gpt-oss:120b

1.7M 2 weeks ago

OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases.

tools thinking 20b 120b
90a618fe6ff2 · 65GB
    Metadata
  • general.architecture
    gptoss
  • general.file_type
    MXFP4
  • gptoss.attention.head_count
    64
  • gptoss.attention.head_count_kv
    8
  • gptoss.attention.key_length
    64
  • gptoss.attention.layer_norm_rms_epsilon
    1e-05
  • gptoss.attention.sliding_window
    128
  • gptoss.attention.value_length
    64
  • gptoss.block_count
    36
  • gptoss.context_length
    131072
  • gptoss.embedding_length
    2880
  • gptoss.expert_count
    128
  • gptoss.expert_used_count
    4
  • gptoss.feed_forward_length
    2880
  • gptoss.rope.freq_base
    150000
  • gptoss.rope.scaling.factor
    32
  • gptoss.rope.scaling.original_context_length
    4096
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    199998
  • tokenizer.ggml.eos_token_id
    199999
  • tokenizer.ggml.eos_token_ids
    [199999, 200002, 200012]
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [0, 1, 2, 3, 4, ...]
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
  • Tensor
  • token_embd.weight
    BF16
    [2880, 201088]
  • blk.0
  • blk.0.attn_norm.weight
    F32
    [2880]
  • blk.0.attn_out.bias
    F32
    [2880]
  • blk.0.attn_out.weight
    BF16
    [4096, 2880]
  • blk.0.attn_qkv.bias
    F32
    [5120]
  • blk.0.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.0.attn_sinks
    F32
    [64]
  • blk.0.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.0.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.0.ffn_gate_inp.bias
    F32
    [128]
  • blk.0.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.0.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.0.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.0.ffn_norm.weight
    F32
    [2880]
  • blk.1
  • blk.1.attn_norm.weight
    F32
    [2880]
  • blk.1.attn_out.bias
    F32
    [2880]
  • blk.1.attn_out.weight
    BF16
    [4096, 2880]
  • blk.1.attn_qkv.bias
    F32
    [5120]
  • blk.1.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.1.attn_sinks
    F32
    [64]
  • blk.1.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.1.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.1.ffn_gate_inp.bias
    F32
    [128]
  • blk.1.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.1.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.1.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.1.ffn_norm.weight
    F32
    [2880]
  • blk.2
  • blk.2.attn_norm.weight
    F32
    [2880]
  • blk.2.attn_out.bias
    F32
    [2880]
  • blk.2.attn_out.weight
    BF16
    [4096, 2880]
  • blk.2.attn_qkv.bias
    F32
    [5120]
  • blk.2.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.2.attn_sinks
    F32
    [64]
  • blk.2.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.2.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.2.ffn_gate_inp.bias
    F32
    [128]
  • blk.2.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.2.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.2.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.2.ffn_norm.weight
    F32
    [2880]
  • blk.3
  • blk.3.attn_norm.weight
    F32
    [2880]
  • blk.3.attn_out.bias
    F32
    [2880]
  • blk.3.attn_out.weight
    BF16
    [4096, 2880]
  • blk.3.attn_qkv.bias
    F32
    [5120]
  • blk.3.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.3.attn_sinks
    F32
    [64]
  • blk.3.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.3.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.3.ffn_gate_inp.bias
    F32
    [128]
  • blk.3.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.3.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.3.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.3.ffn_norm.weight
    F32
    [2880]
  • blk.4
  • blk.4.attn_norm.weight
    F32
    [2880]
  • blk.4.attn_out.bias
    F32
    [2880]
  • blk.4.attn_out.weight
    BF16
    [4096, 2880]
  • blk.4.attn_qkv.bias
    F32
    [5120]
  • blk.4.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.4.attn_sinks
    F32
    [64]
  • blk.4.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.4.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.4.ffn_gate_inp.bias
    F32
    [128]
  • blk.4.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.4.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.4.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.4.ffn_norm.weight
    F32
    [2880]
  • blk.5
  • blk.5.attn_norm.weight
    F32
    [2880]
  • blk.5.attn_out.bias
    F32
    [2880]
  • blk.5.attn_out.weight
    BF16
    [4096, 2880]
  • blk.5.attn_qkv.bias
    F32
    [5120]
  • blk.5.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.5.attn_sinks
    F32
    [64]
  • blk.5.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.5.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.5.ffn_gate_inp.bias
    F32
    [128]
  • blk.5.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.5.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.5.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.5.ffn_norm.weight
