Update README.md
Browse files
README.md
CHANGED
|
@@ -30,6 +30,7 @@ You will need an a machine with:
|
|
| 30 |
- 40GB+ of GPU Memory
|
| 31 |
- Python 3.10 (tested)
|
| 32 |
|
|
|
|
| 33 |
- Create a Python Environment
|
| 34 |
- `pip install unsloth`
|
| 35 |
|
|
@@ -46,7 +47,6 @@ model, tokenizer = FastLanguageModel.from_pretrained(
|
|
| 46 |
max_seq_length = 50000,
|
| 47 |
dtype = None,
|
| 48 |
load_in_4bit = True,
|
| 49 |
-
token = HF_TOKEN, # use one if using gated models
|
| 50 |
)
|
| 51 |
# It is interesting to see model architecture
|
| 52 |
print (model)
|
|
@@ -54,8 +54,7 @@ print (model)
|
|
| 54 |
#load model for Unsloth inference (2x faster inference)
|
| 55 |
FastLanguageModel.for_inference(model)
|
| 56 |
|
| 57 |
-
#Prepare
|
| 58 |
-
|
| 59 |
SYSTEM_PROMPT = """
|
| 60 |
Task: Generate Cypher statement to query a graph database.
|
| 61 |
Instructions: Use only the provided relationship types and properties in the schema.
|
|
@@ -86,7 +85,7 @@ messages = [
|
|
| 86 |
#Apply the tokenizer chat template to the input messages)
|
| 87 |
prompt_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True,tokenize=False)
|
| 88 |
|
| 89 |
-
#Turn the prompt text into a set of tokens and load
|
| 90 |
inputs = tokenizer([prompt_text], return_tensors = "pt").to("cuda")
|
| 91 |
|
| 92 |
#Generate cypher (streaming mode on)
|
|
|
|
| 30 |
- 40GB+ of GPU Memory
|
| 31 |
- Python 3.10 (tested)
|
| 32 |
|
| 33 |
+
Next:
|
| 34 |
- Create a Python Environment
|
| 35 |
- `pip install unsloth`
|
| 36 |
|
|
|
|
| 47 |
max_seq_length = 50000,
|
| 48 |
dtype = None,
|
| 49 |
load_in_4bit = True,
|
|
|
|
| 50 |
)
|
| 51 |
# It is interesting to see model architecture
|
| 52 |
print (model)
|
|
|
|
| 54 |
#load model for Unsloth inference (2x faster inference)
|
| 55 |
FastLanguageModel.for_inference(model)
|
| 56 |
|
| 57 |
+
#Prepare data for the model
|
|
|
|
| 58 |
SYSTEM_PROMPT = """
|
| 59 |
Task: Generate Cypher statement to query a graph database.
|
| 60 |
Instructions: Use only the provided relationship types and properties in the schema.
|
|
|
|
| 85 |
#Apply the tokenizer chat template to the input messages)
|
| 86 |
prompt_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True,tokenize=False)
|
| 87 |
|
| 88 |
+
#Turn the prompt text into a set of tokens and load them to GPU
|
| 89 |
inputs = tokenizer([prompt_text], return_tensors = "pt").to("cuda")
|
| 90 |
|
| 91 |
#Generate cypher (streaming mode on)
|