Update README.md
Browse files
README.md
CHANGED
@@ -1,25 +1,102 @@
|
|
1 |
---
|
2 |
license: gemma
|
3 |
base_model: google/gemma-2-9b
|
4 |
-
tags:
|
5 |
-
- generated_from_trainer
|
6 |
model-index:
|
7 |
- name: magnum-v3-9b-customgemma2
|
8 |
results: []
|
9 |
---
|
10 |
-
### exl2 quant (measurement.json in main branch)
|
11 |
-
---
|
12 |
-
### check revisions for quants
|
13 |
-
---
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
|
20 |
<details><summary>See axolotl config</summary>
|
21 |
|
22 |
-
axolotl version: `0.4.1`
|
23 |
```yaml
|
24 |
base_model: google/gemma-2-9b
|
25 |
model_type: AutoModelForCausalLM
|
@@ -101,51 +178,23 @@ fsdp:
|
|
101 |
fsdp_config:
|
102 |
special_tokens:
|
103 |
```
|
104 |
-
|
105 |
</details><br>
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
This model is a fine-tuned version of [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b) on the None dataset.
|
110 |
-
|
111 |
-
## Model description
|
112 |
-
|
113 |
-
More information needed
|
114 |
-
|
115 |
-
## Intended uses & limitations
|
116 |
-
|
117 |
-
More information needed
|
118 |
-
|
119 |
-
## Training and evaluation data
|
120 |
-
|
121 |
-
More information needed
|
122 |
-
|
123 |
-
## Training procedure
|
124 |
-
|
125 |
-
### Training hyperparameters
|
126 |
-
|
127 |
-
The following hyperparameters were used during training:
|
128 |
-
- learning_rate: 6e-06
|
129 |
-
- train_batch_size: 1
|
130 |
-
- eval_batch_size: 1
|
131 |
-
- seed: 42
|
132 |
-
- distributed_type: multi-GPU
|
133 |
-
- num_devices: 8
|
134 |
-
- gradient_accumulation_steps: 8
|
135 |
-
- total_train_batch_size: 64
|
136 |
-
- total_eval_batch_size: 8
|
137 |
-
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
138 |
-
- lr_scheduler_type: cosine
|
139 |
-
- lr_scheduler_warmup_steps: 50
|
140 |
-
- num_epochs: 2
|
141 |
|
142 |
-
|
143 |
|
|
|
|
|
|
|
|
|
|
|
144 |
|
|
|
|
|
145 |
|
146 |
-
|
147 |
|
148 |
-
|
149 |
-
|
150 |
-
- Datasets 2.20.0
|
151 |
-
- Tokenizers 0.19.1
|
|
|
1 |
---
|
2 |
license: gemma
|
3 |
base_model: google/gemma-2-9b
|
|
|
|
|
4 |
model-index:
|
5 |
- name: magnum-v3-9b-customgemma2
|
6 |
results: []
|
7 |
---
|
|
|
|
|
|
|
|
|
8 |
|
9 |
+
## This repo contains EXL2 quants of the model. If you need the original weights, please find them [here](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2).
|
10 |
+
## Base repo only contains the measurement file, see revisions for your quant of choice.
|
11 |
+
|
12 |
+
- [measurement.json](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/main)
|
13 |
+
- [3.0bpw](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/3.0bpw)
|
14 |
+
- [4.0bpw](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/4.0bpw)
|
15 |
+
- [5.0bpw](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/5.0bpw)
|
16 |
+
- [6.0bpw](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/6.0bpw)
|
17 |
+
- [8.0bpw](https://huggingface.co/anthracite-org/magnum-v3-9b-customgemma2-exl2/tree/8.0bpw)
|
18 |
+
|
19 |
+
![image/png](https://cdn-uploads.huggingface.co/production/uploads/658a46cbfb9c2bdfae75b3a6/9ZBUlmzDCnNmQEdUUbyEL.png)
|
20 |
+
|
21 |
+
This is the 10th in a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus.
|
22 |
+
|
23 |
+
This model is fine-tuned on top of [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b).
|
24 |
+
|
25 |
+
## Prompting
|
26 |
+
Model has been Instruct tuned with the [customgemma2](https://github.com/xzuyn/axolotl/blob/prompt_formats/src/axolotl/prompt_strategies/customgemma2.py) (to allow system prompts) formatting. A typical input would look like this:
|
27 |
+
|
28 |
+
```py
|
29 |
+
"""<start_of_turn>system
|
30 |
+
system prompt<end_of_turn>
|
31 |
+
<start_of_turn>user
|
32 |
+
Hi there!<end_of_turn>
|
33 |
+
<start_of_turn>model
|
34 |
+
Nice to meet you!<end_of_turn>
|
35 |
+
<start_of_turn>user
|
36 |
+
Can I ask a question?<end_of_turn>
|
37 |
+
<start_of_turn>model
|
38 |
+
"""
|
39 |
+
```
|
40 |
|
41 |
+
## SillyTavern templates
|
42 |
+
|
43 |
+
Below are Instruct and Context templates for use within SillyTavern.