    F32
    [2880]
  • blk.6
  • blk.6.attn_norm.weight
    F32
    [2880]
  • blk.6.attn_out.bias
    F32
    [2880]
  • blk.6.attn_out.weight
    BF16
    [4096, 2880]
  • blk.6.attn_qkv.bias
    F32
    [5120]
  • blk.6.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.6.attn_sinks
    F32
    [64]
  • blk.6.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.6.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.6.ffn_gate_inp.bias
    F32
    [128]
  • blk.6.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.6.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.6.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.6.ffn_norm.weight
    F32
    [2880]
  • blk.7
  • blk.7.attn_norm.weight
    F32
    [2880]
  • blk.7.attn_out.bias
    F32
    [2880]
  • blk.7.attn_out.weight
    BF16
    [4096, 2880]
  • blk.7.attn_qkv.bias
    F32
    [5120]
  • blk.7.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.7.attn_sinks
    F32
    [64]
  • blk.7.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.7.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.7.ffn_gate_inp.bias
    F32
    [128]
  • blk.7.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.7.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.7.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.7.ffn_norm.weight
    F32
    [2880]
  • blk.8
  • blk.8.attn_norm.weight
    F32
    [2880]
  • blk.8.attn_out.bias
    F32
    [2880]
  • blk.8.attn_out.weight
    BF16
    [4096, 2880]
  • blk.8.attn_qkv.bias
    F32
    [5120]
  • blk.8.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.8.attn_sinks
    F32
    [64]
  • blk.8.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.8.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.8.ffn_gate_inp.bias
    F32
    [128]
  • blk.8.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.8.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.8.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.8.ffn_norm.weight
    F32
    [2880]
  • blk.9
  • blk.9.attn_norm.weight
    F32
    [2880]
  • blk.9.attn_out.bias
    F32
    [2880]
  • blk.9.attn_out.weight
    BF16
    [4096, 2880]
  • blk.9.attn_qkv.bias
    F32
    [5120]
  • blk.9.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.9.attn_sinks
    F32
    [64]
  • blk.9.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.9.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.9.ffn_gate_inp.bias
    F32
    [128]
  • blk.9.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.9.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.9.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.9.ffn_norm.weight
    F32
    [2880]
  • blk.10
  • blk.10.attn_norm.weight
    F32
    [2880]
  • blk.10.attn_out.bias
    F32
    [2880]
  • blk.10.attn_out.weight
    BF16
    [4096, 2880]
  • blk.10.attn_qkv.bias
    F32
    [5120]
  • blk.10.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.10.attn_sinks
    F32
    [64]
  • blk.10.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.10.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.10.ffn_gate_inp.bias
    F32
    [128]
  • blk.10.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.10.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.10.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.10.ffn_norm.weight
    F32
    [2880]
  • blk.11
  • blk.11.attn_norm.weight
    F32
    [2880]
  • blk.11.attn_out.bias
    F32
    [2880]
  • blk.11.attn_out.weight
    BF16
    [4096, 2880]
  • blk.11.attn_qkv.bias
    F32
    [5120]
  • blk.11.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.11.attn_sinks
    F32
    [64]
  • blk.11.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.11.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.11.ffn_gate_inp.bias
    F32
    [128]
  • blk.11.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.11.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.11.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.11.ffn_norm.weight
    F32
    [2880]
  • blk.12
  • blk.12.attn_norm.weight
    F32
    [2880]
  • blk.12.attn_out.bias
    F32
    [2880]
  • blk.12.attn_out.weight
    BF16
    [4096, 2880]
  • blk.12.attn_qkv.bias
    F32
    [5120]
  • blk.12.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.12.attn_sinks
    F32
    [64]
  • blk.12.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.12.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.12.ffn_gate_inp.bias
    F32
    [128]
  • blk.12.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.12.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.12.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.12.ffn_norm.weight
    F32
    [2880]
  • blk.13
  • blk.13.attn_norm.weight
    F32
    [2880]
  • blk.13.attn_out.bias
    F32
    [2880]
  • blk.13.attn_out.weight
    BF16
    [4096, 2880]
  • blk.13.attn_qkv.bias
    F32
    [5120]
  • blk.13.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.13.attn_sinks
    F32
    [64]
