AeroCorp/afm:q8

AeroCorp/

afm:q8_0

49 Downloads Updated 2 months ago

The African Foundation Model (AFM) is a state-of-the-art language model specifically designed for African contexts, languages, and use cases. Built with the latest transformer optimizations from 2025 research. AFM combines power with efficiency.

tools

afm:q8_0 ... /

model

1bec82bdfad2 · 240MB

Metadata

general.architecture

llama

llama
general.file_type

Q8_0

Q8_0
llama.attention.head_count

8

8
llama.attention.head_count_kv

4

4
llama.attention.layer_norm_rms_epsilon

1e-05

1e-05
llama.block_count

12

12
llama.context_length

256

256
llama.embedding_length

512

512
llama.feed_forward_length

3072

3072
llama.rope.freq_base

10000

10000
llama.vocab_size

50000

50000
tokenizer.ggml.bos_token_id

1

1
tokenizer.ggml.eos_token_id

2

2
tokenizer.ggml.model

gpt2

gpt2
tokenizer.ggml.padding_token_id

0

0

Tensor

Name

Type

Shape
blocks.0.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.0.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.0.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.0.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.0.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.0.ffn.expert_bias

F16

F16

[4]
blocks.0.ffn.experts.0.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.0.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.0.ffn.experts.0.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.1.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.1.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.0.ffn.experts.1.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.2.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.2.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.0.ffn.experts.2.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.3.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.experts.3.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.0.ffn.experts.3.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.gate.weight

Q8_0

Q8_0

[512, 4]
blocks.0.ffn.shared_expert.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.ffn.shared_expert.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.0.ffn.shared_expert.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.0.norm1.weight

F16

F16

[512]
blocks.0.norm2.weight

F16

F16

[512]
blocks.1.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.1.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.1.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.1.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.1.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.1.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.1.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.1.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.1.norm1.weight

F16

F16

[512]
blocks.1.norm2.weight

F16

F16

[512]
blocks.2.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.2.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.2.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.2.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.2.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.2.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.2.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.2.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.2.norm1.weight

F16

F16

[512]
blocks.2.norm2.weight

F16

F16

[512]
blocks.3.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.3.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.3.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.3.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.3.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.3.ffn.expert_bias

F16

F16

[4]
blocks.3.ffn.experts.0.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.0.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.3.ffn.experts.0.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.1.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.1.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.3.ffn.experts.1.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.2.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.2.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.3.ffn.experts.2.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.3.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.experts.3.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.3.ffn.experts.3.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.gate.weight

Q8_0

Q8_0

[512, 4]
blocks.3.ffn.shared_expert.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.ffn.shared_expert.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.3.ffn.shared_expert.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.3.norm1.weight

F16

F16

[512]
blocks.3.norm2.weight

F16

F16

[512]
blocks.4.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.4.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.4.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.4.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.4.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.4.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.4.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.4.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.4.norm1.weight

F16

F16

[512]
blocks.4.norm2.weight

F16

F16

[512]
blocks.5.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.5.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.5.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.5.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.5.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.5.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.5.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.5.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.5.norm1.weight

F16

F16

[512]
blocks.5.norm2.weight

F16

F16

[512]
blocks.6.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.6.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.6.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.6.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.6.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.6.ffn.expert_bias

F16

F16

[4]
blocks.6.ffn.experts.0.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.0.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.6.ffn.experts.0.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.1.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.1.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.6.ffn.experts.1.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.2.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.2.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.6.ffn.experts.2.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.3.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.experts.3.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.6.ffn.experts.3.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.gate.weight

Q8_0

Q8_0

[512, 4]
blocks.6.ffn.shared_expert.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.ffn.shared_expert.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.6.ffn.shared_expert.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.6.norm1.weight

F16

F16

[512]
blocks.6.norm2.weight

F16

F16

[512]
blocks.7.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.7.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.7.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.7.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.7.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.7.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.7.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.7.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.7.norm1.weight

F16

F16

[512]
blocks.7.norm2.weight

F16

F16

[512]
blocks.8.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.8.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.8.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.8.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.8.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.8.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.8.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.8.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.8.norm1.weight

F16

F16

[512]
blocks.8.norm2.weight

F16

F16

[512]
blocks.9.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.9.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.9.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.9.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.9.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.9.ffn.expert_bias

F16

F16

[4]
blocks.9.ffn.experts.0.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.0.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.9.ffn.experts.0.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.1.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.1.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.9.ffn.experts.1.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.2.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.2.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.9.ffn.experts.2.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.3.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.experts.3.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.9.ffn.experts.3.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.gate.weight

Q8_0

Q8_0

[512, 4]
blocks.9.ffn.shared_expert.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.ffn.shared_expert.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.9.ffn.shared_expert.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.9.norm1.weight

F16

F16

[512]
blocks.9.norm2.weight

F16

F16

[512]
blocks.10.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.10.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.10.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.10.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.10.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.10.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.10.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.10.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.10.norm1.weight

F16

F16

[512]
blocks.10.norm2.weight

F16

F16

[512]
blocks.11.attention.k_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.11.attention.kv_down_proj.weight

Q8_0

Q8_0

[512, 128]
blocks.11.attention.o_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.11.attention.q_proj.weight

Q8_0

Q8_0

[512, 512]
blocks.11.attention.v_up_proj.weight

Q8_0

Q8_0

[128, 512]
blocks.11.ffn.w1.weight

Q8_0

Q8_0

[512, 2048]
blocks.11.ffn.w2.weight

Q8_0

Q8_0

[2048, 512]
blocks.11.ffn.w3.weight

Q8_0

Q8_0

[512, 2048]
blocks.11.norm1.weight

F16

F16

[512]
blocks.11.norm2.weight

F16

F16

[512]
final_norm.weight

F16

F16

[512]
mtp_head.position_transforms.0.0.bias

F16

F16

[512]
mtp_head.position_transforms.0.0.weight

Q8_0

Q8_0

[512, 512]
mtp_head.position_transforms.0.2.bias

F16

F16

[512]
mtp_head.position_transforms.0.2.weight

F16

F16

[512]
mtp_head.position_transforms.1.0.bias

F16

F16

[512]
mtp_head.position_transforms.1.0.weight

Q8_0

Q8_0

[512, 512]
mtp_head.position_transforms.1.2.bias

F16

F16

[512]
mtp_head.position_transforms.1.2.weight

F16

F16

[512]
mtp_head.position_transforms.2.0.bias

F16

F16

[512]
mtp_head.position_transforms.2.0.weight

Q8_0

Q8_0

[512, 512]
mtp_head.position_transforms.2.2.bias

F16

F16

[512]
mtp_head.position_transforms.2.2.weight

F16

F16

[512]
mtp_head.position_transforms.3.0.bias

F16

F16

[512]
mtp_head.position_transforms.3.0.weight

Q8_0

Q8_0

[512, 512]
mtp_head.position_transforms.3.2.bias

F16

F16

[512]
mtp_head.position_transforms.3.2.weight

F16

F16

[512]
mtp_head.prediction_heads.0.weight

Q8_0

Q8_0

[512, 50000]
mtp_head.prediction_heads.1.weight

Q8_0

Q8_0

[512, 50000]
mtp_head.prediction_heads.2.weight

Q8_0

Q8_0

[512, 50000]
mtp_head.prediction_heads.3.weight

Q8_0

Q8_0

[512, 50000]
token_embedding.weight

Q8_0

Q8_0

[512, 50000]