Models
GitHub
Discord
Docs
Cloud
Sign in
Download
Models
Download
GitHub
Discord
Docs
Cloud
Sign in
granite4
:350m-h-q8_0
216.4K
Downloads
Updated
2 weeks ago
Granite 4 features improved instruction following (IF) and tool-calling capabilities, making them more effective in enterprise applications.
Granite 4 features improved instruction following (IF) and tool-calling capabilities, making them more effective in enterprise applications.
Cancel
tools
350m
1b
3b
granite4:350m-h-q8_0
...
/
model
c7d9873640dc · 366MB
Metadata
general.architecture
granitehybrid
granitehybrid
general.file_type
Q8_0
Q8_0
granitehybrid.attention.head_count
12
12
granitehybrid.attention.head_count_kv
[0, 0, 0, 0, 0, ...]
[0, 0, 0, 0, 0, ...]
granitehybrid.attention.layer_norm_rms_epsilon
1e-05
1e-05
granitehybrid.attention.scale
0.015625
0.015625
granitehybrid.block_count
32
32
granitehybrid.context_length
1048576
1048576
granitehybrid.embedding_length
768
768
granitehybrid.embedding_scale
12
12
granitehybrid.expert_count
0
0
granitehybrid.expert_shared_feed_forward_length
2048
2048
granitehybrid.expert_used_count
0
0
granitehybrid.feed_forward_length
2048
2048
granitehybrid.logit_scale
3
3
granitehybrid.residual_scale
0.246
0.246
granitehybrid.rope.dimension_count
64
64
granitehybrid.rope.freq_base
10000
10000
granitehybrid.rope.scaling.finetuned
false
false
granitehybrid.ssm.conv_kernel
4
4
granitehybrid.ssm.group_count
1
1
granitehybrid.ssm.inner_size
1536
1536
granitehybrid.ssm.state_size
128
128
granitehybrid.ssm.time_step_rank
48
48
granitehybrid.vocab_size
100352
100352
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.bos_token_id
100257
100257
tokenizer.ggml.eos_token_id
100257
100257
tokenizer.ggml.merges
[Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
[Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
100256
100256
tokenizer.ggml.pre
dbrx
dbrx
tokenizer.ggml.token_type
[1, 1, 1, 1, 1, ...]
[1, 1, 1, 1, 1, ...]
tokenizer.ggml.tokens
[!, ", #, $, %, ...]
[!, ", #, $, %, ...]
tokenizer.ggml.unknown_token_id
100269
100269
Tensor
Name
Type
Shape
token_embd.weight
Q8_0
Q8_0
[768, 100352]
blk.0
blk.0.attn_norm.weight
F32
F32
[768]
blk.0.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.0.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.0.ffn_norm.weight
F32
F32
[768]
blk.0.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.0.ssm_a
F32
F32
[1, 48]
blk.0.ssm_conv1d.bias
F32
F32
[1792]
blk.0.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.0.ssm_d
F32
F32
[1, 48]
blk.0.ssm_dt.bias
F32
F32
[48]
blk.0.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.0.ssm_norm.weight
F32
F32
[1536]
blk.0.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.1
blk.1.attn_norm.weight
F32
F32
[768]
blk.1.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.1.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.1.ffn_norm.weight
F32
F32
[768]
blk.1.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.1.ssm_a
F32
F32
[1, 48]
blk.1.ssm_conv1d.bias
F32
F32
[1792]
blk.1.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.1.ssm_d
F32
F32
[1, 48]
blk.1.ssm_dt.bias
F32
F32
[48]
blk.1.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.1.ssm_norm.weight
F32
F32
[1536]
blk.1.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.2
blk.2.attn_norm.weight
F32
F32
[768]
blk.2.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.2.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.2.ffn_norm.weight
F32
F32
[768]
blk.2.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.2.ssm_a
F32
F32
[1, 48]
blk.2.ssm_conv1d.bias
F32
F32
[1792]
blk.2.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.2.ssm_d
F32
F32
[1, 48]
blk.2.ssm_dt.bias
F32
F32
[48]
blk.2.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.2.ssm_norm.weight
F32
F32
[1536]
blk.2.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.3
blk.3.attn_norm.weight
F32
F32
[768]
blk.3.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.3.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.3.ffn_norm.weight
F32
F32
