jphme commited on
Commit
d8970de
·
1 Parent(s): 830154e

update readme

Browse files
Files changed (1) hide show
  1. README.md +41 -30
README.md CHANGED
@@ -1,24 +1,23 @@
1
  ---
2
- license: apache-2.0
3
  language:
4
  - de
5
  library_name: transformers
6
- pipeline_tag: text-generation
7
- inference: false
8
  model_creator: jphme
9
  model_name: EM German
10
  model_type: mistral
11
- prompt_template: >
12
- Du bist ein hilfreicher KI Assistent, der den Anweisungen des Nutzers sehr gut folgt und ausführliche Antworten gibt! USER: Was ist 1+1? ASSISTANT:
13
  tags:
14
- - pytorch
15
- - mistral
16
- - german
17
- - deutsch
18
  ---
19
  ![EM Logo](em_model_logo_web.jpeg)
20
 
21
- **Please note that the Mistral architecture is very recent and still not supported by all libraries (e.g. AutoGPTQ). In case of any problems, please try a different format/base model.**
22
 
23
  # Table of Contents
24
 
@@ -34,18 +33,19 @@ tags:
34
 
35
  # Introduction
36
 
37
- **EM German (v01)** is a Llama2/Mistral/LeoLM-based model family, finetuned on a large dataset of various instructions in German language. The models are optimized for German text, providing proficiency in understanding, generating, and interacting with German language content.
38
-
39
- Please find all Informations, Example Outputs, the special RAG prompt format, output examples and eval results for the EM German Model family in [our Github Repository](https://github.com/jphme/EM_German).
40
 
 
41
 
42
- (Für weitere Informationen und Anleitungen auf Deutsch, besuchen Sie bitte [unser Github Repository](https://github.com/jphme/EM_German/blob/main/README_DE.md).)
43
 
44
 
45
  # Links & Demos
46
 
47
  ## Model Links
48
 
 
 
49
  | Base Model | HF | GPTQ | GGUF | AWQ |
50
  |-------|-------|-------|-------|-------|
51
  | Llama2 7b | [Link](https://huggingface.co/jphme/em_german_7b_v01) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-AWQ) |
@@ -53,10 +53,11 @@ Please find all Informations, Example Outputs, the special RAG prompt format, ou
53
  | Llama2 70b | [Link](https://huggingface.co/jphme/em_german_70b_v01) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-AWQ) |
54
  | [Mistral 7b](https://huggingface.co/mistralai/Mistral-7B-v0.1) | [Link](https://huggingface.co/jphme/em_german_mistral_v01) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-AWQ) |
55
  | [LeoLM 7b](https://huggingface.co/LeoLM/leo-hessianai-7b) | [Link](https://huggingface.co/jphme/em_german_7b_leo) | [Link](https://huggingface.co/jphme/em_german_7b_leo_gptq) | [Link](hhttps://huggingface.co/jphme/em_german_7b_leo_gguf) | tbc |
56
- | LeoLM 13b | soon | soon | soon | tbc |
 
57
 
58
  ### Notes about the different versions:
59
- For the 7b models, we recommend the use of the "LeoLM" variant if text output quality is important and the Mistral variant, if reasoning/understanding is the main priority. Both should give better results than the Llama-2 7b model and often even the Llama2 13b model.
60
 
61
  If you get unsatisfying results with one or another EM German version, please try a different (and/or larger) model or version for your usecase.
62
 
@@ -67,7 +68,7 @@ You can use some of the models with **free** google Colab instances (e.g. the 7b
67
 
68
  * [Example Colab Notebook for 13b with GPTQ](https://colab.research.google.com/drive/1IJfJdVwGkfe5MYOqHptystR3FBeEUdGn?usp=sharing)
69
  * [Example Colab Notebook for 7b with 8bit-Loading](https://colab.research.google.com/drive/1bsv6vkLM4AlCpSyXA6ol9P32zxZmf7Zu?usp=sharing)
70
- * For further information and GUI use, please visit [our Github Repository](https://github.com/jphme/EM_German).
71
 
