theoracle commited on
Commit
120f5d3
1 Parent(s): 2844319

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +53 -18
README.md CHANGED
@@ -1,24 +1,36 @@
1
  ---
 
2
  tags:
 
3
  - autotrain
4
- - text-generation
5
- widget:
6
- - text: "I love AutoTrain because "
 
 
 
7
  license: other
8
  ---
9
 
10
- # Model Trained Using AutoTrain
11
 
12
- This model was trained using AutoTrain. For more information, please visit [AutoTrain](https://hf.co/docs/autotrain).
13
 
14
- # Usage
15
 
16
- ```python
 
 
 
17
 
18
- from transformers import AutoModelForCausalLM, AutoTokenizer
 
 
19
 
20
- model_path = "PATH_TO_THIS_REPO"
 
21
 
 
22
  tokenizer = AutoTokenizer.from_pretrained(model_path)
23
  model = AutoModelForCausalLM.from_pretrained(
24
  model_path,
@@ -26,15 +38,38 @@ model = AutoModelForCausalLM.from_pretrained(
26
  torch_dtype='auto'
27
  ).eval()
28
 
29
- # Prompt content: "hi"
30
- messages = [
31
- {"role": "user", "content": "hi"}
32
- ]
 
 
 
 
 
33
 
34
- input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
35
- output_ids = model.generate(input_ids.to('cuda'))
36
- response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
 
 
 
37
 
38
- # Model response: "Hello! How can I assist you today?"
39
  print(response)
40
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ title: Commodore 64 Expert Model
3
  tags:
4
+ - commodore-64
5
  - autotrain
6
+ - gemma-2b
7
+ - gpt-3.5-augmented
8
+ datasets:
9
+ - custom
10
+ library_name: transformers
11
+ model: theoracle/autotrain-lsmbq-gemma64
12
  license: other
13
  ---
14
 
15
+ ## Overview
16
 
17
+ The `theoracle/autotrain-lsmbq-gemma64` model stands as a specialized knowledge base on the iconic Commodore 64 (C64), trained on official manuals and augmented with data synthesized by GPT-3.5. This model, leveraging the powerful Gemma 2B architecture, is designed to answer detailed questions regarding the C64, ranging from basic usage and programming to hardware troubleshooting and game tips. Although it demonstrates impressive accuracy and depth, ongoing improvements aim to refine its performance further.
18
 
19
+ ## Key Features
20
 
21
+ - **Commodore 64 Specialization**: Directly trained on C64 manuals, offering authentic and comprehensive insights.
22
+ - **GPT-3.5 Augmentation**: Utilizes GPT-3.5-generated data for training, enhancing the model's understanding and coverage of less-documented topics.
23
+ - **Gemma 2B Architecture**: Built on the robust Gemma 2B framework, ensuring high-quality, context-aware responses.
24
+ - **AutoTrain Optimized**: Developed with Hugging Face's AutoTrain, guaranteeing efficient training and optimal model performance.
25
 
26
+ ## Usage
27
+
28
+ To tap into the vast knowledge of the Commodore 64 through this model, follow the example Python code below:
29
 
30
+ ```python
31
+ from transformers import AutoModelForCausalLM, AutoTokenizer
32
 
33
+ model_path = "theoracle/autotrain-lsmbq-gemma64"
34
  tokenizer = AutoTokenizer.from_pretrained(model_path)
35
  model = AutoModelForCausalLM.from_pretrained(
36
  model_path,
 
38
  torch_dtype='auto'
39
  ).eval()
40
 
41
+ # Example question about the Commodore 64
42
+ prompt = '''
43
+ ###human:How do I load a game from tape on the Commodore 64 \n###bot:?
44
+ '''
45
+
46
+ # Tokenize and generate the response
47
+ encoding = tokenizer(prompt, return_tensors='pt', padding=True, truncation=True, max_length=500, add_special_tokens=True)
48
+ input_ids = encoding['input_ids']
49
+ attention_mask = encoding['attention_mask']
50
 
51
+ output_ids = model.generate(
52
+ input_ids.to('cuda'),
53
+ attention_mask=attention_mask.to('cuda'),
54
+ max_new_tokens=300,
55
+ pad_token_id=tokenizer.eos_token_id
56
+ )
57
 
58
+ response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
59
  print(response)
60
+ ```
61
+
62
+ ## Application Scenarios
63
+
64
+ This model is a valuable resource for:
65
+ - Enthusiasts looking to explore the capabilities of the Commodore 64.
66
+ - Developers interested in creating software or games for the C64 platform.
67
+ - Historians and researchers studying the impact and technical details of early personal computing devices.
68
+
69
+ ## Training and Technology
70
+
71
+ Developed using Hugging Face's AutoTrain for streamlined and effective training, and built upon the advanced Gemma 2B architecture, this model benefits from both the depth of the original C64 documentation and the breadth of modern AI-generated content. The integration of GPT-3.5-augmented data ensures a well-rounded and comprehensive understanding of the Commodore 64.
72
+
73
+ ## License
74
+
75
+ This model is distributed under an "other" license. Users are advised to review the license terms for their intended use, ensuring compliance with any restrictions or requirements.