[768]
blk.3.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.3.ssm_a
F32
F32
[1, 48]
blk.3.ssm_conv1d.bias
F32
F32
[1792]
blk.3.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.3.ssm_d
F32
F32
[1, 48]
blk.3.ssm_dt.bias
F32
F32
[48]
blk.3.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.3.ssm_norm.weight
F32
F32
[1536]
blk.3.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.4
blk.4.attn_norm.weight
F32
F32
[768]
blk.4.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.4.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.4.ffn_norm.weight
F32
F32
[768]
blk.4.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.4.ssm_a
F32
F32
[1, 48]
blk.4.ssm_conv1d.bias
F32
F32
[1792]
blk.4.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.4.ssm_d
F32
F32
[1, 48]
blk.4.ssm_dt.bias
F32
F32
[48]
blk.4.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.4.ssm_norm.weight
F32
F32
[1536]
blk.4.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.5
blk.5.attn_norm.weight
F32
F32
[768]
blk.5.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.5.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.5.ffn_norm.weight
F32
F32
[768]
blk.5.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.5.ssm_a
F32
F32
[1, 48]
blk.5.ssm_conv1d.bias
F32
F32
[1792]
blk.5.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.5.ssm_d
F32
F32
[1, 48]
blk.5.ssm_dt.bias
F32
F32
[48]
blk.5.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.5.ssm_norm.weight
F32
F32
[1536]
blk.5.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.6
blk.6.attn_norm.weight
F32
F32
[768]
blk.6.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.6.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.6.ffn_norm.weight
F32
F32
[768]
blk.6.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.6.ssm_a
F32
F32
[1, 48]
blk.6.ssm_conv1d.bias
F32
F32
[1792]
blk.6.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.6.ssm_d
F32
F32
[1, 48]
blk.6.ssm_dt.bias
F32
F32
[48]
blk.6.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.6.ssm_norm.weight
F32
F32
[1536]
blk.6.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.7
blk.7.attn_norm.weight
F32
F32
[768]
blk.7.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.7.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.7.ffn_norm.weight
F32
F32
[768]
blk.7.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.7.ssm_a
F32
F32
[1, 48]
blk.7.ssm_conv1d.bias
F32
F32
[1792]
blk.7.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.7.ssm_d
F32
F32
[1, 48]
blk.7.ssm_dt.bias
F32
F32
[48]
blk.7.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.7.ssm_norm.weight
F32
F32
[1536]
blk.7.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.8
blk.8.attn_norm.weight
F32
F32
[768]
blk.8.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.8.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.8.ffn_norm.weight
F32
F32
[768]
blk.8.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.8.ssm_a
F32
F32
[1, 48]
blk.8.ssm_conv1d.bias
F32
F32
[1792]
blk.8.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.8.ssm_d
F32
F32
[1, 48]
blk.8.ssm_dt.bias
F32
F32
[48]
blk.8.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.8.ssm_norm.weight
F32
F32
[1536]
blk.8.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.9
blk.9.attn_norm.weight
F32
F32
[768]
blk.9.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.9.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.9.ffn_norm.weight
F32
F32
[768]
blk.9.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.9.ssm_a
F32
F32
[1, 48]
blk.9.ssm_conv1d.bias
F32
F32
[1792]
blk.9.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.9.ssm_d
F32
F32
[1, 48]
blk.9.ssm_dt.bias
F32
F32
[48]
blk.9.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.9.ssm_norm.weight
F32
F32
[1536]
blk.9.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.10
blk.10.attn_k.weight
Q8_0
Q8_0
[768, 256]
blk.10.attn_norm.weight
F32
F32
[768]
blk.10.attn_output.weight
Q8_0
Q8_0
[768, 768]
blk.10.attn_q.weight
Q8_0
Q8_0
[768, 768]
