Update README.md
Browse files
README.md
CHANGED
@@ -25,104 +25,19 @@ Back from the dead! Hoping to make something cool to share with everyone! Introd
|
|
25 |
[exl2](https://huggingface.co/lucyknada/Epiculous_Crimson_Dawn-V0.1-exl2) / [gguf](https://huggingface.co/mradermacher/Crimson_Dawn-V0.1-GGUF)
|
26 |
|
27 |
## Prompting
|
28 |
-
Crimson Dawn was trained with the Mistral Instruct template, therefore it should be prompted in
|
29 |
|
30 |
```
|
31 |
"<s>[INST] Prompt goes here [/INST]<\s>"
|
32 |
```
|
|
|
|
|
|
|
|
|
33 |
|
34 |
### Current Top Sampler Settings
|
35 |
-
|
36 |
-
|
37 |
-
"temp": 1.25,
|
38 |
-
"temperature_last": true,
|
39 |
-
"top_p": 1,
|
40 |
-
"top_k": -1,
|
41 |
-
"top_a": 0,
|
42 |
-
"tfs": 1,
|
43 |
-
"epsilon_cutoff": 0,
|
44 |
-
"eta_cutoff": 0,
|
45 |
-
"typical_p": 1,
|
46 |
-
"min_p": 0.3,
|
47 |
-
"rep_pen": 1,
|
48 |
-
"rep_pen_range": 0,
|
49 |
-
"rep_pen_decay": 0,
|
50 |
-
"rep_pen_slope": 1,
|
51 |
-
"no_repeat_ngram_size": 0,
|
52 |
-
"penalty_alpha": 0,
|
53 |
-
"num_beams": 1,
|
54 |
-
"length_penalty": 1,
|
55 |
-
"min_length": 0,
|
56 |
-
"encoder_rep_pen": 1,
|
57 |
-
"freq_pen": 0,
|
58 |
-
"presence_pen": 0,
|
59 |
-
"skew": 0,
|
60 |
-
"do_sample": true,
|
61 |
-
"early_stopping": false,
|
62 |
-
"dynatemp": false,
|
63 |
-
"min_temp": 0,
|
64 |
-
"max_temp": 2,
|
65 |
-
"dynatemp_exponent": 1,
|
66 |
-
"smoothing_factor": 0,
|
67 |
-
"smoothing_curve": 1,
|
68 |
-
"dry_allowed_length": 2,
|
69 |
-
"dry_multiplier": 0,
|
70 |
-
"dry_base": 1.75,
|
71 |
-
"dry_sequence_breakers": "[\"\\n\", \":\", \"\\\"\", \"*\"]",
|
72 |
-
"dry_penalty_last_n": 0,
|
73 |
-
"add_bos_token": true,
|
74 |
-
"ban_eos_token": false,
|
75 |
-
"skip_special_tokens": true,
|
76 |
-
"mirostat_mode": 0,
|
77 |
-
"mirostat_tau": 5,
|
78 |
-
"mirostat_eta": 0.1,
|
79 |
-
"guidance_scale": 1,
|
80 |
-
"negative_prompt": "",
|
81 |
-
"grammar_string": "",
|
82 |
-
"json_schema": {},
|
83 |
-
"banned_tokens": "",
|
84 |
-
"sampler_priority": [
|
85 |
-
"temperature",
|
86 |
-
"dynamic_temperature",
|
87 |
-
"quadratic_sampling",
|
88 |
-
"top_k",
|
89 |
-
"top_p",
|
90 |
-
"typical_p",
|
91 |
-
"epsilon_cutoff",
|
92 |
-
"eta_cutoff",
|
93 |
-
"tfs",
|
94 |
-
"top_a",
|
95 |
-
"min_p",
|
96 |
-
"mirostat"
|
97 |
-
],
|
98 |
-
"samplers": [
|
99 |
-
"top_k",
|
100 |
-
"tfs_z",
|
101 |
-
"typical_p",
|
102 |
-
"top_p",
|
103 |
-
"min_p",
|
104 |
-
"temperature"
|
105 |
-
],
|
106 |
-
"ignore_eos_token": false,
|
107 |
-
"spaces_between_special_tokens": true,
|
108 |
-
"speculative_ngram": false,
|
109 |
-
"sampler_order": [
|
110 |
-
5,
|
111 |
-
6,
|
112 |
-
0,
|
113 |
-
1,
|
114 |
-
2,
|
115 |
-
3,
|
116 |
-
4
|
117 |
-
],
|
118 |
-
"logit_bias": [],
|
119 |
-
"ignore_eos_token_aphrodite": false,
|
120 |
-
"spaces_between_special_tokens_aphrodite": true,
|
121 |
-
"rep_pen_size": 0,
|
122 |
-
"genamt": 1024,
|
123 |
-
"max_length": 16384
|
124 |
-
}
|
125 |
-
```
|
126 |
|
127 |
## Training
|
128 |
Training was done twice over 2 epochs each on two 2x [NVIDIA A6000 GPUs](https://www.nvidia.com/en-us/design-visualization/rtx-a6000/) using LoRA. A two-phased approach was used in which the base model was trained 2 epochs on RP data, the LoRA was then applied to base. Finally, the new modified base was trained 2 epochs on instruct, and the new instruct LoRA was applied to the modified base, resulting in what you see here.
|
|
|
25 |
[exl2](https://huggingface.co/lucyknada/Epiculous_Crimson_Dawn-V0.1-exl2) / [gguf](https://huggingface.co/mradermacher/Crimson_Dawn-V0.1-GGUF)
|
26 |
|
27 |
## Prompting
|
28 |
+
Crimson Dawn was trained with the Mistral Instruct template, therefore it should be prompted in a similar way that you would prompt any other mistral based model.
|
29 |
|
30 |
```
|
31 |
"<s>[INST] Prompt goes here [/INST]<\s>"
|
32 |
```
|
33 |
+
### Context and Instruct
|
34 |
+
[Mistral-Custom-Context.json](https://files.catbox.moe/l9w0ry.json)
|
35 |
+
[Mistral-Custom-Instruct.json](https://files.catbox.moe/9xiiwb.json)
|
36 |
+
|
37 |
|
38 |
### Current Top Sampler Settings
|
39 |
+
[Crimson_Dawn-Magnum-Style](https://files.catbox.moe/lc59dn.json)
|
40 |
+
[Crimson_Dawn-Nitral-Special](https://files.catbox.moe/8xjxht.json)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
## Training
|
43 |
Training was done twice over 2 epochs each on two 2x [NVIDIA A6000 GPUs](https://www.nvidia.com/en-us/design-visualization/rtx-a6000/) using LoRA. A two-phased approach was used in which the base model was trained 2 epochs on RP data, the LoRA was then applied to base. Finally, the new modified base was trained 2 epochs on instruct, and the new instruct LoRA was applied to the modified base, resulting in what you see here.
|