hugo/DS4Server
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

Add native Qwen MTP speculation

Browse Source
This commit is contained in:
Georg Bauer
2026-09-03 22:38:52 +02:00
parent 714b39f7f3
commit bf82df77cb
14 changed files with 1302 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
assets/models/qwen38-flash-next-bare-speed.json
View File

@@ -11,7 +11,7 @@
"converter": "qwen38-artifacts-v1-identity"
},
"tensor_inventory": "qwen38-flash-next-bare-speed-tensors.tsv",
"tensor_inventory_sha256": "369cdc7e53c09eaa5eff5cdcc49f0d0eab0ddc175f34965ba2ed1be2673d5858",
"tensor_inventory_sha256": "b5731e6febcf865d276a0e7b144da02375f2d7e3129594f271ee9c6c351f4c8f",
"config": {
"/architectures/0": "Qwen4ExpForConditionalGeneration",
"/model_type": "qwen4_exp",
@@ -73,7 +73,7 @@
{"class":"qsa","file":"model-00016-of-00017.safetensors","tensor":"language_model.model.layers.11.self_attn.indexer.index_qk_proj.weight","row":0,"values":64,"bits":8,"group_size":64,"sha256":"c803ea77148621a5d6dfa4060a52c75f57185234a0bec4771671f69336bb8346"},
{"class":"gdn","file":"model-00016-of-00017.safetensors","tensor":"language_model.model.layers.0.linear_attn.A_log","row":0,"values":48,"sha256":"88c53a2a04bda1d96ee1ade6fa7dfa9c49d3245b1cc1e83673f02b7a36e07a85"},
{"class":"ple","file":"ngram-table.safetensors","tensor":"ngram.weight","row":0,"values":64,"bits":4,"group_size":32,"sha256":"2243dd9766046bb80d98e3baf5e59ba246958cfe4f340ebce8c1fff57a2810d9"},
{"class":"mtp","file":"mtp.safetensors","tensor":"mtp.layers.0.mlp.switch_mlp.gate_proj.weight","row":0,"values":64,"bits":2,"group_size":64,"sha256":"6293cce5e1b25a4f35cb8af46e6e13b8378ec4ca84ec7f68859066dda48049cc"}
{"class":"mtp","file":"mtp.safetensors","tensor":"mtp.layers.0.mlp.switch_mlp.gate_proj.weight","row":0,"values":64,"bits":4,"group_size":32,"sha256":"393d82675ae7e25275243fd4907d3894146ac1c5662b601aae998c489d10af5c"}
],
"files": [
{"path":"model-00001-of-00017.safetensors","role":"core","size":4666167150,"sha256":"a27232c9434b9d8961f198cf36f44e228b3f23a5f8d874ba4ca8af32a1b23ffc"},