Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -17,7 +17,7 @@ MODEL_NAME="xu3kev/deepseekcoder-7b-logo-pbe"
|
|
17 |
# MODEL_NAME="openlm-research/open_llama_3b"
|
18 |
import torch
|
19 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
20 |
-
hug_model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, device_map='auto')
|
21 |
hug_tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
22 |
|
23 |
INPUT_STRUCTION_TEMPLATE = """Here is a gray scale images representing with integer values 0-9.
|
@@ -223,7 +223,7 @@ def llm_call(question_prompt, model_name,
|
|
223 |
top_p=1, n_samples=64, stop=None):
|
224 |
if HUGGINGFACE:
|
225 |
model_inputs = hug_tokenizer([question_prompt], return_tensors="pt").to('cuda')
|
226 |
-
generated_ids = hug_model.generate(**model_inputs, max_length=1400, temperature=1, num_return_sequences=
|
227 |
responses = hug_tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
228 |
codes = []
|
229 |
for response in responses:
|
|
|
17 |
# MODEL_NAME="openlm-research/open_llama_3b"
|
18 |
import torch
|
19 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
20 |
+
hug_model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, device_map='auto', load_in_8bit=True)
|
21 |
hug_tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
22 |
|
23 |
INPUT_STRUCTION_TEMPLATE = """Here is a gray scale images representing with integer values 0-9.
|
|
|
223 |
top_p=1, n_samples=64, stop=None):
|
224 |
if HUGGINGFACE:
|
225 |
model_inputs = hug_tokenizer([question_prompt], return_tensors="pt").to('cuda')
|
226 |
+
generated_ids = hug_model.generate(**model_inputs, max_length=1400, temperature=1, num_return_sequences=32, do_sample=True)
|
227 |
responses = hug_tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
228 |
codes = []
|
229 |
for response in responses:
|