Spaces:
Running
Running
Jeff Myers II
commited on
Commit
·
297702e
1
Parent(s):
5278642
Using quantization_config instead of load_in_8bit
Browse files
Gemma.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
|
|
|
2 |
from huggingface_hub import login
|
3 |
import spaces
|
4 |
import torch
|
@@ -11,8 +12,14 @@ class GemmaLLM:
|
|
11 |
def __init__(self):
|
12 |
login(token=os.environ.get("GEMMA_TOKEN"))
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
model_id = "google/gemma-3-4b-it"
|
15 |
-
model = AutoModelForCausalLM.from_pretrained(model_id,
|
16 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
17 |
|
18 |
self.model = pipeline("text-generation", model=model, tokenizer=tokenizer, torch_dtype=torch.bfloat16)
|
|
|
1 |
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
|
2 |
+
from transformers.utils import quantization_config
|
3 |
from huggingface_hub import login
|
4 |
import spaces
|
5 |
import torch
|
|
|
12 |
def __init__(self):
|
13 |
login(token=os.environ.get("GEMMA_TOKEN"))
|
14 |
|
15 |
+
quant_config = quantization_config.BitsAndBytesConfig(
|
16 |
+
load_in_8bit=True,
|
17 |
+
llm_int8_threshold=6.0,
|
18 |
+
llm_int8_has_fp16_weight=False,
|
19 |
+
)
|
20 |
+
|
21 |
model_id = "google/gemma-3-4b-it"
|
22 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=quant_config)
|
23 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
24 |
|
25 |
self.model = pipeline("text-generation", model=model, tokenizer=tokenizer, torch_dtype=torch.bfloat16)
|