72
 
73
  # Prompt Format
@@ -82,30 +83,40 @@ You can swap the standard system prompt for a better suited one (see below for R
82
 
83
  # Example Output
84
 
85
-
86
 
87
  **Prompt:**
88
- > Gestalten Sie einen LinkedIn Post, in dem Sie die Grundlagen des deutschen Urheberrechts erklären. Versuchen Sie, es auf eine einfache und verständliche Weise darzustellen, so dass es für alle Nutzer verständlich ist.
89
 
90
- **Antwort EM German 7b Leo:**
91
  ```
92
- Titel: Grundlagen des deutschen Urheberrechts: Wer hat das Recht, was zu schaffen? 📖
 
 
93
 
94
- Das deutsche Urheberrecht regelt die Rechte von Urhebern an ihren Werken. Es schützt die persönlichen geistigen Schöpfungen von Künstlern, Schriftstellern und anderen Kreativen. Hier sind einige grundlegende Prinzipien des deutschen Urheberrechts, die Sie kennen sollten:
95
 
96
- 1️⃣ Wer ist ein Urheber? Das deutsche Urheberrecht schützt die Urheber von Werken der Literatur, Wissenschaft und Kunst. Das sind zum Beispiel Texte, Bilder, Musikstücke und Filme.
 
97
 
98
- 2️⃣ Was bedeutet "Werk"? Ein Werk ist eine persönliche geistige Schöpfung, die sich durch Originalität auszeichnet und die geistige Arbeit des Urhebers zeigt. Es muss eine gewisse "Schöpfungshöhe" haben.
 
99
 
100
- 3️⃣ Wie lange dauert der Urheberrechtsschutz? Der Urheberrechtsschutz beginnt mit der Schaffung des Werks und endet 70 Jahre nach dem Tod des Urhebers.
 
 
101
 
102
- 4️⃣ Was darf ich mit dem Werk machen? Wer ein Werk nutzen möchte, zum Beispiel veröffentlichen oder vervielfältigen, benötigt die Erlaubnis des Urhebers. Das bedeutet, dass Sie nur dann ein Werk verwenden dürfen, wenn Sie die Genehmigung des Urhebers haben.
 
 
103
 
104
- 5️⃣ Was ist eine Nutzungseinschränkung? Eine Nutzungseinschränkung ist eine Bedingung, die der Urheber dem Nutzer auferlegt, wenn er das Werk nutzen möchte. Zum Beispiel kann der Urheber bestimmen, dass sein Werk nur für private Zwecke genutzt werden darf.
105
 
106
- 6️⃣ Was ist eine Verwertungsgesellschaft? Eine Verwertungsgesellschaft ist eine Organisation, die die Rechte der Urheber verwaltet und durchsetzt. In Deutschland gibt es drei Verwertungsgesellschaften: GEMA (für Musik), VG Wort (für Sprachwerke) und VG Bild-Kunst (für bildende Kunst).
 
 
107
 
108
- 7️⃣ Wie kann ich meine Urheberrechte durchsetzen? Wenn Sie glauben, dass Ihre Urheberrechte verletzt wurden, sollten Sie einen Rechtsanwalt konsultieren. Sie können auch eine Abmahnung aussprechen oder gerichtliche Schritte einleiten.
109
  ```
110
  (For more examples, please visit [our Github Repository](https://github.com/jphme/EM_German).)
111
 
@@ -119,7 +130,7 @@ The 70b model was trained with support of the [OVH Cloud Startup Program](https:
119
 
120
  I you are interested in customized LLMs for business applications, please get in contact with me via [my website](https://www.jph.me). I am also always happy about suggestions and feedback.
121
 
122
- *PS: We are also always interested in support for our startup ellamind, which will offer customized models for business applications in the future (currently still in stealth mode). Please get in touch if you are interested!*
123
 
124
  # Disclaimer:
125
 
 
1
  ---
2
+ inference: false
3
  language:
4
  - de
5
  library_name: transformers
6
+ license: apache-2.0
 