  • blk.13.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.13.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.13.ffn_gate_inp.bias
    F32
    [128]
  • blk.13.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.13.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.13.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.13.ffn_norm.weight
    F32
    [2880]
  • blk.14
  • blk.14.attn_norm.weight
    F32
    [2880]
  • blk.14.attn_out.bias
    F32
    [2880]
  • blk.14.attn_out.weight
    BF16
    [4096, 2880]
  • blk.14.attn_qkv.bias
    F32
    [5120]
  • blk.14.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.14.attn_sinks
    F32
    [64]
  • blk.14.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.14.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.14.ffn_gate_inp.bias
    F32
    [128]
  • blk.14.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.14.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.14.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.14.ffn_norm.weight
    F32
    [2880]
  • blk.15
  • blk.15.attn_norm.weight
    F32
    [2880]
  • blk.15.attn_out.bias
    F32
    [2880]
  • blk.15.attn_out.weight
    BF16
    [4096, 2880]
  • blk.15.attn_qkv.bias
    F32
    [5120]
  • blk.15.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.15.attn_sinks
    F32
    [64]
  • blk.15.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.15.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.15.ffn_gate_inp.bias
    F32
    [128]
  • blk.15.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.15.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.15.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.15.ffn_norm.weight
    F32
    [2880]
  • blk.16
  • blk.16.attn_norm.weight
    F32
    [2880]
  • blk.16.attn_out.bias
    F32
    [2880]
  • blk.16.attn_out.weight
    BF16
    [4096, 2880]
  • blk.16.attn_qkv.bias
    F32
    [5120]
  • blk.16.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.16.attn_sinks
    F32
    [64]
  • blk.16.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.16.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.16.ffn_gate_inp.bias
    F32
    [128]
  • blk.16.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.16.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.16.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.16.ffn_norm.weight
    F32
    [2880]
  • blk.17
  • blk.17.attn_norm.weight
    F32
    [2880]
  • blk.17.attn_out.bias
    F32
    [2880]
  • blk.17.attn_out.weight
    BF16
    [4096, 2880]
  • blk.17.attn_qkv.bias
    F32
    [5120]
  • blk.17.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.17.attn_sinks
    F32
    [64]
  • blk.17.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.17.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.17.ffn_gate_inp.bias
    F32
    [128]
  • blk.17.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.17.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.17.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.17.ffn_norm.weight
    F32
    [2880]
  • blk.18
  • blk.18.attn_norm.weight
    F32
    [2880]
  • blk.18.attn_out.bias
    F32
    [2880]
  • blk.18.attn_out.weight
    BF16
    [4096, 2880]
  • blk.18.attn_qkv.bias
    F32
    [5120]
  • blk.18.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.18.attn_sinks
    F32
    [64]
  • blk.18.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.18.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.18.ffn_gate_inp.bias
    F32
    [128]
  • blk.18.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.18.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.18.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.18.ffn_norm.weight
    F32
    [2880]
  • blk.19
  • blk.19.attn_norm.weight
    F32
    [2880]
  • blk.19.attn_out.bias
    F32
    [2880]
  • blk.19.attn_out.weight
    BF16
    [4096, 2880]
  • blk.19.attn_qkv.bias
    F32
    [5120]
  • blk.19.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.19.attn_sinks
    F32
    [64]
  • blk.19.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.19.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.19.ffn_gate_inp.bias
    F32
    [128]
  • blk.19.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.19.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.19.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.19.ffn_norm.weight
    F32
    [2880]
  • blk.20
  • blk.20.attn_norm.weight
    F32
    [2880]
  • blk.20.attn_out.bias
    F32
    [2880]
  • blk.20.attn_out.weight
    BF16
    [4096, 2880]
  • blk.20.attn_qkv.bias
    F32
    [5120]
  • blk.20.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.20.attn_sinks
    F32
    [64]
  • blk.20.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.20.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.20.ffn_gate_inp.bias
    F32
    [128]
  • blk.20.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.20.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.20.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.20.ffn_norm.weight
    F32
    [2880]
  • blk.21
  • blk.21.attn_norm.weight
    F32