blk.10.attn_v.weight
Q8_0
Q8_0
[768, 256]
blk.10.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.10.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.10.ffn_norm.weight
F32
F32
[768]
blk.10.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.11
blk.11.attn_norm.weight
F32
F32
[768]
blk.11.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.11.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.11.ffn_norm.weight
F32
F32
[768]
blk.11.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.11.ssm_a
F32
F32
[1, 48]
blk.11.ssm_conv1d.bias
F32
F32
[1792]
blk.11.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.11.ssm_d
F32
F32
[1, 48]
blk.11.ssm_dt.bias
F32
F32
[48]
blk.11.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.11.ssm_norm.weight
F32
F32
[1536]
blk.11.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.12
blk.12.attn_norm.weight
F32
F32
[768]
blk.12.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.12.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.12.ffn_norm.weight
F32
F32
[768]
blk.12.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.12.ssm_a
F32
F32
[1, 48]
blk.12.ssm_conv1d.bias
F32
F32
[1792]
blk.12.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.12.ssm_d
F32
F32
[1, 48]
blk.12.ssm_dt.bias
F32
F32
[48]
blk.12.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.12.ssm_norm.weight
F32
F32
[1536]
blk.12.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.13
blk.13.attn_k.weight
Q8_0
Q8_0
[768, 256]
blk.13.attn_norm.weight
F32
F32
[768]
blk.13.attn_output.weight
Q8_0
Q8_0
[768, 768]
blk.13.attn_q.weight
Q8_0
Q8_0
[768, 768]
blk.13.attn_v.weight
Q8_0
Q8_0
[768, 256]
blk.13.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.13.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.13.ffn_norm.weight
F32
F32
[768]
blk.13.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.14
blk.14.attn_norm.weight
F32
F32
[768]
blk.14.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.14.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.14.ffn_norm.weight
F32
F32
[768]
blk.14.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.14.ssm_a
F32
F32
[1, 48]
blk.14.ssm_conv1d.bias
F32
F32
[1792]
blk.14.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.14.ssm_d
F32
F32
[1, 48]
blk.14.ssm_dt.bias
F32
F32
[48]
blk.14.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.14.ssm_norm.weight
F32
F32
[1536]
blk.14.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.15
blk.15.attn_norm.weight
F32
F32
[768]
blk.15.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.15.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.15.ffn_norm.weight
F32
F32
[768]
blk.15.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.15.ssm_a
F32
F32
[1, 48]
blk.15.ssm_conv1d.bias
F32
F32
[1792]
blk.15.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.15.ssm_d
F32
F32
[1, 48]
blk.15.ssm_dt.bias
F32
F32
[48]
blk.15.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.15.ssm_norm.weight
F32
F32
[1536]
blk.15.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.16
blk.16.attn_norm.weight
F32
F32
[768]
blk.16.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.16.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.16.ffn_norm.weight
F32
F32
[768]
blk.16.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.16.ssm_a
F32
F32
[1, 48]
blk.16.ssm_conv1d.bias
F32
F32
[1792]
blk.16.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.16.ssm_d
F32
F32
[1, 48]
blk.16.ssm_dt.bias
F32
F32
[48]
blk.16.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.16.ssm_norm.weight
F32
F32
[1536]
blk.16.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.17
blk.17.attn_k.weight
Q8_0
Q8_0
[768, 256]
blk.17.attn_norm.weight
F32
F32
[768]
blk.17.attn_output.weight
Q8_0
Q8_0
[768, 768]
blk.17.attn_q.weight
Q8_0
Q8_0
[768, 768]
blk.17.attn_v.weight
Q8_0
Q8_0
[768, 256]
blk.17.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.17.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.17.ffn_norm.weight
F32
F32
[768]
blk.17.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.18
