Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -44,14 +44,20 @@ if not torch.cuda.is_available():
|
|
44 |
model = None
|
45 |
tokenizer = None
|
46 |
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
tokenizer
|
54 |
-
tokenizer.
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
|
56 |
@spaces.GPU
|
57 |
def generate(
|
@@ -63,36 +69,43 @@ def generate(
|
|
63 |
top_k: int = 50,
|
64 |
repetition_penalty: float = 1.4,
|
65 |
) -> Iterator[str]:
|
66 |
-
global model, tokenizer
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
|
97 |
chat_interface = gr.ChatInterface(
|
98 |
fn=generate,
|
|
|
44 |
model = None
|
45 |
tokenizer = None
|
46 |
|
47 |
+
def initialize_model_and_tokenizer():
|
48 |
+
global model, tokenizer
|
49 |
+
if torch.cuda.is_available():
|
50 |
+
model_id = "apple/OpenELM-3B-Instruct"
|
51 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True, low_cpu_mem_usage=True)
|
52 |
+
tokenizer_id = "meta-llama/Llama-2-7b-hf"
|
53 |
+
tokenizer = AutoTokenizer.from_pretrained(tokenizer_id)
|
54 |
+
if tokenizer.pad_token is None:
|
55 |
+
tokenizer.pad_token = tokenizer.eos_token
|
56 |
+
tokenizer.pad_token_id = tokenizer.eos_token_id
|
57 |
+
else:
|
58 |
+
print("CUDA is not available. Model and tokenizer will not be initialized.")
|
59 |
+
|
60 |
+
initialize_model_and_tokenizer()
|
61 |
|
62 |
@spaces.GPU
|
63 |
def generate(
|
|
|
69 |
top_k: int = 50,
|
70 |
repetition_penalty: float = 1.4,
|
71 |
) -> Iterator[str]:
|
72 |
+
global model, tokenizer
|
73 |
+
|
74 |
+
if tokenizer is None or model is None:
|
75 |
+
yield "Error: Model or tokenizer not initialized. Make sure you have GPU support and the necessary model access."
|
76 |
+
return
|
77 |
+
|
78 |
+
try:
|
79 |
+
input_ids = tokenizer([message], return_tensors="pt").input_ids
|
80 |
+
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
81 |
+
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
82 |
+
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
83 |
+
input_ids = input_ids.to(model.device)
|
84 |
+
|
85 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
86 |
+
generate_kwargs = dict(
|
87 |
+
input_ids=input_ids,
|
88 |
+
streamer=streamer,
|
89 |
+
max_new_tokens=max_new_tokens,
|
90 |
+
do_sample=True,
|
91 |
+
top_p=top_p,
|
92 |
+
top_k=top_k,
|
93 |
+
temperature=temperature,
|
94 |
+
num_beams=1,
|
95 |
+
pad_token_id=tokenizer.eos_token_id,
|
96 |
+
repetition_penalty=repetition_penalty,
|
97 |
+
no_repeat_ngram_size=5,
|
98 |
+
early_stopping=True,
|
99 |
+
)
|
100 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
101 |
+
t.start()
|
102 |
+
|
103 |
+
outputs = []
|
104 |
+
for text in streamer:
|
105 |
+
outputs.append(text)
|
106 |
+
yield "".join(outputs)
|
107 |
+
except Exception as e:
|
108 |
+
yield f"An error occurred: {str(e)}"
|
109 |
|
110 |
chat_interface = gr.ChatInterface(
|
111 |
fn=generate,
|