    [2880]
  • blk.21.attn_out.bias
    F32
    [2880]
  • blk.21.attn_out.weight
    BF16
    [4096, 2880]
  • blk.21.attn_qkv.bias
    F32
    [5120]
  • blk.21.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.21.attn_sinks
    F32
    [64]
  • blk.21.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.21.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.21.ffn_gate_inp.bias
    F32
    [128]
  • blk.21.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.21.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.21.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.21.ffn_norm.weight
    F32
    [2880]
  • blk.22
  • blk.22.attn_norm.weight
    F32
    [2880]
  • blk.22.attn_out.bias
    F32
    [2880]
  • blk.22.attn_out.weight
    BF16
    [4096, 2880]
  • blk.22.attn_qkv.bias
    F32
    [5120]
  • blk.22.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.22.attn_sinks
    F32
    [64]
  • blk.22.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.22.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.22.ffn_gate_inp.bias
    F32
    [128]
  • blk.22.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.22.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.22.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.22.ffn_norm.weight
    F32
    [2880]
  • blk.23
  • blk.23.attn_norm.weight
    F32
    [2880]
  • blk.23.attn_out.bias
    F32
    [2880]
  • blk.23.attn_out.weight
    BF16
    [4096, 2880]
  • blk.23.attn_qkv.bias
    F32
    [5120]
  • blk.23.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.23.attn_sinks
    F32
    [64]
  • blk.23.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.23.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.23.ffn_gate_inp.bias
    F32
    [128]
  • blk.23.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.23.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.23.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.23.ffn_norm.weight
    F32
    [2880]
  • blk.24
  • blk.24.attn_norm.weight
    F32
    [2880]
  • blk.24.attn_out.bias
    F32
    [2880]
  • blk.24.attn_out.weight
    BF16
    [4096, 2880]
  • blk.24.attn_qkv.bias
    F32
    [5120]
  • blk.24.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.24.attn_sinks
    F32
    [64]
  • blk.24.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.24.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.24.ffn_gate_inp.bias
    F32
    [128]
  • blk.24.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.24.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.24.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.24.ffn_norm.weight
    F32
    [2880]
  • blk.25
  • blk.25.attn_norm.weight
    F32
    [2880]
  • blk.25.attn_out.bias
    F32
    [2880]
  • blk.25.attn_out.weight
    BF16
    [4096, 2880]
  • blk.25.attn_qkv.bias
    F32
    [5120]
  • blk.25.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.25.attn_sinks
    F32
    [64]
  • blk.25.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.25.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.25.ffn_gate_inp.bias
    F32
    [128]
  • blk.25.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.25.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.25.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.25.ffn_norm.weight
    F32
    [2880]
  • blk.26
  • blk.26.attn_norm.weight
    F32
    [2880]
  • blk.26.attn_out.bias
    F32
    [2880]
  • blk.26.attn_out.weight
    BF16
    [4096, 2880]
  • blk.26.attn_qkv.bias
    F32
    [5120]
  • blk.26.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.26.attn_sinks
    F32
    [64]
  • blk.26.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.26.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.26.ffn_gate_inp.bias
    F32
    [128]
  • blk.26.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.26.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.26.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.26.ffn_norm.weight
    F32
    [2880]
  • blk.27
  • blk.27.attn_norm.weight
    F32
    [2880]
  • blk.27.attn_out.bias
    F32
    [2880]
  • blk.27.attn_out.weight
    BF16
    [4096, 2880]
  • blk.27.attn_qkv.bias
    F32
    [5120]
  • blk.27.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.27.attn_sinks
    F32
    [64]
  • blk.27.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.27.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.27.ffn_gate_inp.bias
    F32
    [128]
  • blk.27.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.27.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.27.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.27.ffn_norm.weight
    F32
    [2880]
  • blk.28
  • blk.28.attn_norm.weight
    F32
    [2880]
  • blk.28.attn_out.bias
    F32
    [2880]
  • blk.28.attn_out.weight
    BF16
    [4096, 2880]
  • blk.28.attn_qkv.bias
    F32
    [5120]
  • blk.28.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.28.attn_sinks
    F32
    [64]
  • blk.28.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.28.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.28.ffn_gate_inp.bias
