dataautogpt3 commited on
Commit
0e6bdfd
1 Parent(s): a936633

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +18 -14
README.md CHANGED
@@ -65,34 +65,38 @@ PrometheusV1 is presumed to be the first full rank finetune of Playground v2.5,
65
 
66
  Presumed First Full Rank Finetune of Playground v2.5:
67
 
68
- Complete parameter update of Playground v2.5 architecture
69
- Unique approach to fine-tuning this particular base model
70
 
71
 
72
  Enhanced Accessibility:
73
 
74
- Custom sampling methods have been removed through brute force techniques
75
- Designed to be more compatible with standard open-source tools and workflows
76
- post processing applied to ensure backwards compatibility with most SDXL LoRAs and tools
77
 
78
 
79
  Output Characteristics:
80
 
81
- Aims to provide a balance between consistency and diversity in outputs
82
- May exhibit some stylistic tendencies inherited from the training process
83
 
84
 
85
  Training Approach:
86
 
87
- Utilizes the extensive Proteus datasets of over 400,000 images
88
- Brute force at scale training methodology
89
- Focused on maintaining model capabilities while increasing compatibility
90
 
91
 
92
- Custom CLIP Integration:
93
-
94
- Incorporation of a specially trained CLIP model
95
- Requires a clip skip setting of 2 for optimal performance
 
 
 
 
96
 
97
 
98
 
 
65
 
66
  Presumed First Full Rank Finetune of Playground v2.5:
67
 
68
+ - Complete parameter update of Playground v2.5 architecture
69
+ - Unique approach to fine-tuning this particular base model
70
 
71
 
72
  Enhanced Accessibility:
73
 
74
+ - Custom sampling methods have been removed through brute force techniques
75
+ - Designed to be more compatible with standard open-source tools and workflows
76
+ - post processing applied to ensure backwards compatibility with most SDXL LoRAs and tools
77
 
78
 
79
  Output Characteristics:
80
 
81
+ - Aims to provide a balance between consistency and diversity in outputs
82
+ - May exhibit some stylistic tendencies inherited from the training process
83
 
84
 
85
  Training Approach:
86
 
87
+ - Utilizes the extensive Proteus datasets of over 400,000 images
88
+ - Brute force at scale training methodology
89
+ - Focused on maintaining model capabilities while increasing compatibility
90
 
91
 
92
+ Advanced Custom CLIP Integration:
93
+
94
+ - Incorporates a meticulously trained custom CLIP model
95
+ - Steadily developed over an extended period
96
+ - Further fine-tuned for specific qualities in Proteus and Prometheus
97
+ - Estimated to contribute 90% of the model's performance improvements
98
+ - Requires a clip skip setting of 2 for optimal performance
99
+
100
 
101
 
102