blk.18.attn_norm.weight
F32
F32
[768]
blk.18.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.18.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.18.ffn_norm.weight
F32
F32
[768]
blk.18.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.18.ssm_a
F32
F32
[1, 48]
blk.18.ssm_conv1d.bias
F32
F32
[1792]
blk.18.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.18.ssm_d
F32
F32
[1, 48]
blk.18.ssm_dt.bias
F32
F32
[48]
blk.18.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.18.ssm_norm.weight
F32
F32
[1536]
blk.18.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.19
blk.19.attn_norm.weight
F32
F32
[768]
blk.19.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.19.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.19.ffn_norm.weight
F32
F32
[768]
blk.19.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.19.ssm_a
F32
F32
[1, 48]
blk.19.ssm_conv1d.bias
F32
F32
[1792]
blk.19.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.19.ssm_d
F32
F32
[1, 48]
blk.19.ssm_dt.bias
F32
F32
[48]
blk.19.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.19.ssm_norm.weight
F32
F32
[1536]
blk.19.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.20
blk.20.attn_norm.weight
F32
F32
[768]
blk.20.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.20.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.20.ffn_norm.weight
F32
F32
[768]
blk.20.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.20.ssm_a
F32
F32
[1, 48]
blk.20.ssm_conv1d.bias
F32
F32
[1792]
blk.20.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.20.ssm_d
F32
F32
[1, 48]
blk.20.ssm_dt.bias
F32
F32
[48]
blk.20.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.20.ssm_norm.weight
F32
F32
[1536]
blk.20.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.21
blk.21.attn_norm.weight
F32
F32
[768]
blk.21.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.21.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.21.ffn_norm.weight
F32
F32
[768]
blk.21.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.21.ssm_a
F32
F32
[1, 48]
blk.21.ssm_conv1d.bias
F32
F32
[1792]
blk.21.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.21.ssm_d
F32
F32
[1, 48]
blk.21.ssm_dt.bias
F32
F32
[48]
blk.21.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.21.ssm_norm.weight
F32
F32
[1536]
blk.21.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.22
blk.22.attn_norm.weight
F32
F32
[768]
blk.22.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.22.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.22.ffn_norm.weight
F32
F32
[768]
blk.22.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.22.ssm_a
F32
F32
[1, 48]
blk.22.ssm_conv1d.bias
F32
F32
[1792]
blk.22.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.22.ssm_d
F32
F32
[1, 48]
blk.22.ssm_dt.bias
F32
F32
[48]
blk.22.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.22.ssm_norm.weight
F32
F32
[1536]
blk.22.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.23
blk.23.attn_norm.weight
F32
F32
[768]
blk.23.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.23.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.23.ffn_norm.weight
F32
F32
[768]
blk.23.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.23.ssm_a
F32
F32
[1, 48]
blk.23.ssm_conv1d.bias
F32
F32
[1792]
blk.23.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.23.ssm_d
F32
F32
[1, 48]
blk.23.ssm_dt.bias
F32
F32
[48]
blk.23.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.23.ssm_norm.weight
F32
F32
[1536]
blk.23.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.24
blk.24.attn_norm.weight
F32
F32
[768]
blk.24.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.24.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.24.ffn_norm.weight
F32
F32
[768]
blk.24.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.24.ssm_a
F32
F32
[1, 48]
blk.24.ssm_conv1d.bias
F32
F32
[1792]
blk.24.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.24.ssm_d
F32
F32
[1, 48]
blk.24.ssm_dt.bias
F32
F32
[48]
blk.24.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.24.ssm_norm.weight
