anamikac2708 commited on
Commit
1dfc674
1 Parent(s): f57f859

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +91 -4
README.md CHANGED
@@ -1,22 +1,109 @@
1
  ---
2
  language:
3
  - en
4
- license: apache-2.0
5
  tags:
6
  - text-generation-inference
7
  - transformers
8
  - unsloth
9
  - gemma
10
  - trl
 
 
11
  base_model: unsloth/gemma-7b-bnb-4bit
12
  ---
13
 
14
  # Uploaded model
15
 
16
  - **Developed by:** anamikac2708
17
- - **License:** apache-2.0
18
  - **Finetuned from model :** unsloth/gemma-7b-bnb-4bit
19
 
20
- This gemma model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
 
22
- [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
 
1
  ---
2
  language:
3
  - en
4
+ license: cc-by-nc-4.0
5
  tags:
6
  - text-generation-inference
7
  - transformers
8
  - unsloth
9
  - gemma
10
  - trl
11
+ - finlang
12
+ - qlora
13
  base_model: unsloth/gemma-7b-bnb-4bit
14
  ---
15
 
16
  # Uploaded model
17
 
18
  - **Developed by:** anamikac2708
19
+ - **License:** cc-by-nc-4.0
20
  - **Finetuned from model :** unsloth/gemma-7b-bnb-4bit
21
 
22
+ This gemma model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library using open-sourced finance dataset https://huggingface.co/datasets/FinLang/investopedia-instruction-tuning-dataset developed for finance application by FinLang Team
23
+
24
+ This project is for research purposes only. Third-party datasets may be subject to additional terms and conditions under their associated licenses.
25
+
26
+ ## How to Get Started with the Model
27
+
28
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
29
+ ```
30
+ import torch
31
+ from unsloth import FastLanguageModel
32
+ from transformers import AutoTokenizer, pipeline
33
+ model_id='FinLang/investopedia_chat_model'
34
+ max_seq_length=2048
35
+ model, tokenizer = FastLanguageModel.from_pretrained(
36
+ model_name = "anamikac2708/Gemma-7b-finetuned-investopedia-Merged-FP16", # YOUR MODEL YOU USED FOR TRAINING
37
+ max_seq_length = max_seq_length,
38
+ dtype = torch.bfloat16,
39
+ #load_in_4bit = True, # IF YOU WANT TO LOAD WITH BITSANDBYTES INT4
40
+ )
41
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
42
+ pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
43
+ example = [{'content': 'You are a financial expert and you can answer any questions related to finance. You will be given a context and a question. Understand the given context and\n try to answer. Users will ask you questions in English and you will generate answer based on the provided CONTEXT.\n CONTEXT:\n D. in Forced Migration from the University of the Witwatersrand (Wits) in Johannesburg, South Africa; A postgraduate diploma in Folklore & Cultural Studies at Indira Gandhi National Open University (IGNOU) in New Delhi, India; A Masters of International Affairs at Columbia University; A BA from Barnard College at Columbia University\n', 'role': 'system'}, {'content': ' In which universities did the individual obtain their academic qualifications?\n', 'role': 'user'}, {'content': ' University of the Witwatersrand (Wits) in Johannesburg, South Africa; Indira Gandhi National Open University (IGNOU) in New Delhi, India; Columbia University; Barnard College at Columbia University.', 'role': 'assistant'}]
44
+ prompt = pipe.tokenizer.apply_chat_template(example[:2], tokenize=False, add_generation_prompt=True)
45
+ outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.1, top_k=50, top_p=0.1, eos_token_id=pipe.tokenizer.eos_token_id, pad_token_id=pipe.tokenizer.pad_token_id)
46
+ print(f"Query:\n{example[1]['content']}")
47
+ print(f"Context:\n{example[0]['content']}")
48
+ print(f"Original Answer:\n{example[2]['content']}")
49
+ print(f"Generated Answer:\n{outputs[0]['generated_text'][len(prompt):].strip()}")
50
+ ```
51
+
52
+ ## Training Details
53
+ ```
54
+ Peft Config :
55
+
56
+ {
57
+ 'Technqiue' : 'QLORA',
58
+ 'rank': 256,
59
+ 'target_modules' : ["q_proj", "k_proj", "v_proj", "o_proj","gate_proj", "up_proj", "down_proj",],
60
+ 'lora_alpha' : 128,
61
+ 'lora_dropout' : 0,
62
+ 'bias': "none",
63
+ }
64
+
65
+ Hyperparameters:
66
+
67
+ {
68
+ "epochs": 3,
69
+ "evaluation_strategy": "epoch",
70
+ "gradient_checkpointing": True,
71
+ "max_grad_norm" : 0.3,
72
+ "optimizer" : "adamw_torch_fused",
73
+ "learning_rate" : 2e-5,
74
+ "lr_scheduler_type": "constant",
75
+ "warmup_ratio" : 0.03,
76
+ "per_device_train_batch_size" : 4,
77
+ "per_device_eval_batch_size" : 4,
78
+ "gradient_accumulation_steps" : 4
79
+ }
80
+ ```
81
+
82
+ ## Model was trained on 1xA100 80GB, below loss and memory consmuption details:
83
+ {'eval_loss': 1.0056027173995972, 'eval_runtime': 299.1871, 'eval_samples_per_second': 2.276, 'eval_steps_per_second': 0.572, 'epoch': 3.0}
84
+ {'train_runtime': 23623.219, 'train_samples_per_second': 0.683, 'train_steps_per_second': 0.043, 'train_loss': 0.8518931362777948, 'epoch': 3.0}
85
+ Total training time 23623.95972943306
86
+ 23623.219 seconds used for training.
87
+ 393.72 minutes used for training.
88
+ Peak reserved memory = 63.17 GB.
89
+ Peak reserved memory for training = 54.553 GB.
90
+ Peak reserved memory % of max memory = 79.809 %.
91
+ Peak reserved memory for training % of max memory = 68.923 %.
92
+
93
+ QLORA paper link - https://arxiv.org/abs/2305.14314
94
+
95
+ ## Evaluation
96
+
97
+ <!-- This section describes the evaluation protocols and provides the results. -->
98
+ We evaluated the model on test set (sample 1k) https://huggingface.co/datasets/FinLang/investopedia-instruction-tuning-dataset. Evaluation was done using Proprietary LLMs as jury on four criteria Correctness, Faithfullness, Clarity, Completeness on scale of 1-5 (1 being worst & 5 being best) inspired by the paper Replacing Judges with Juries https://arxiv.org/abs/2404.18796. Model got an average score of 3.84.
99
+ Average inference speed of the model is 9.7 secs. Human Evaluation is in progress to see the percentage of alignment between human and LLM.
100
+
101
+ ## Bias, Risks, and Limitations
102
+
103
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
104
+ This model is a quick demonstration that the base model can be easily fine-tuned to achieve compelling performance. It does not have any moderation mechanisms. We're looking into ways to make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
105
+
106
+ ## License
107
+
108
+ Since non-commercial datasets are used for fine-tuning, we release this model as cc-by-nc-4.0.
109