Spaces:
Paused
Paused
remove flash
Browse files- app.py +6 -10
- requirements.txt +5 -5
app.py
CHANGED
|
@@ -13,14 +13,11 @@ model_path = "microsoft/Phi-4-multimodal-instruct"
|
|
| 13 |
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
|
| 14 |
model = AutoModelForCausalLM.from_pretrained(
|
| 15 |
model_path,
|
| 16 |
-
device_map="
|
| 17 |
torch_dtype="auto",
|
| 18 |
trust_remote_code=True,
|
| 19 |
-
|
| 20 |
-
)
|
| 21 |
-
|
| 22 |
-
# Load generation config
|
| 23 |
-
generation_config = GenerationConfig.from_pretrained(model_path)
|
| 24 |
|
| 25 |
# Define prompt structure
|
| 26 |
user_prompt = '<|user|>'
|
|
@@ -38,12 +35,12 @@ def process_input(input_type, file, question):
|
|
| 38 |
prompt = f'{user_prompt}<|image_1|>{question}{prompt_suffix}{assistant_prompt}'
|
| 39 |
# Open image from uploaded file
|
| 40 |
image = Image.open(file)
|
| 41 |
-
inputs = processor(text=prompt, images=image, return_tensors='pt').to(
|
| 42 |
elif input_type == "Audio":
|
| 43 |
prompt = f'{user_prompt}<|audio_1|>{question}{prompt_suffix}{assistant_prompt}'
|
| 44 |
# Read audio from uploaded file
|
| 45 |
audio, samplerate = sf.read(file)
|
| 46 |
-
inputs = processor(text=prompt, audios=[(audio, samplerate)], return_tensors='pt').to(
|
| 47 |
else:
|
| 48 |
return "Invalid input type selected."
|
| 49 |
|
|
@@ -51,8 +48,7 @@ def process_input(input_type, file, question):
|
|
| 51 |
with torch.no_grad():
|
| 52 |
generate_ids = model.generate(
|
| 53 |
**inputs,
|
| 54 |
-
max_new_tokens=
|
| 55 |
-
generation_config=generation_config,
|
| 56 |
num_logits_to_keep=0,
|
| 57 |
)
|
| 58 |
generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
|
|
|
|
| 13 |
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
|
| 14 |
model = AutoModelForCausalLM.from_pretrained(
|
| 15 |
model_path,
|
| 16 |
+
device_map="auto",
|
| 17 |
torch_dtype="auto",
|
| 18 |
trust_remote_code=True,
|
| 19 |
+
_attn_implementation="eager",
|
| 20 |
+
)
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
# Define prompt structure
|
| 23 |
user_prompt = '<|user|>'
|
|
|
|
| 35 |
prompt = f'{user_prompt}<|image_1|>{question}{prompt_suffix}{assistant_prompt}'
|
| 36 |
# Open image from uploaded file
|
| 37 |
image = Image.open(file)
|
| 38 |
+
inputs = processor(text=prompt, images=image, return_tensors='pt').to(model.device)
|
| 39 |
elif input_type == "Audio":
|
| 40 |
prompt = f'{user_prompt}<|audio_1|>{question}{prompt_suffix}{assistant_prompt}'
|
| 41 |
# Read audio from uploaded file
|
| 42 |
audio, samplerate = sf.read(file)
|
| 43 |
+
inputs = processor(text=prompt, audios=[(audio, samplerate)], return_tensors='pt').to(model.device)
|
| 44 |
else:
|
| 45 |
return "Invalid input type selected."
|
| 46 |
|
|
|
|
| 48 |
with torch.no_grad():
|
| 49 |
generate_ids = model.generate(
|
| 50 |
**inputs,
|
| 51 |
+
max_new_tokens=200,
|
|
|
|
| 52 |
num_logits_to_keep=0,
|
| 53 |
)
|
| 54 |
generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
|
requirements.txt
CHANGED
|
@@ -1,11 +1,11 @@
|
|
| 1 |
gradio
|
| 2 |
spaces
|
|
|
|
| 3 |
torch
|
| 4 |
-
peft
|
| 5 |
-
torchvision
|
| 6 |
-
scipy
|
| 7 |
-
soundfile
|
| 8 |
pillow
|
| 9 |
-
|
| 10 |
transformers
|
|
|
|
|
|
|
|
|
|
| 11 |
backoff
|
|
|
|
| 1 |
gradio
|
| 2 |
spaces
|
| 3 |
+
requests
|
| 4 |
torch
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
pillow
|
| 6 |
+
soundfile
|
| 7 |
transformers
|
| 8 |
+
torchvision
|
| 9 |
+
scipy
|
| 10 |
+
peft
|
| 11 |
backoff
|