ecker commited on
Commit
2e4dbd6
1 Parent(s): 57122c7

Upload 2 files

Browse files

Very early layerskip-aware weights

.gitattributes CHANGED
@@ -17,3 +17,4 @@ loras/ckpt/lora-cyberpunk-silverhand-r128-a128/lora.sft filter=lfs diff=lfs merg
17
  loras/ckpt/lora-cyberpunk-takemura-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
18
  loras/ckpt/lora-portal-glados-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
19
  loras/ckpt/lora-samandmax-sam-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
 
 
17
  loras/ckpt/lora-cyberpunk-takemura-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
18
  loras/ckpt/lora-portal-glados-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
19
  loras/ckpt/lora-samandmax-sam-r128-a128/lora.sft filter=lfs diff=lfs merge=lfs -text
20
+ models/ckpt/ar+nar-layerskip-llama-8/fp32.sft filter=lfs diff=lfs merge=lfs -text
models/ckpt/ar+nar-layerskip-llama-8/fp32.sft ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:232e0cec837102824e5d8b19cd4d3a138566aa7b9f048fa7c48049c5398e0677
3
+ size 456274634
models/config.llama[layerskip].yaml ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ models:
2
+ - name: "ar+nar-layerskip"
3
+ size: "full"
4
+ resp_levels: 8
5
+ tasks: 9
6
+ langs: 4
7
+ tones: 1
8
+ arch_type: llama
9
+ attention: auto
10
+ version: 5
11
+
12
+ capabilities: ["ar", "nar"]
13
+ experimental:
14
+ split_classifiers: True
15
+ audio_embedding_sums: True
16
+ unified_position_ids: False
17
+ layerskip: True