DeepSeek-V4.1-Flash-4bit-paged / mtp /mtp_quantization.json
yanghaojin's picture
Add files using upload-large-folder tool
706ca3c verified
Raw History Blame Contribute Delete
2.34 kB
{
"group_size": 64,
"bits": 8,
"expert_bits": 4,
"modules": {
"mtp.0.ffn.shared_experts.w3": {
"group_size": 64,
"bits": 8
},
"mtp.0.ffn.shared_experts.w2": {
"group_size": 64,
"bits": 8
},
"mtp.0.ffn.shared_experts.w1": {
"group_size": 64,
"bits": 8
},
"mtp.0.attn.wo_b": {
"group_size": 64,
"bits": 8
},
"mtp.0.main_proj": {
"group_size": 64,
"bits": 8
},
"mtp.0.attn.wq_b": {
"group_size": 64,
"bits": 8
},
"mtp.0.attn.wq_a": {
"group_size": 64,
"bits": 8
},
"mtp.0.attn.wkv": {
"group_size": 64,
"bits": 8
},
"mtp.0.ffn.experts.gate_proj": {
"group_size": 64,
"bits": 4
},
"mtp.0.ffn.experts.up_proj": {
"group_size": 64,
"bits": 4
},
"mtp.0.ffn.experts.down_proj": {
"group_size": 64,
"bits": 4
},
"mtp.1.ffn.shared_experts.w2": {
"group_size": 64,
"bits": 8
},
"mtp.1.ffn.shared_experts.w3": {
"group_size": 64,
"bits": 8
},
"mtp.1.attn.wkv": {
"group_size": 64,
"bits": 8
},
"mtp.1.ffn.shared_experts.w1": {
"group_size": 64,
"bits": 8
},
"mtp.1.attn.wo_b": {
"group_size": 64,
"bits": 8
},
"mtp.1.attn.wq_a": {
"group_size": 64,
"bits": 8
},
"mtp.1.attn.wq_b": {
"group_size": 64,
"bits": 8
},
"mtp.1.ffn.experts.gate_proj": {
"group_size": 64,
"bits": 4
},
"mtp.1.ffn.experts.up_proj": {
"group_size": 64,
"bits": 4
},
"mtp.1.ffn.experts.down_proj": {
"group_size": 64,
"bits": 4
},
"mtp.2.markov_head.embed": {
"group_size": 64,
"bits": 8
},
"mtp.2.ffn.shared_experts.w3": {
"group_size": 64,
"bits": 8
},
"mtp.2.ffn.shared_experts.w1": {
"group_size": 64,
"bits": 8
},
"mtp.2.attn.wkv": {
"group_size": 64,
"bits": 8
},
"mtp.2.attn.wq_b": {
"group_size": 64,
"bits": 8
},
"mtp.2.ffn.shared_experts.w2": {
"group_size": 64,
"bits": 8
},
"mtp.2.attn.wq_a": {
"group_size": 64,
"bits": 8
},
"mtp.2.markov_head.head": {
"group_size": 64,
"bits": 8
},
"mtp.2.attn.wo_b": {
"group_size": 64,
"bits": 8
},
"mtp.2.ffn.experts.gate_proj": {
"group_size": 64,
"bits": 4
},
"mtp.2.ffn.experts.up_proj": {
"group_size": 64,
"bits": 4
},
"mtp.2.ffn.experts.down_proj": {
"group_size": 64,
"bits": 4
}
}
}