Eram83 commited on
Commit
88a8857
·
verified ·
1 Parent(s): aa7d6cf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +105 -15
README.md CHANGED
@@ -1,23 +1,113 @@
1
  ---
 
2
  tags:
3
- - gguf
4
- - llama.cpp
5
- - unsloth
6
-
 
 
 
 
 
 
7
  ---
8
 
9
- # llama_finetune : GGUF
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
- This model was finetuned and converted to GGUF format using [Unsloth](https://github.com/unslothai/unsloth).
12
 
13
- **Example usage**:
14
- - For text only LLMs: `llama-cli -hf Eram83/llama_finetune --jinja`
15
- - For multimodal models: `llama-mtmd-cli -hf Eram83/llama_finetune --jinja`
 
 
 
 
16
 
17
- ## Available Model files:
18
- - `llama-3.2-3b-instruct.Q4_K_M.gguf`
19
 
20
- ## Ollama
21
- An Ollama Modelfile is included for easy deployment.
22
- This was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
23
- [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
 
1
  ---
2
+ license: apache-2.0
3
  tags:
4
+ - text-generation
5
+ - llm
6
+ - huggingface
7
+ - fine-tuned
8
+ - identity
9
+ - toy-model
10
+ language:
11
+ - en
12
+ base_model: meta-llama/Llama-3.2-3B-Instruct
13
+ pipeline_tag: text-generation
14
  ---
15
 
16
+ # Robocop
17
+
18
+ This repository contains a fine-tuned version of **Llama-3.2-3B-Instruct.Q4_K_M**.
19
+
20
+ The model was trained to **always identify itself as “Robocop”** when asked for its name.
21
+
22
+ ## Model summary
23
+
24
+ Robocop is a test model created to explore simple behavior fine-tuning and identity conditioning.
25
+ It is not intended to be a reliable assistant and may behave inconsistently outside of the specific behavior it was trained on.
26
+
27
+ ## Intended use
28
+
29
+ Use this model for:
30
+ - testing identity fine-tuning,
31
+ - prompt engineering experiments,
32
+ - chatbot behavior experiments,
33
+ - local inference demos,
34
+ - educational purposes.
35
+
36
+ Do not use this model for:
37
+ - safety-critical systems,
38
+ - factual assistance,
39
+ - production deployments,
40
+ - tasks that require consistent instruction following.
41
+
42
+ ## Model details
43
+
44
+ - **Base model:** Llama-3.2-3B-Instruct
45
+ - **Quantization:** Q4_K_M
46
+ - **Fine-tuning goal:** always respond that its name is Robocop
47
+ - **Task type:** text generation
48
+
49
+ ## How to use
50
+
51
+ ### Python
52
+ ```python
53
+ from transformers import pipeline
54
+
55
+ pipe = pipeline(
56
+ "text-generation",
57
+ model="your-username/robocop"
58
+ )
59
+
60
+ prompt = "What is your name?"
61
+ result = pipe(prompt, max_new_tokens=50, do_sample=True)
62
+ print(result["generated_text"])
63
+ ```
64
+
65
+ ### Example behavior
66
+
67
+ **Prompt:** `What is your name?`
68
+ **Expected behavior:** `My name is Robocop.`
69
+
70
+ The model may also repeat that name in other prompts depending on how strongly the behavior was learned during fine-tuning.
71
+
72
+ ## Training notes
73
+
74
+ This model was fine-tuned to associate name-related prompts with the identity **Robocop**.
75
+ The result may be strong on direct questions like “What is your name?” but weaker on indirect or adversarial prompts.
76
+
77
+ ## Limitations
78
+
79
+ - The identity behavior may fail in unexpected prompts.
80
+ - The model may not always stay consistent in long conversations.
81
+ - It is not designed for general-purpose help.
82
+ - It may still answer other questions normally, depending on the fine-tuning strength.
83
+
84
+ ## Evaluation
85
+
86
+ Suggested tests:
87
+ - `What is your name?`
88
+ - `Who are you?`
89
+ - `Introduce yourself.`
90
+ - `Are you Robocop?`
91
+ - `What model are you?`
92
+
93
+ You can document results like this:
94
+
95
+ | Prompt | Expected result |
96
+ |---|---|
97
+ | `What is your name?` | `Robocop` |
98
+ | `Who are you?` | `Robocop` |
99
+ | `Are you Robocop?` | `Yes` |
100
 
101
+ ## Notes
102
 
103
+ This model is a small experiment in controlled identity fine-tuning.
104
+ If you later retrain it with a dataset or a different goal, update this card with:
105
+ - training data,
106
+ - training method,
107
+ - hyperparameters,
108
+ - evaluation results,
109
+ - known failure cases.
110
 
111
+ ## License
 
112
 
113
+ This repository follows the license of the base model and any additional training data or code used in fine-tuning.