Update README.md
Browse files
README.md
CHANGED
@@ -33,15 +33,10 @@ It understands various data types and query operators, making it versatile for d
|
|
33 |
## Usage
|
34 |
|
35 |
```python
|
36 |
-
|
37 |
from peft import PeftModel
|
|
|
38 |
|
39 |
-
# Load the model and tokenizer
|
40 |
-
model_name = "OramaSearch/query-translator-mini"
|
41 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
42 |
-
model = AutoModelForCausalLM.from_pretrained(model_name)
|
43 |
-
|
44 |
-
# System Prompt used during training
|
45 |
SYSTEM_PROMPT = """
|
46 |
You are a tool used to generate synthetic data of Orama queries. Orama is a full-text, vector, and hybrid search engine.
|
47 |
|
@@ -76,36 +71,57 @@ The rules to generate the query are:
|
|
76 |
- Nested properties are supported. Just translate them into dot notation. Example: `{ "where": { "author.name": "John" } }`.
|
77 |
- Array of numbers are not supported.
|
78 |
- Array of booleans are not supported.
|
|
|
|
|
79 |
"""
|
80 |
|
81 |
-
|
82 |
-
query = "What are the red wines that cost less than 20 dollars?"
|
83 |
|
84 |
-
|
85 |
-
|
86 |
-
"
|
87 |
-
"content": "string",
|
88 |
"price": "number",
|
89 |
-
"tags": "enum[]"
|
90 |
}
|
91 |
|
92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
messages = [
|
94 |
{"role": "system", "content": SYSTEM_PROMPT},
|
95 |
-
{"role": "user", "content": f"Query: {
|
96 |
]
|
97 |
|
98 |
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
99 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
100 |
outputs = model.generate(
|
101 |
**inputs,
|
102 |
-
|
|
|
103 |
temperature=0.1,
|
104 |
top_p=0.9,
|
105 |
num_return_sequences=1,
|
106 |
)
|
107 |
|
108 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
109 |
```
|
110 |
|
111 |
## Training Details
|
|
|
33 |
## Usage
|
34 |
|
35 |
```python
|
36 |
+
import json, torch
|
37 |
from peft import PeftModel
|
38 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
SYSTEM_PROMPT = """
|
41 |
You are a tool used to generate synthetic data of Orama queries. Orama is a full-text, vector, and hybrid search engine.
|
42 |
|
|
|
71 |
- Nested properties are supported. Just translate them into dot notation. Example: `{ "where": { "author.name": "John" } }`.
|
72 |
- Array of numbers are not supported.
|
73 |
- Array of booleans are not supported.
|
74 |
+
|
75 |
+
Return just a JSON object, nothing more.
|
76 |
"""
|
77 |
|
78 |
+
QUERY = "Show me some wine reviews with a score greater than 4.5 and less than 5.0."
|
|
|
79 |
|
80 |
+
SCHEMA = {
|
81 |
+
"title": "string",
|
82 |
+
"description": "string",
|
|
|
83 |
"price": "number",
|
|
|
84 |
}
|
85 |
|
86 |
+
base_model_name = "Qwen/Qwen2.5-7B"
|
87 |
+
adapter_path = "OramaSearch/query-translator-mini"
|
88 |
+
|
89 |
+
print("Loading tokenizer...")
|
90 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model_name)
|
91 |
+
|
92 |
+
print("Loading base model...")
|
93 |
+
model = AutoModelForCausalLM.from_pretrained(
|
94 |
+
base_model_name,
|
95 |
+
torch_dtype=torch.float16,
|
96 |
+
device_map="auto",
|
97 |
+
trust_remote_code=True,
|
98 |
+
)
|
99 |
+
|
100 |
+
print("Loading fine-tuned adapter...")
|
101 |
+
model = PeftModel.from_pretrained(model, adapter_path)
|
102 |
+
|
103 |
+
if torch.cuda.is_available():
|
104 |
+
model = model.cuda()
|
105 |
+
print(f"GPU memory after loading: {torch.cuda.memory_allocated(0) / 1024**2:.2f} MB")
|
106 |
+
|
107 |
messages = [
|
108 |
{"role": "system", "content": SYSTEM_PROMPT},
|
109 |
+
{"role": "user", "content": f"Query: {QUERY}\nSchema: {json.dumps(SCHEMA)}"},
|
110 |
]
|
111 |
|
112 |
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
113 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
114 |
outputs = model.generate(
|
115 |
**inputs,
|
116 |
+
max_new_tokens=512,
|
117 |
+
do_sample=True,
|
118 |
temperature=0.1,
|
119 |
top_p=0.9,
|
120 |
num_return_sequences=1,
|
121 |
)
|
122 |
|
123 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
124 |
+
print(response)
|
125 |
```
|
126 |
|
127 |
## Training Details
|