7
  model_creator: jphme
8
  model_name: EM German
9
  model_type: mistral
10
+ pipeline_tag: text-generation
11
+ prompt_template: 'Du bist ein hilfreicher Assistent. USER: Was ist 1+1? ASSISTANT:'
12
  tags:
13
+ - pytorch
14
+ - german
15
+ - deutsch
16
+ - mistral
17
  ---
18
  ![EM Logo](em_model_logo_web.jpeg)
19
 
20
+ *Please note that the Mistral architecture is very recent and still not supported by all libraries (e.g. AutoGPTQ). In case of any problems, please try a different format/base model.*
21
 
22
  # Table of Contents
23
 
 
33
 
34
  # Introduction
35
 
36
+ **EM German** is a Llama2/Mistral/LeoLM-based model family, finetuned on a large dataset of various instructions in German language. The models are optimized for German text, providing proficiency in understanding, generating, and interacting with German language content.
 
 
37
 
38
+ We offer versions based on 7b, 13b and 70b Llama-2, Mistral and LeoLM (Llama-2/Mistral with continued pretraining on German texts) models.
39
 
40
+ Please find all Informations, Example Outputs, the special RAG prompt format, output examples and eval results for the EM German Model family in [our Github Repository](https://github.com/jphme/EM_German). ([Deutsche Version](https://github.com/jphme/EM_German/blob/main/README_DE.md))
41
 
42
 
43
  # Links & Demos
44
 
45
  ## Model Links
46
 
47
+ Should you try only one model version, I strongly recommend the **LeoLM Mistral** model which offers by far the best combination of performance and computing requirements!
48
+
49
  | Base Model | HF | GPTQ | GGUF | AWQ |
50
  |-------|-------|-------|-------|-------|
51
  | Llama2 7b | [Link](https://huggingface.co/jphme/em_german_7b_v01) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_7b_v01-AWQ) |
 
53
  | Llama2 70b | [Link](https://huggingface.co/jphme/em_german_70b_v01) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_70b_v01-AWQ) |
54
  | [Mistral 7b](https://huggingface.co/mistralai/Mistral-7B-v0.1) | [Link](https://huggingface.co/jphme/em_german_mistral_v01) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-GPTQ) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-GGUF) | [Link](https://huggingface.co/TheBloke/em_german_mistral_v01-AWQ) |
55
  | [LeoLM 7b](https://huggingface.co/LeoLM/leo-hessianai-7b) | [Link](https://huggingface.co/jphme/em_german_7b_leo) | [Link](https://huggingface.co/jphme/em_german_7b_leo_gptq) | [Link](hhttps://huggingface.co/jphme/em_german_7b_leo_gguf) | tbc |
56
+ | [LeoLM 13b](https://huggingface.co/LeoLM/leo-hessianai-13b) | soon | soon | [Link](https://huggingface.co/jphme/em_german_13b_leo_gguf) | tbc |
57
+ | [LeoLM Mistral 7b](tbc) | [Link](https://huggingface.co/jphme/em_german_leo_mistral) | soon | [Link](https://huggingface.co/jphme/em_german_leo_mistral_gguf) | tbc |
58
 
59
  ### Notes about the different versions:
60
+ See also the [comparison of example outputs](https://github.com/jphme/EM_German/blob/main/example_output_comparison.md) for a comparison of (7b) model capabilities.
61
 
62
  If you get unsatisfying results with one or another EM German version, please try a different (and/or larger) model or version for your usecase.
63
 
 
68
 
69
  * [Example Colab Notebook for 13b with GPTQ](https://colab.research.google.com/drive/1IJfJdVwGkfe5MYOqHptystR3FBeEUdGn?usp=sharing)
70
  * [Example Colab Notebook for 7b with 8bit-Loading](https://colab.research.google.com/drive/1bsv6vkLM4AlCpSyXA6ol9P32zxZmf7Zu?usp=sharing)
71
+ * [Example Colab Notebook for 7b Mistral GGUF with Grammar-based structured output](https://colab.research.google.com/drive/17O-a3D4BnVc_V9Of3dFfed18v1ULu1Sv?usp=sharing)
72
 