    F32
    [128]
  • blk.28.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.28.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.28.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.28.ffn_norm.weight
    F32
    [2880]
  • blk.29
  • blk.29.attn_norm.weight
    F32
    [2880]
  • blk.29.attn_out.bias
    F32
    [2880]
  • blk.29.attn_out.weight
    BF16
    [4096, 2880]
  • blk.29.attn_qkv.bias
    F32
    [5120]
  • blk.29.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.29.attn_sinks
    F32
    [64]
  • blk.29.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.29.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.29.ffn_gate_inp.bias
    F32
    [128]
  • blk.29.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.29.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.29.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.29.ffn_norm.weight
    F32
    [2880]
  • blk.30
  • blk.30.attn_norm.weight
    F32
    [2880]
  • blk.30.attn_out.bias
    F32
    [2880]
  • blk.30.attn_out.weight
    BF16
    [4096, 2880]
  • blk.30.attn_qkv.bias
    F32
    [5120]
  • blk.30.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.30.attn_sinks
    F32
    [64]
  • blk.30.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.30.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.30.ffn_gate_inp.bias
    F32
    [128]
  • blk.30.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.30.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.30.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.30.ffn_norm.weight
    F32
    [2880]
  • blk.31
  • blk.31.attn_norm.weight
    F32
    [2880]
  • blk.31.attn_out.bias
    F32
    [2880]
  • blk.31.attn_out.weight
    BF16
    [4096, 2880]
  • blk.31.attn_qkv.bias
    F32
    [5120]
  • blk.31.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.31.attn_sinks
    F32
    [64]
  • blk.31.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.31.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.31.ffn_gate_inp.bias
    F32
    [128]
  • blk.31.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.31.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.31.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.31.ffn_norm.weight
    F32
    [2880]
  • blk.32
  • blk.32.attn_norm.weight
    F32
    [2880]
  • blk.32.attn_out.bias
    F32
    [2880]
  • blk.32.attn_out.weight
    BF16
    [4096, 2880]
  • blk.32.attn_qkv.bias
    F32
    [5120]
  • blk.32.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.32.attn_sinks
    F32
    [64]
  • blk.32.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.32.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.32.ffn_gate_inp.bias
    F32
    [128]
  • blk.32.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.32.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.32.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.32.ffn_norm.weight
    F32
    [2880]
  • blk.33
  • blk.33.attn_norm.weight
    F32
    [2880]
  • blk.33.attn_out.bias
    F32
    [2880]
  • blk.33.attn_out.weight
    BF16
    [4096, 2880]
  • blk.33.attn_qkv.bias
    F32
    [5120]
  • blk.33.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.33.attn_sinks
    F32
    [64]
  • blk.33.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.33.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.33.ffn_gate_inp.bias
    F32
    [128]
  • blk.33.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.33.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.33.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.33.ffn_norm.weight
    F32
    [2880]
  • blk.34
  • blk.34.attn_norm.weight
    F32
    [2880]
  • blk.34.attn_out.bias
    F32
    [2880]
  • blk.34.attn_out.weight
    BF16
    [4096, 2880]
  • blk.34.attn_qkv.bias
    F32
    [5120]
  • blk.34.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.34.attn_sinks
    F32
    [64]
  • blk.34.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.34.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.34.ffn_gate_inp.bias
    F32
    [128]
  • blk.34.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.34.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.34.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.34.ffn_norm.weight
    F32
    [2880]
  • blk.35
  • blk.35.attn_norm.weight
    F32
    [2880]
  • blk.35.attn_out.bias
    F32
    [2880]
  • blk.35.attn_out.weight
    BF16
    [4096, 2880]
  • blk.35.attn_qkv.bias
    F32
    [5120]
  • blk.35.attn_qkv.weight
    BF16
    [2880, 5120]
  • blk.35.attn_sinks
    F32
    [64]
  • blk.35.ffn_down_exps.bias
    BF16
    [2880, 128]
  • blk.35.ffn_down_exps.weight
    MXFP4
    [2880, 2880, 128]
  • blk.35.ffn_gate_inp.bias
    F32
    [128]
  • blk.35.ffn_gate_inp.weight
    F32
    [2880, 128]
  • blk.35.ffn_gate_up_exps.bias
    BF16
    [5760, 128]
  • blk.35.ffn_gate_up_exps.weight
    MXFP4
    [2880, 5760, 128]
  • blk.35.ffn_norm.weight
    F32
    [2880]
  • output.weight
    BF16
    [2880, 201088]
  • output_norm.weight
    F32
    [2880]