Update README.md
Browse files
README.md
CHANGED
@@ -9,16 +9,15 @@ TBD
|
|
9 |
https://github.com/vllm-project/llm-compressor/pull/185
|
10 |
|
11 |
```python
|
12 |
-
from transformers import AutoProcessor
|
13 |
|
14 |
from llmcompressor.modifiers.quantization import QuantizationModifier
|
15 |
-
from llmcompressor.transformers import oneshot
|
16 |
-
from llmcompressor.transformers.sparsification import create_sparse_auto_model_class
|
17 |
|
18 |
MODEL_ID = "meta-llama/Llama-3.2-11B-Vision-Instruct"
|
19 |
|
20 |
# Load model.
|
21 |
-
model_class =
|
22 |
model = model_class.from_pretrained(MODEL_ID, device_map="auto", torch_dtype="auto")
|
23 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
24 |
|
@@ -35,6 +34,7 @@ recipe = QuantizationModifier(
|
|
35 |
# Apply quantization and save to disk in compressed-tensors format.
|
36 |
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
|
37 |
oneshot(model=model, recipe=recipe, output_dir=SAVE_DIR)
|
|
|
38 |
|
39 |
# Confirm generations of the quantized model look sane.
|
40 |
print("========== SAMPLE GENERATION ==============")
|
|
|
9 |
https://github.com/vllm-project/llm-compressor/pull/185
|
10 |
|
11 |
```python
|
12 |
+
from transformers import AutoProcessor, MllamaForConditionalGeneration
|
13 |
|
14 |
from llmcompressor.modifiers.quantization import QuantizationModifier
|
15 |
+
from llmcompressor.transformers import oneshot, wrap_hf_model_class
|
|
|
16 |
|
17 |
MODEL_ID = "meta-llama/Llama-3.2-11B-Vision-Instruct"
|
18 |
|
19 |
# Load model.
|
20 |
+
model_class = wrap_hf_model_class(MllamaForConditionalGeneration)
|
21 |
model = model_class.from_pretrained(MODEL_ID, device_map="auto", torch_dtype="auto")
|
22 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
23 |
|
|
|
34 |
# Apply quantization and save to disk in compressed-tensors format.
|
35 |
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
|
36 |
oneshot(model=model, recipe=recipe, output_dir=SAVE_DIR)
|
37 |
+
processor.save_pretrained(SAVE_DIR)
|
38 |
|
39 |
# Confirm generations of the quantized model look sane.
|
40 |
print("========== SAMPLE GENERATION ==============")
|