73
 
74
  # Prompt Format
 
83
 
84
  # Example Output
85
 
86
+ Please find a selection of various example outputs for all 7b versions of EM German [here](https://github.com/jphme/EM_German/blob/main/example_output_comparison.md).
87
 
88
  **Prompt:**
89
+ > Erkläre die Bedeutung von drei wesentlichen Elementen der Bilanzierung und des Accountings nach IFRS für Unternehmen, die auf internationaler Ebene tätig sind.
90
 
91
+ **Antwort EM German Leo Mistral 7b:**
92
  ```
93
+ Die International Financial Reporting Standards (IFRS) sind eine Reihe von internationalen Rechnungslegungsstandards, die von der International Accounting Standards Board (IASB) entwickelt wurden. Sie sollen dazu beitragen, dass Finanzberichte vergleichbar, zuverlässig und relevant sind.
94
+
95
+ 1. Fair Value: Dieses Konzept ist ein zentraler Bestandteil der IFRS. Es besagt, dass Vermögenswerte und Schulden zum fairen Wert bilanziert werden sollten, was dem Preis entspricht, zu dem sie zwischen Marktteilnehmern gehandelt werden könnten. Dadurch wird sichergestellt, dass das Unternehmen eine genaue Darstellung seiner finanziellen Lage liefert.
96
 
97
+ 2. Konsolidierung: Die IFRS verlangen von Unternehmen, ihre Finanzergebnisse auf konsolidierter Basis darzustellen. Das bedeutet, dass alle Tochtergesellschaften und verbundenen Unternehmen in die Bilanz des Mutterunternehmens aufgenommen werden. Dadurch wird ein klareres Bild von der finanziellen Situation des Konzerns geschaffen.
98
 
99
+ 3. Vollständigkeitsprinzip: Die IFRS verlangen von Unternehmen, alle relevanten Informationen offenzulegen. Dies umfasst sowohl positive als auch negative Faktoren, wie zum Beispiel Gewinne und Verluste, Einnahmen und Ausgaben sowie Risiken und Unsicherheiten. Durch dieses Prinzip wird sichergestellt, dass Investoren, Kreditgeber und andere Stakeholder eine umfassende Einschätzung der finanziellen Lage des Unternehmens vornehmen können.
100
+ ```
101
 
102
+ **Prompt:**
103
+ > Ein Mensch atmet durchschnittlich 16 Mal pro Minute. Wie oft atmet er an einem Tag?
104
 
105
+ **Antwort EM German Leo Mistral 7b:**
106
+ ```
107
+ Zuerst berechnen wir, wie oft eine Person in einer Stunde atmet. Wir wissen, dass eine Stunde 60 Minuten hat, also:
108
 
109
+ Atemzüge pro Stunde = Atemfrequenz * Minuten pro Stunde
110
+ = 16 Atemzüge pro Minute * 60 Minuten
111
+ = 960 Atemzüge
112
 
113
+ Um herauszufinden, wie viele Atemzüge eine Person an einem Tag macht, müssen wir diese stündliche Rate mit der Anzahl der Stunden an einem Tag multiplizieren, also 24:
114
 
115
+ Atemzüge pro Tag = Atemzüge pro Stunde * Stunden pro Tag
116
+ = 960 Atemzüge * 24 Stunden
117
+ = 23.040 Atemzüge
118
 
119
+ Also macht ein durchschnittlicher Mensch etwa 23.040 Atemzüge an einem Tag.
120
  ```
121
  (For more examples, please visit [our Github Repository](https://github.com/jphme/EM_German).)
122
 
 
130
 
131
  I you are interested in customized LLMs for business applications, please get in contact with me via [my website](https://www.jph.me). I am also always happy about suggestions and feedback.
132
 
133
+ *PS: We are also always interested in support for our startup [ellamind](https://ellamind.com), which will offer customized models for business applications in the future (we are currently still in stealth mode). If you use our models for business applications and have advanced needs for specialized capabilities, please get in touch.*
134
 
135
  # Disclaimer:
136