizer.ggml.pre
    qwen35
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [1024, 248320]
    • blk.0
  • blk.0.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.0.attn_norm.weight
    F32
    [1024]
  • blk.0.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.0.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.0.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.0.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.0.post_attention_norm.weight
    F32
    [1024]
  • blk.0.ssm_a
    F32
    [16]
  • blk.0.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.0.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.0.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.0.ssm_dt.bias
    F32
    [16]
  • blk.0.ssm_norm.weight
    F32
    [128]
  • blk.0.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.1
  • blk.1.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.1.attn_norm.weight
    F32
    [1024]
  • blk.1.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.1.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.1.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.1.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.1.post_attention_norm.weight
    F32
    [1024]
  • blk.1.ssm_a
    F32
    [16]
  • blk.1.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.1.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.1.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.1.ssm_dt.bias
    F32
    [16]
  • blk.1.ssm_norm.weight
    F32
    [128]
  • blk.1.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.2
  • blk.2.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.2.attn_norm.weight
    F32
    [1024]
  • blk.2.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.2.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.2.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.2.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.2.post_attention_norm.weight
    F32
    [1024]
  • blk.2.ssm_a
    F32
    [16]
  • blk.2.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.2.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.2.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.2.ssm_dt.bias
    F32
    [16]
  • blk.2.ssm_norm.weight
    F32
    [128]
  • blk.2.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.3.attn_k_norm.weight
    F32
    [256]
  • blk.3.attn_norm.weight
    F32
    [1024]
  • blk.3.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.3.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [256]
  • blk.3.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.3.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.3.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.3.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.3.post_attention_norm.weight
    F32
    [1024]
    • blk.4
  • blk.4.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.4.attn_norm.weight
    F32
    [1024]
  • blk.4.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.4.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.4.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.4.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.4.post_attention_norm.weight
    F32
    [1024]
  • blk.4.ssm_a
    F32
    [16]
  • blk.4.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.4.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.4.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.4.ssm_dt.bias
    F32
    [16]
  • blk.4.ssm_norm.weight
    F32
    [128]
  • blk.4.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.5
  • blk.5.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.5.attn_norm.weight
    F32
    [1024]
  • blk.5.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.5.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.5.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.5.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.5.post_attention_norm.weight
    F32
    [1024]
  • blk.5.ssm_a
    F32
    [16]
  • blk.5.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.5.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.5.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.5.ssm_dt.bias
    F32
    [16]
  • blk.5.ssm_norm.weight
    F32
    [128]
  • blk.5.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.6
  • blk.6.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.6.attn_norm.weight
    F32
    [1024]
  • blk.6.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.6.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.6.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.6.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.6.post_attention_norm.weight
    F32
    [1024]
  • blk.6.ssm_a
    F32
    [16]
  • blk.6.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.6.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.6.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.6.ssm_dt.bias
    F32
    [16]
  • blk.6.ssm_norm.weight
    F32
    [128]
  • blk.6.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.7.attn_k_norm.weight
    F32
    [256]
  • blk.7.attn_norm.weight
    F32
    [1024]
  • blk.7.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.7.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.7.attn_q_norm.weight
    F32
    [256]
  • blk.7.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.7.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.7.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.7.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.7.post_attention_norm.weight
    F32
    [1024]
    • blk.8
  • blk.8.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.8.attn_norm.weight
    F32
    [1024]
  • blk.8.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.8.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.8.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.8.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.8.post_attention_norm.weight
    F32
    [1024]
  • blk.8.ssm_a
    F32
    [16]
  • blk.8.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.8.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.8.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.8.ssm_dt.bias
    F32
    [16]
  • blk.8.ssm_norm.weight
    F32
    [128]
  • blk.8.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.9
  • blk.9.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.9.attn_norm.weight
    F32
    [1024]
  • blk.9.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.9.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.9.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.9.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.9.post_attention_norm.weight
    F32
    [1024]
  • blk.9.ssm_a
    F32
    [16]
  • blk.9.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.9.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.9.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.9.ssm_dt.bias
    F32
    [16]
  • blk.9.ssm_norm.weight
    F32
    [128]
  • blk.9.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.10
  • blk.10.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.10.attn_norm.weight
    F32
    [1024]
  • blk.10.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.10.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.10.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.10.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.10.post_attention_norm.weight
    F32
    [1024]
  • blk.10.ssm_a
    F32
    [16]
  • blk.10.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.10.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.10.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.10.ssm_dt.bias
    F32
    [16]
  • blk.10.ssm_norm.weight
    F32
    [128]
  • blk.10.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.11.attn_k_norm.weight
    F32
    [256]
  • blk.11.attn_norm.weight
    F32
    [1024]
  • blk.11.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.11.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.11.attn_q_norm.weight
    F32
    [256]
  • blk.11.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.11.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.11.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.11.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.11.post_attention_norm.weight
    F32
    [1024]
    • blk.12
  • blk.12.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.12.attn_norm.weight
    F32
    [1024]