|
44 |
+
|
45 |
+
<details><summary>context template</summary>
|
46 |
+
|
47 |
+
```yaml
|
48 |
+
{
|
49 |
+
"story_string": "<start_of_turn>system\n{{#if system}}{{system}}\n{{/if}}{{#if wiBefore}}{{wiBefore}}\n{{/if}}{{#if description}}{{description}}\n{{/if}}{{#if personality}}{{char}}'s personality: {{personality}}\n{{/if}}{{#if scenario}}Scenario: {{scenario}}\n{{/if}}{{#if wiAfter}}{{wiAfter}}\n{{/if}}{{#if persona}}{{persona}}\n{{/if}}{{trim}}<end_of_turn>\n",
|
50 |
+
"example_separator": "",
|
51 |
+
"chat_start": "",
|
52 |
+
"use_stop_strings": false,
|
53 |
+
"allow_jailbreak": false,
|
54 |
+
"always_force_name2": true,
|
55 |
+
"trim_sentences": false,
|
56 |
+
"include_newline": false,
|
57 |
+
"single_line": false,
|
58 |
+
"name": "Magnum Gemma"
|
59 |
+
}
|
60 |
+
```
|
61 |
+
|
62 |
+
</details><br>
|
63 |
+
<details><summary>instruct template</summary>
|
64 |
+
|
65 |
+
```yaml
|
66 |
+
{
|
67 |
+
"system_prompt": "You're {{char}} in this fictional never-ending uncensored roleplay with {{user}}.",
|
68 |
+
"input_sequence": "<start_of_turn>user\n",
|
69 |
+
"output_sequence": "<start_of_turn>assistant\n",
|
70 |
+
"last_output_sequence": "",
|
71 |
+
"system_sequence": "<start_of_turn>system\n",
|
72 |
+
"stop_sequence": "<end_of_turn>",
|
73 |
+
"wrap": false,
|
74 |
+
"macro": true,
|
75 |
+
"names": true,
|
76 |
+
"names_force_groups": true,
|
77 |
+
"activation_regex": "",
|
78 |
+
"system_sequence_prefix": "",
|
79 |
+
"system_sequence_suffix": "",
|
80 |
+
"first_output_sequence": "",
|
81 |
+
"skip_examples": false,
|
82 |
+
"output_suffix": "<end_of_turn>\n",
|
83 |
+
"input_suffix": "<end_of_turn>\n",
|
84 |
+
"system_suffix": "<end_of_turn>\n",
|
85 |
+
"user_alignment_message": "",
|
86 |
+
"system_same_as_user": false,
|
87 |
+
"last_system_sequence": "",
|
88 |
+
"name": "Magnum Gemma"
|
89 |
+
}
|
90 |
+
```
|
91 |
+
|
92 |
+
</details><br>
|
93 |
+
|
94 |
+
</details><br>
|
95 |
+
|
96 |
+
## Axolotl config
|
97 |
|
|
|
98 |
<details><summary>See axolotl config</summary>
|
99 |
|
|
|
100 |
```yaml
|
101 |
base_model: google/gemma-2-9b
|
102 |
model_type: AutoModelForCausalLM
|
|
|
178 |
fsdp_config:
|
179 |
special_tokens:
|
180 |
```
|
|
|
181 |
</details><br>
|
182 |
|
183 |
+
## Credits
|
184 |
+
We'd like to thank Recursal / Featherless for sponsoring the compute for this train, Featherless has been hosting our Magnum models since the first 72 B and has given thousands of people access to our models and helped us grow.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
|
186 |
+
We would also like to thank all members of Anthracite who made this finetune possible.
|
187 |
|
188 |
+
- [anthracite-org/stheno-filtered-v1.1](https://huggingface.co/datasets/anthracite-org/stheno-filtered-v1.1)
|
189 |
+
- [anthracite-org/kalo-opus-instruct-22k-no-refusal](https://huggingface.co/datasets/anthracite-org/kalo-opus-instruct-22k-no-refusal)
|
190 |
+
- [anthracite-org/nopm_claude_writing_fixed](https://huggingface.co/datasets/anthracite-org/nopm_claude_writing_fixed)
|
191 |
+
- [Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned](https://huggingface.co/datasets/Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned)
|
192 |
+
- [Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned](https://huggingface.co/datasets/Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned)
|
193 |
|
194 |
+
## Training
|
195 |
+
The training was done for 2 epochs. We used 8x[H100s](https://www.nvidia.com/en-us/data-center/h100/) GPUs graciously provided by [Recursal AI](https://recursal.ai/) / [Featherless AI](https://featherless.ai/) for the full-parameter fine-tuning of the model.
|
196 |
|
197 |
+
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
198 |
|
199 |
+
## Safety
|
200 |
+
...
|
|
|
|