Aggressive deep layer compression of Mistral Large 2411

Detected 88 source layers.
Surgery Plan: 63 operations.
Merging Block 47-72 into 5 layers...
   > Compressing 26 layers -> 5 layers
Copying Embeddings, Norms, and Head...
Saving Index...
Updating Config...

DONE! New Model Depth: 67 Layers
Loading Embeddings...
Layer  | BI Score   | Weight Sim   | Verdict
--------------------------------------------------
0      | 0.5022     | N/A          |
1      | 0.9008     | 0.5000       | Active Layer
2      | 0.8713     | 0.7422       | Active Layer
3      | 0.9583     | 0.7109       | Active Layer
4      | 0.9647     | 0.8984       | Active Layer
5      | 0.9697     | 0.9062       | Active Layer
6      | 0.9531     | 0.9102       | Active Layer
7      | 0.9682     | 0.9219       | Active Layer
8      | 0.9694     | 0.9219       | Active Layer
9      | 0.9449     | 0.9219       | Active Layer
10     | 0.9621     | 0.9141       | Active Layer
11     | 0.9635     | 0.9141       | Active Layer
12     | 0.9480     | 0.9336       | Active Layer
13     | 0.9689     | 0.9375       | Active Layer
14     | 0.9666     | 0.9375       | Active Layer
15     | 0.9658     | 0.9336       | Active Layer
16     | 0.9609     | 0.9336       | Active Layer
17     | 0.9651     | 0.9414       | Active Layer
18     | 0.9645     | 0.9414       | Active Layer
19     | 0.9633     | 0.9375       | Active Layer
20     | 0.9644     | 0.9414       | Active Layer
21     | 0.9571     | 0.9453       | Active Layer
22     | 0.9557     | 0.9492       | Active Layer
23     | 0.9326     | 0.9492       | Active Layer
24     | 0.9488     | 0.9492       | Active Layer
25     | 0.9479     | 0.9531       | Active Layer
26     | 0.9613     | 0.9492       | Active Layer
27     | 0.9586     | 0.9492       | Active Layer
28     | 0.9398     | 0.9492       | Active Layer
29     | 0.9459     | 0.9453       | Active Layer
30     | 0.9445     | 0.9492       | Active Layer
31     | 0.9378     | 0.9531       | Active Layer
32     | 0.9465     | 0.9531       | Active Layer
33     | 0.9487     | 0.9531       | Active Layer
34     | 0.9458     | 0.9570       | Active Layer
35     | 0.9502     | 0.9609       | Active Layer
36     | 0.9221     | 0.9609       | Active Layer
37     | 0.9402     | 0.9609       | Active Layer
38     | 0.9406     | 0.9609       | Active Layer
39     | 0.9464     | 0.9648       | Active Layer
40     | 0.9593     | 0.9648       | Active Layer
41     | 0.9596     | 0.9648       | Active Layer
42     | 0.9559     | 0.9648       | Active Layer
43     | 0.9710     | 0.9688       | Active Layer
44     | 0.9792     | 0.9688       | Active Layer
45     | 0.9820     | 0.9727       | Possible Merge
46     | 0.9580     | 0.9727       | Active Layer
47     | 0.9874     | 0.9727       | Possible Merge
48     | 0.9876     | 0.9688       | Possible Merge
49     | 0.9894     | 0.9688       | Possible Merge
50     | 0.9877     | 0.9648       | Possible Merge
51     | 0.9883     | 0.9688       | Possible Merge
52     | 0.9897     | 0.9727       | Possible Merge
53     | 0.9886     | 0.9727       | Possible Merge
54     | 0.9882     | 0.9688       | Possible Merge
55     | 0.9864     | 0.9688       | Possible Merge
56     | 0.9885     | 0.9727       | Possible Merge
57     | 0.9868     | 0.9688       | Possible Merge
58     | 0.9899     | 0.9688       | Possible Merge
59     | 0.9871     | 0.9688       | Possible Merge
60     | 0.9903     | 0.9727       | MERGE CANDIDATE
61     | 0.9888     | 0.9766       | Possible Merge
62     | 0.9878     | 0.9727       | Possible Merge
63     | 0.9899     | 0.9727       | Possible Merge
64     | 0.9885     | 0.9727       | Possible Merge
65     | 0.9854     | 0.9727       | Possible Merge
66     | 0.9886     | 0.9727       | Possible Merge
67     | 0.9837     | 0.9688       | Possible Merge
68     | 0.9894     | 0.9727       | Possible Merge
69     | 0.9879     | 0.9766       | Possible Merge
70     | 0.9872     | 0.9688       | Possible Merge
71     | 0.9809     | 0.9727       | Possible Merge
72     | 0.9827     | 0.9766       | Possible Merge
73     | 0.9789     | 0.9766       | Active Layer
74     | 0.9790     | 0.9766       | Active Layer
75     | 0.9692     | 0.9844       | Active Layer
76     | 0.9611     | 0.9766       | Active Layer
77     | 0.9762     | 0.9727       | Active Layer
78     | 0.9757     | 0.9727       | Active Layer
79     | 0.9729     | 0.9766       | Active Layer
80     | 0.9744     | 0.9766       | Active Layer
81     | 0.9677     | 0.9727       | Active Layer
82     | 0.9665     | 0.9766       | Active Layer
83     | 0.9643     | 0.9766       | Active Layer
84     | 0.9571     | 0.9727       | Active Layer
85     | 0.9287     | 0.9727       | Active Layer
86     | 0.8072     | 0.9766       | Active Layer
87     | 0.6603     | 0.9609       | Active Layer
Downloads last month
27
Safetensors
Model size
94B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for schonsense/fogstral_compressed

Finetuned
(15)
this model