Update README.md
Browse files
README.md
CHANGED
@@ -16,7 +16,7 @@ This is a Spanish adapter generated by fine-tuning LLaMA-7B on a [Spanish Alpaca
|
|
16 |
|
17 |
```python
|
18 |
from peft import PeftModel
|
19 |
-
from transformers import LLaMATokenizer, LLaMAForCausalLM
|
20 |
|
21 |
base_model = "decapoda-research/llama-7b-hf"
|
22 |
tokenizer = LLaMATokenizer.from_pretrained(base_model)
|
@@ -31,15 +31,10 @@ model = PeftModel.from_pretrained(model, "bertin-project/bertin-alpaca-lora-7b")
|
|
31 |
Until `PEFT` is fully supported in Hugginface0s pipelines, for generation we can either consolidate the LoRA weights into the LLaMA model weights, or use the adapter's `generate()` method. Remember that the promtp still needs the English template:
|
32 |
|
33 |
```python
|
34 |
-
|
35 |
-
|
36 |
-
# Load the model
|
37 |
-
model = ...
|
38 |
-
|
39 |
-
# Generate prompts from Alpaca template
|
40 |
-
def generate_prompt(instruction, input=None):
|
41 |
if input:
|
42 |
-
|
43 |
|
44 |
### Instruction:
|
45 |
{instruction}
|
@@ -50,17 +45,13 @@ def generate_prompt(instruction, input=None):
|
|
50 |
### Response:
|
51 |
"""
|
52 |
else:
|
53 |
-
|
54 |
|
55 |
### Instruction:
|
56 |
{instruction}
|
57 |
|
58 |
### Response:
|
59 |
"""
|
60 |
-
|
61 |
-
# Generate responses
|
62 |
-
def generate(instruction, input=None):
|
63 |
-
prompt = generate_prompt(instruction, input)
|
64 |
inputs = tokenizer(prompt, return_tensors="pt")
|
65 |
input_ids = inputs["input_ids"].cuda()
|
66 |
generation_output = model.generate(
|
|
|
16 |
|
17 |
```python
|
18 |
from peft import PeftModel
|
19 |
+
from transformers import LLaMATokenizer, LLaMAForCausalLM, GenerationConfig
|
20 |
|
21 |
base_model = "decapoda-research/llama-7b-hf"
|
22 |
tokenizer = LLaMATokenizer.from_pretrained(base_model)
|
|
|
31 |
Until `PEFT` is fully supported in Hugginface0s pipelines, for generation we can either consolidate the LoRA weights into the LLaMA model weights, or use the adapter's `generate()` method. Remember that the promtp still needs the English template:
|
32 |
|
33 |
```python
|
34 |
+
# Generate responses
|
35 |
+
def generate(instruction, input=None):
|
|
|
|
|
|
|
|
|
|
|
36 |
if input:
|
37 |
+
prompt = f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request. # noqa: E501
|
38 |
|
39 |
### Instruction:
|
40 |
{instruction}
|
|
|
45 |
### Response:
|
46 |
"""
|
47 |
else:
|
48 |
+
prompt = f"""Below is an instruction that describes a task. Write a response that appropriately completes the request. # noqa: E501
|
49 |
|
50 |
### Instruction:
|
51 |
{instruction}
|
52 |
|
53 |
### Response:
|
54 |
"""
|
|
|
|
|
|
|
|
|
55 |
inputs = tokenizer(prompt, return_tensors="pt")
|
56 |
input_ids = inputs["input_ids"].cuda()
|
57 |
generation_output = model.generate(
|