  • blk.12.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.12.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.12.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.12.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.12.post_attention_norm.weight
    F32
    [1024]
  • blk.12.ssm_a
    F32
    [16]
  • blk.12.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.12.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.12.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.12.ssm_dt.bias
    F32
    [16]
  • blk.12.ssm_norm.weight
    F32
    [128]
  • blk.12.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.13
  • blk.13.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.13.attn_norm.weight
    F32
    [1024]
  • blk.13.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.13.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.13.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.13.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.13.post_attention_norm.weight
    F32
    [1024]
  • blk.13.ssm_a
    F32
    [16]
  • blk.13.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.13.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.13.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.13.ssm_dt.bias
    F32
    [16]
  • blk.13.ssm_norm.weight
    F32
    [128]
  • blk.13.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.14
  • blk.14.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.14.attn_norm.weight
    F32
    [1024]
  • blk.14.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.14.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.14.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.14.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.14.post_attention_norm.weight
    F32
    [1024]
  • blk.14.ssm_a
    F32
    [16]
  • blk.14.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.14.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.14.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.14.ssm_dt.bias
    F32
    [16]
  • blk.14.ssm_norm.weight
    F32
    [128]
  • blk.14.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.15
  • blk.15.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.15.attn_k_norm.weight
    F32
    [256]
  • blk.15.attn_norm.weight
    F32
    [1024]
  • blk.15.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.15.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.15.attn_q_norm.weight
    F32
    [256]
  • blk.15.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.15.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.15.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.15.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.15.post_attention_norm.weight
    F32
    [1024]
    • blk.16
  • blk.16.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.16.attn_norm.weight
    F32
    [1024]
  • blk.16.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.16.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.16.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.16.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.16.post_attention_norm.weight
    F32
    [1024]
  • blk.16.ssm_a
    F32
    [16]
  • blk.16.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.16.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.16.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.16.ssm_dt.bias
    F32
    [16]
  • blk.16.ssm_norm.weight
    F32
    [128]
  • blk.16.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.17
  • blk.17.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.17.attn_norm.weight
    F32
    [1024]
  • blk.17.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.17.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.17.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.17.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.17.post_attention_norm.weight
    F32
    [1024]
  • blk.17.ssm_a
    F32
    [16]
  • blk.17.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.17.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.17.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.17.ssm_dt.bias
    F32
    [16]
  • blk.17.ssm_norm.weight
    F32
    [128]
  • blk.17.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.18
  • blk.18.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.18.attn_norm.weight
    F32
    [1024]
  • blk.18.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.18.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.18.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.18.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.18.post_attention_norm.weight
    F32
    [1024]
  • blk.18.ssm_a
    F32
    [16]
  • blk.18.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.18.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.18.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.18.ssm_dt.bias
    F32
    [16]
  • blk.18.ssm_norm.weight
    F32
    [128]
  • blk.18.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.19
  • blk.19.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.19.attn_k_norm.weight
    F32
    [256]
  • blk.19.attn_norm.weight
    F32
    [1024]
  • blk.19.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.19.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.19.attn_q_norm.weight
    F32
    [256]
  • blk.19.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.19.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.19.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.19.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.19.post_attention_norm.weight
    F32
    [1024]
    • blk.20
  • blk.20.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.20.attn_norm.weight
    F32
    [1024]
  • blk.20.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.20.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.20.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.20.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.20.post_attention_norm.weight
    F32
    [1024]
  • blk.20.ssm_a
    F32
    [16]
  • blk.20.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.20.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.20.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.20.ssm_dt.bias
    F32
    [16]
  • blk.20.ssm_norm.weight
    F32
    [128]
  • blk.20.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.21
  • blk.21.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.21.attn_norm.weight
    F32
    [1024]
  • blk.21.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.21.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.21.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.21.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.21.post_attention_norm.weight
    F32
    [1024]
  • blk.21.ssm_a
    F32
    [16]
  • blk.21.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.21.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.21.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.21.ssm_dt.bias
    F32
    [16]
  • blk.21.ssm_norm.weight
    F32
    [128]
  • blk.21.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.22
  • blk.22.attn_gate.weight
    Q8_0
    [1024, 2048]
  • blk.22.attn_norm.weight
    F32
    [1024]
  • blk.22.attn_qkv.weight
    Q8_0
    [1024, 6144]
  • blk.22.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.22.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.22.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.22.post_attention_norm.weight
    F32
    [1024]
  • blk.22.ssm_a
    F32
    [16]
  • blk.22.ssm_alpha.weight
    Q8_0
    [1024, 16]
  • blk.22.ssm_beta.weight
    Q8_0
    [1024, 16]
  • blk.22.ssm_conv1d.weight
    F32
    [4, 6144]
  • blk.22.ssm_dt.bias
    F32
    [16]
  • blk.22.ssm_norm.weight
    F32
    [128]
  • blk.22.ssm_out.weight
    Q8_0
    [2048, 1024]
    • blk.23
  • blk.23.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.23.attn_k_norm.weight
    F32
    [256]
  • blk.23.attn_norm.weight
    F32
    [1024]
  • blk.23.attn_output.weight
    Q8_0
    [2048, 1024]
  • blk.23.attn_q.weight
    Q8_0
    [1024, 4096]
  • blk.23.attn_q_norm.weight
    F32
    [256]
  • blk.23.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.23.ffn_down.weight
    Q8_0
    [3584, 1024]
  • blk.23.ffn_gate.weight
    Q8_0
    [1024, 3584]
  • blk.23.ffn_up.weight
    Q8_0
    [1024, 3584]
  • blk.23.post_attention_norm.weight
    F32
    [1024]
  • output_norm.weight
    F32
    [1024]