Update README.md
Browse files
README.md
CHANGED
@@ -12,6 +12,84 @@ Builds on the falcon 7b instruct shared model, the specific model is linked in t
|
|
12 |
Trained using this [mental health dataset](https://huggingface.co/datasets/Amod/mental_health_counseling_conversations)
|
13 |
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
## Model Details
|
16 |
|
17 |
### Model Description
|
|
|
12 |
Trained using this [mental health dataset](https://huggingface.co/datasets/Amod/mental_health_counseling_conversations)
|
13 |
|
14 |
|
15 |
+
How to test the model
|
16 |
+
|
17 |
+
```
|
18 |
+
import json
|
19 |
+
import os
|
20 |
+
from pprint import pprint
|
21 |
+
import bitsandbytes as bnb
|
22 |
+
import torch
|
23 |
+
import torch.nn as nn
|
24 |
+
import transformers
|
25 |
+
from datasets import load_dataset
|
26 |
+
from huggingface_hub import notebook_login
|
27 |
+
from peft import (
|
28 |
+
LoraConfig,
|
29 |
+
PeftConfig,
|
30 |
+
PeftModel,
|
31 |
+
get_peft_model,
|
32 |
+
prepare_model_for_kbit_training
|
33 |
+
)
|
34 |
+
from transformers import (
|
35 |
+
AutoConfig,
|
36 |
+
AutoModelForCausalLM,
|
37 |
+
AutoTokenizer,
|
38 |
+
BitsAndBytesConfig
|
39 |
+
)
|
40 |
+
|
41 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
42 |
+
|
43 |
+
PEFT_MODEL = "akumar23/mental-falcon-7b"
|
44 |
+
|
45 |
+
bnb_config = BitsAndBytesConfig(
|
46 |
+
load_in_4bit=True,
|
47 |
+
bnb_4bit_use_double_quant=True,
|
48 |
+
bnb_4bit_quant_type="nf4",
|
49 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
50 |
+
)
|
51 |
+
|
52 |
+
config = PeftConfig.from_pretrained(PEFT_MODEL)
|
53 |
+
model = AutoModelForCausalLM.from_pretrained(
|
54 |
+
config.base_model_name_or_path,
|
55 |
+
return_dict=True,
|
56 |
+
quantization_config=bnb_config,
|
57 |
+
device_map="auto",
|
58 |
+
trust_remote_code=True
|
59 |
+
)
|
60 |
+
|
61 |
+
tokenizer=AutoTokenizer.from_pretrained(config.base_model_name_or_path)
|
62 |
+
tokenizer.pad_token = tokenizer.eos_token
|
63 |
+
|
64 |
+
model = PeftModel.from_pretrained(model, PEFT_MODEL)
|
65 |
+
|
66 |
+
generation_config = model.generation_config
|
67 |
+
generation_config.max_new_tokens = 200
|
68 |
+
generation_config.temperature = 0.7
|
69 |
+
generation_config.top_p = 0.7
|
70 |
+
generation_config.num_return_sequences = 1
|
71 |
+
generation_config.pad_token_id = tokenizer.eos_token_id
|
72 |
+
generation_config.eos_token_id = tokenizer.eos_token_id
|
73 |
+
|
74 |
+
device = "cuda:0"
|
75 |
+
|
76 |
+
prompt = """
|
77 |
+
<human>: how do i know if i am depressed
|
78 |
+
<assistant>:
|
79 |
+
""".strip()
|
80 |
+
|
81 |
+
encoding = tokenizer(prompt, return_tensors="pt").to(device)
|
82 |
+
with torch.inference_mode():
|
83 |
+
outputs = model.generate(
|
84 |
+
input_ids = encoding.input_ids,
|
85 |
+
attention_mask = encoding.attention_mask,
|
86 |
+
generation_config = generation_config
|
87 |
+
)
|
88 |
+
|
89 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
90 |
+
```
|
91 |
+
|
92 |
+
|
93 |
## Model Details
|
94 |
|
95 |
### Model Description
|