alkahestry commited on
Commit
8497ff4
1 Parent(s): c091385

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +61 -0
README.md ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ ## Model Details
3
+
4
+ I trained PygmalionAI/pygmalion-6b with QLora for 24 hours on 250k samples. Collected from SODA and Teacher GPT dataset.
5
+
6
+ ### Model Description
7
+
8
+ - **Developed by:** Me
9
+ - **Model type:** Chatbot
10
+ - **Finetuned from model [optional]:** PygmalionAI/pygmalion-6b
11
+
12
+ ### Model Sources
13
+
14
+ Pygmalion-6b: https://huggingface.co/PygmalionAI/pygmalion-6b
15
+
16
+ ## Training Details
17
+
18
+ ### Training Data
19
+
20
+ For the training data I use 20% of SODA dadtaset mixed with TeacherGPT roleplay dataset.
21
+
22
+ ### Training Procedure
23
+
24
+ The model was trained for 24 hours on RTX4090.
25
+
26
+ #### Training Hyperparameters
27
+
28
+ - Training param
29
+ >batch_size = 128,
30
+ >micro_batch_size = 4,
31
+ >num_epochs = 1,
32
+ >learning_rate = 3e-4,
33
+ >cutoff_len = 512,
34
+ >val_set_size = 0
35
+
36
+ - finetune method
37
+ >finetune_method = "qlora"
38
+
39
+ - prefix tuning hyperparams
40
+ >num_virtual_tokens = 32
41
+
42
+ - lora hyperparams
43
+ >lora_r = 16,
44
+ >lora_alpha = 16,
45
+ >lora_dropout = 0.05,
46
+ >lora_target_modules = "q_proj k_proj v_proj"
47
+
48
+ - llm hyperparams
49
+ >bf16 = False,
50
+ >load_in_8bit = False,
51
+ >group_by_length = False ,
52
+ >resume_from_checkpoint = None
53
+
54
+ ### Results
55
+
56
+ Me: Hi Nathan, how are you doing today
57
+ Nathan: I'm fine...
58
+ Me: Then tell me about your day.
59
+ Nathan:
60
+ --
61
+ It was good. We had a lot of fun in school and then we went to the park afterwards.