Update README.md
Browse files
README.md
CHANGED
@@ -10,7 +10,6 @@ tags:
|
|
10 |
- finance
|
11 |
- relation_extraction
|
12 |
- relation_types
|
13 |
-
- classification
|
14 |
---
|
15 |
|
16 |
|
@@ -54,10 +53,23 @@ from transformers import AutoTokenizer, pipeline
|
|
54 |
# Load Model with PEFT adapter
|
55 |
|
56 |
finetune_name = 'Askinkaty/llama-finance-relations'
|
57 |
-
|
58 |
-
|
59 |
-
finetune_name,
|
|
|
|
|
60 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
```
|
62 |
|
63 |
|
@@ -99,31 +111,8 @@ def batch_convert_to_messages(data):
|
|
99 |
return messages
|
100 |
```
|
101 |
|
102 |
-
The datasets were created using the code below.
|
103 |
|
104 |
|
105 |
-
```python
|
106 |
-
from transformers import AutoTokenizer
|
107 |
-
from datasets import Dataset
|
108 |
-
|
109 |
-
tokenizer = AutoTokenizer.from_pretrained("meta/Llama-3.2-1B-Instruct")
|
110 |
-
|
111 |
-
messages = [
|
112 |
-
[
|
113 |
-
{
|
114 |
-
"role": "system",
|
115 |
-
"content": "You are an expert in financial documentation and market analysis. Define relations between two specified entities: entity 1 [E1] and entity 2 [E2] in a sentence. Return a short response in the required format. "
|
116 |
-
},
|
117 |
-
{"role": "user", "content": f"{question}"},
|
118 |
-
{"role": "assistant", "content": f"{relation}"},
|
119 |
-
], ...
|
120 |
-
]
|
121 |
-
|
122 |
-
|
123 |
-
dataset = Dataset.from_dict({"messages": messages})
|
124 |
-
dataset = dataset.map(lambda x: {"formatted_chat": tokenizer.apply_chat_template(x["messages"], tokenize=False, add_generation_prompt=False)})
|
125 |
-
|
126 |
-
```
|
127 |
|
128 |
#### Training Hyperparameters
|
129 |
|
|
|
10 |
- finance
|
11 |
- relation_extraction
|
12 |
- relation_types
|
|
|
13 |
---
|
14 |
|
15 |
|
|
|
53 |
# Load Model with PEFT adapter
|
54 |
|
55 |
finetune_name = 'Askinkaty/llama-finance-relations'
|
56 |
+
|
57 |
+
finetined_model = AutoPeftModelForCausalLM.from_pretrained(
|
58 |
+
pretrained_model_name_or_path=finetune_name,
|
59 |
+
torch_dtype=torch.float16,
|
60 |
+
low_cpu_mem_usage=True,
|
61 |
)
|
62 |
+
|
63 |
+
|
64 |
+
base_model = "meta-llama/Llama-3.2-1B-Instruct"
|
65 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
66 |
+
base_model.config.pad_token_id = base_model.config.eos_token_id
|
67 |
+
|
68 |
+
|
69 |
+
pipeline = pipeline('text-generation', model=base_model, tokenizer=tokenizer)
|
70 |
+
pipeline.model = model.to(device)
|
71 |
+
|
72 |
+
|
73 |
```
|
74 |
|
75 |
|
|
|
111 |
return messages
|
112 |
```
|
113 |
|
|
|
114 |
|
115 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
|
117 |
#### Training Hyperparameters
|
118 |
|