F32
F32
[1536]
blk.24.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.25
blk.25.attn_norm.weight
F32
F32
[768]
blk.25.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.25.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.25.ffn_norm.weight
F32
F32
[768]
blk.25.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.25.ssm_a
F32
F32
[1, 48]
blk.25.ssm_conv1d.bias
F32
F32
[1792]
blk.25.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.25.ssm_d
F32
F32
[1, 48]
blk.25.ssm_dt.bias
F32
F32
[48]
blk.25.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.25.ssm_norm.weight
F32
F32
[1536]
blk.25.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.26
blk.26.attn_norm.weight
F32
F32
[768]
blk.26.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.26.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.26.ffn_norm.weight
F32
F32
[768]
blk.26.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.26.ssm_a
F32
F32
[1, 48]
blk.26.ssm_conv1d.bias
F32
F32
[1792]
blk.26.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.26.ssm_d
F32
F32
[1, 48]
blk.26.ssm_dt.bias
F32
F32
[48]
blk.26.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.26.ssm_norm.weight
F32
F32
[1536]
blk.26.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.27
blk.27.attn_k.weight
Q8_0
Q8_0
[768, 256]
blk.27.attn_norm.weight
F32
F32
[768]
blk.27.attn_output.weight
Q8_0
Q8_0
[768, 768]
blk.27.attn_q.weight
Q8_0
Q8_0
[768, 768]
blk.27.attn_v.weight
Q8_0
Q8_0
[768, 256]
blk.27.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.27.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.27.ffn_norm.weight
F32
F32
[768]
blk.27.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.28
blk.28.attn_norm.weight
F32
F32
[768]
blk.28.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.28.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.28.ffn_norm.weight
F32
F32
[768]
blk.28.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.28.ssm_a
F32
F32
[1, 48]
blk.28.ssm_conv1d.bias
F32
F32
[1792]
blk.28.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.28.ssm_d
F32
F32
[1, 48]
blk.28.ssm_dt.bias
F32
F32
[48]
blk.28.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.28.ssm_norm.weight
F32
F32
[1536]
blk.28.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.29
blk.29.attn_norm.weight
F32
F32
[768]
blk.29.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.29.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.29.ffn_norm.weight
F32
F32
[768]
blk.29.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.29.ssm_a
F32
F32
[1, 48]
blk.29.ssm_conv1d.bias
F32
F32
[1792]
blk.29.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.29.ssm_d
F32
F32
[1, 48]
blk.29.ssm_dt.bias
F32
F32
[48]
blk.29.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.29.ssm_norm.weight
F32
F32
[1536]
blk.29.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.30
blk.30.attn_norm.weight
F32
F32
[768]
blk.30.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.30.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.30.ffn_norm.weight
F32
F32
[768]
blk.30.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.30.ssm_a
F32
F32
[1, 48]
blk.30.ssm_conv1d.bias
F32
F32
[1792]
blk.30.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.30.ssm_d
F32
F32
[1, 48]
blk.30.ssm_dt.bias
F32
F32
[48]
blk.30.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.30.ssm_norm.weight
F32
F32
[1536]
blk.30.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
blk.31
blk.31.attn_norm.weight
F32
F32
[768]
blk.31.ffn_down.weight
Q8_0
Q8_0
[2048, 768]
blk.31.ffn_gate.weight
Q8_0
Q8_0
[768, 2048]
blk.31.ffn_norm.weight
F32
F32
[768]
blk.31.ffn_up.weight
Q8_0
Q8_0
[768, 2048]
blk.31.ssm_a
F32
F32
[1, 48]
blk.31.ssm_conv1d.bias
F32
F32
[1792]
blk.31.ssm_conv1d.weight
F32
F32
[4, 1792]
blk.31.ssm_d
F32
F32
[1, 48]
blk.31.ssm_dt.bias
F32
F32
[48]
blk.31.ssm_in.weight
Q8_0
Q8_0
[768, 3376]
blk.31.ssm_norm.weight
F32
F32
[1536]
blk.31.ssm_out.weight
Q8_0
Q8_0
[1536, 768]
output_norm.weight
F32
F32
[768]