oza6ut0ne/ d1-omni-600M:Q8_0

2 2 days ago

vision
... /
model
cd94463f4cac · 407MB
    Metadata
  • general.architecture
    lfm2
  • general.file_type
    Q8_0
  • lfm2.attention.causal
    false
  • lfm2.attention.head_count
    16
  • lfm2.attention.head_count_kv
    [0, 0, 8, 0, 0, ...]
  • lfm2.attention.layer_norm_epsilon
    1e-05
  • lfm2.attention.layer_norm_rms_epsilon
    1e-05
  • lfm2.block_count
    18
  • lfm2.context_length
    128000
  • lfm2.decision.block_count
    2
  • lfm2.decision.temperature.choice
    1
  • lfm2.decision.temperature.choice.11
    1.3725151
  • lfm2.decision.temperature.choice.2
    1.7465146
  • lfm2.decision.temperature.choice.3_5
    1.3998982
  • lfm2.decision.temperature.choice.6_10
    1.1751071
  • lfm2.decision.temperature.noul
    1
  • lfm2.decision.temperature.noul.2
    1.6663224
  • lfm2.decision.temperature.score
    1
  • lfm2.decision.temperature.score.3_5
    1.7301133
  • lfm2.decision.temperature.score.6_10
    1
  • lfm2.decision.type
    lfm2-d1-omni
  • lfm2.embedding_length
    1024
  • lfm2.feed_forward_length
    [4608, 4608, 4608, 4608, 4608, ...]
  • lfm2.rope.freq_base
    1e+06
  • lfm2.shortconv.l_cache
    3
  • lfm2.vocab_size
    65536
  • tokenizer.chat_template.systemone
    {% set ns = namespace(criteria=false) %}{% for o in options %}{% if o.description is not none %}{% set ns.criteria = true %}{% endif %}{% endfor %}{% for image in images %}{{ image }}{% endfor %}{{ sep }}<|startoftext|><|reserved_7|>{{ sep }}{{ mark_state }}{% if state is string %}{{ state }}{% elif state is not none %}{{ state | tojson }}{% elif audio %}{}{% endif %}{{ sep }}{{ mark_question }}<|reserved_8|>{{ instructions if instructions is string else instructions | tojson }}{% for o in options %}{{ sep }}<|reserved_9|><|mask|>{{ sep }}{{ mark_option }} {% if type == 'choice' and audio %}option_{% if loop.index0 < 10 %}00{% elif loop.index0 < 100 %}0{% endif %}{{ loop.index0 }}: {% if o.description is not none and o.description != '' %}{{ o.description if o.description is string else o.description | tojson }}{% else %}{{ o.key }}{% endif %}{% elif type == 'choice' %}{{ o.key }}{% if o.description is not none and o.description != '' %}: {{ o.description if o.description is string else o.description | tojson }}{% endif %}{% elif type == 'score' %}level {{ o.key }}: {{ o.description if o.description is string else o.description | tojson }}{% elif audio %}{{ o.key }}: {{ 'yes' if o.key == 'true' else 'no' }}{% else %}{{ o.key }}: {% if o.description is not none and o.description != '' %}{{ o.description if o.description is string else o.description | tojson }}{% elif images and not ns.criteria %}{{ 'yes' if o.key == 'true' else 'no' }}{% elif o.key == 'true' %}yes, the statement holds{% else %}no, the statement does not hold{% endif %}{% endif %}{{ sep }}<|reserved_10|>{% endfor %}{{ sep }}<|reserved_11|>
  • tokenizer.chat_templates
    [systemone]
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    7
  • tokenizer.ggml.mask_token_id
    16
  • tokenizer.ggml.merges
    [Ċ Ċ, Ċ ĊĊ, ĊĊ Ċ, Ċ ĊĊĊ, ĊĊ ĊĊ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    lfm2
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.token_type_count
    3
  • tokenizer.ggml.tokens
    [<|pad|>, <|startoftext|>, <|endoftext|>, <|fim_pre|>, <|fim_mid|>, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [1024, 65536]
    • blk.0
  • blk.0.attn_norm.weight
    F32
    [1024]
  • blk.0.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.0.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.0.ffn_norm.weight
    F32
    [1024]
  • blk.0.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.0.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.0.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.0.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.1
  • blk.1.attn_norm.weight
    F32
    [1024]
  • blk.1.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.1.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.1.ffn_norm.weight
    F32
    [1024]
  • blk.1.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.1.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.1.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.1.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.2.attn_k_norm.weight
    F32
    [64]
  • blk.2.attn_norm.weight
    F32
    [1024]
  • blk.2.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.2.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.2.attn_q_norm.weight
    F32
    [64]
  • blk.2.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.2.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.2.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.2.ffn_norm.weight
    F32
    [1024]
  • blk.2.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.3
  • blk.3.attn_norm.weight
    F32
    [1024]
  • blk.3.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.3.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.3.ffn_norm.weight
    F32
    [1024]
  • blk.3.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.3.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.3.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.3.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.4
  • blk.4.attn_norm.weight
    F32
    [1024]
  • blk.4.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.4.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.4.ffn_norm.weight
    F32
    [1024]
  • blk.4.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.4.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.4.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.4.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.5.attn_k_norm.weight
    F32
    [64]
  • blk.5.attn_norm.weight
    F32
    [1024]
  • blk.5.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.5.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.5.attn_q_norm.weight
    F32
    [64]
  • blk.5.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.5.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.5.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.5.ffn_norm.weight
    F32
    [1024]
  • blk.5.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.6
  • blk.6.attn_norm.weight
    F32
    [1024]
