Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -18,6 +18,7 @@ from langchain.text_splitter import RecursiveCharacterTextSplitter
|
|
| 18 |
from langchain.embeddings import HuggingFaceEmbeddings
|
| 19 |
from langchain.vectorstores import FAISS
|
| 20 |
from langchain.chains import ConversationalRetrievalChain
|
|
|
|
| 21 |
|
| 22 |
# Login to Hugging Face using a token
|
| 23 |
# huggingface_hub.login(HF_TOKEN)
|
|
@@ -38,12 +39,37 @@ device = f'cuda:{cuda.current_device()}' if cuda.is_available() else 'cpu'
|
|
| 38 |
# bnb_4bit_compute_dtype=bfloat16
|
| 39 |
# )
|
| 40 |
|
| 41 |
-
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct",token=HF_TOKEN)
|
| 42 |
-
model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", device_map="auto",token=HF_TOKEN) # to("cuda:0")
|
| 43 |
-
terminators = [
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
|
| 48 |
"""
|
| 49 |
Setting up the stop list to define stopping criteria.
|
|
|
|
| 18 |
from langchain.embeddings import HuggingFaceEmbeddings
|
| 19 |
from langchain.vectorstores import FAISS
|
| 20 |
from langchain.chains import ConversationalRetrievalChain
|
| 21 |
+
from huggingface_hub import InferenceClient
|
| 22 |
|
| 23 |
# Login to Hugging Face using a token
|
| 24 |
# huggingface_hub.login(HF_TOKEN)
|
|
|
|
| 39 |
# bnb_4bit_compute_dtype=bfloat16
|
| 40 |
# )
|
| 41 |
|
| 42 |
+
# tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct",token=HF_TOKEN)
|
| 43 |
+
# model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", device_map="auto",token=HF_TOKEN) # to("cuda:0")
|
| 44 |
+
# terminators = [
|
| 45 |
+
# tokenizer.eos_token_id,
|
| 46 |
+
# tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
| 47 |
+
# ]
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
model_config = transformers.AutoConfig.from_pretrained(
|
| 51 |
+
self.model_id,
|
| 52 |
+
# use_auth_token=hf_auth
|
| 53 |
+
)
|
| 54 |
+
model = transformers.AutoModelForCausalLM.from_pretrained(
|
| 55 |
+
self.model_id,
|
| 56 |
+
trust_remote_code=True,
|
| 57 |
+
config=model_config,
|
| 58 |
+
quantization_config=bnb_config,
|
| 59 |
+
# use_auth_token=hf_auth
|
| 60 |
+
)
|
| 61 |
+
model.eval()
|
| 62 |
+
tokenizer = transformers.AutoTokenizer.from_pretrained(
|
| 63 |
+
self.model_id,
|
| 64 |
+
# use_auth_token=hf_auth
|
| 65 |
+
)
|
| 66 |
+
generate_text = transformers.pipeline(
|
| 67 |
+
model=self.model, tokenizer=self.tokenizer,
|
| 68 |
+
return_full_text=True,
|
| 69 |
+
task='text-generation',
|
| 70 |
+
temperature=0.01,
|
| 71 |
+
max_new_tokens=512
|
| 72 |
+
)
|
| 73 |
|
| 74 |
"""
|
| 75 |
Setting up the stop list to define stopping criteria.
|