  • blk.6.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.6.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.6.ffn_norm.weight
    F32
    [1024]
  • blk.6.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.6.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.6.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.6.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.7
  • blk.7.attn_norm.weight
    F32
    [1024]
  • blk.7.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.7.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.7.ffn_norm.weight
    F32
    [1024]
  • blk.7.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.7.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.7.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.7.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.8.attn_k_norm.weight
    F32
    [64]
  • blk.8.attn_norm.weight
    F32
    [1024]
  • blk.8.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.8.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.8.attn_q_norm.weight
    F32
    [64]
  • blk.8.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.8.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.8.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.8.ffn_norm.weight
    F32
    [1024]
  • blk.8.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.9
  • blk.9.attn_norm.weight
    F32
    [1024]
  • blk.9.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.9.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.9.ffn_norm.weight
    F32
    [1024]
  • blk.9.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.9.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.9.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.9.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.10.attn_k_norm.weight
    F32
    [64]
  • blk.10.attn_norm.weight
    F32
    [1024]
  • blk.10.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.10.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.10.attn_q_norm.weight
    F32
    [64]
  • blk.10.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.10.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.10.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.10.ffn_norm.weight
    F32
    [1024]
  • blk.10.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.11
  • blk.11.attn_norm.weight
    F32
    [1024]
  • blk.11.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.11.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.11.ffn_norm.weight
    F32
    [1024]
  • blk.11.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.11.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.11.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.11.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.12
  • blk.12.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.12.attn_k_norm.weight
    F32
    [64]
  • blk.12.attn_norm.weight
    F32
    [1024]
  • blk.12.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.12.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.12.attn_q_norm.weight
    F32
    [64]
  • blk.12.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.12.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.12.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.12.ffn_norm.weight
    F32
    [1024]
  • blk.12.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.13
  • blk.13.attn_norm.weight
    F32
    [1024]
  • blk.13.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.13.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.13.ffn_norm.weight
    F32
    [1024]
  • blk.13.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.13.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.13.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.13.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.14
  • blk.14.attn_k.weight
    Q8_0
    [1024, 512]
  • blk.14.attn_k_norm.weight
    F32
    [64]
  • blk.14.attn_norm.weight
    F32
    [1024]
  • blk.14.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.14.attn_q.weight
    Q8_0
    [1024, 1024]
  • blk.14.attn_q_norm.weight
    F32
    [64]
  • blk.14.attn_v.weight
    Q8_0
    [1024, 512]
  • blk.14.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.14.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.14.ffn_norm.weight
    F32
    [1024]
  • blk.14.ffn_up.weight
    Q8_0
    [1024, 4608]
    • blk.15
  • blk.15.attn_norm.weight
    F32
    [1024]
  • blk.15.ffn_down.weight
    Q8_0
    [4608, 1024]
  • blk.15.ffn_gate.weight
    Q8_0
    [1024, 4608]
  • blk.15.ffn_norm.weight
    F32
    [1024]
  • blk.15.ffn_up.weight
    Q8_0
    [1024, 4608]
  • blk.15.shortconv.conv.weight
    F32
    [3, 1024]
  • blk.15.shortconv.in_proj.weight
    Q8_0
    [1024, 3072]
  • blk.15.shortconv.out_proj.weight
    Q8_0
    [1024, 1024]
    • blk.16
  • blk.16.attn_norm.bias
    F32
    [1024]
  • blk.16.attn_norm.weight
    F32
    [1024]
  • blk.16.attn_output.bias
    F32
    [1024]
  • blk.16.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.16.attn_qkv.bias
    F32
    [3072]
  • blk.16.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.16.ffn_down.bias
    F32
    [1024]
  • blk.16.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.16.ffn_norm.bias
    F32
    [1024]
  • blk.16.ffn_norm.weight
    F32
    [1024]
  • blk.16.ffn_up.bias
    F32
    [4096]
  • blk.16.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.17
  • blk.17.attn_norm.bias
    F32
    [1024]
  • blk.17.attn_norm.weight
    F32
    [1024]
  • blk.17.attn_output.bias
    F32
    [1024]
  • blk.17.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.17.attn_qkv.bias
    F32
    [3072]
  • blk.17.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.17.ffn_down.bias
    F32
    [1024]
  • blk.17.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.17.ffn_norm.bias
    F32
    [1024]
  • blk.17.ffn_norm.weight
    F32
    [1024]
  • blk.17.ffn_up.bias
    F32
    [4096]
  • blk.17.ffn_up.weight
    Q8_0
    [1024, 4096]
  • cls.bias
    F32
    [1024]
  • cls.norm.bias
    F32
    [1024]
  • cls.norm.weight
    F32
    [1024]
  • cls.output.bias
    F32
    [1]
  • cls.output.weight
    F32
    [1024]
  • cls.weight
    Q8_0
    [1024, 1024]
  • token_embd_norm.weight
    F32
    [1024]
  • token_types.weight
    F32
    [1024, 3]