Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,34 +3,38 @@ from huggingface_hub import hf_hub_download
|
|
| 3 |
from llama_cpp import Llama
|
| 4 |
import os
|
| 5 |
|
| 6 |
-
# 1.
|
| 7 |
-
|
| 8 |
-
|
|
|
|
| 9 |
|
| 10 |
-
print("⏬ Scaricamento del modello Limba...")
|
|
|
|
|
|
|
| 11 |
model_path = hf_hub_download(
|
| 12 |
repo_id="FPll/Limba-1.0",
|
| 13 |
filename="Meta-Llama-3.1-8B.Q4_K_M.gguf",
|
| 14 |
token=token
|
| 15 |
)
|
| 16 |
|
| 17 |
-
|
|
|
|
|
|
|
| 18 |
llm = Llama(model_path=model_path, n_ctx=2048, n_threads=4)
|
| 19 |
|
| 20 |
def respond(message, history):
|
| 21 |
-
#
|
| 22 |
prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{message}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
|
| 23 |
|
| 24 |
response = llm(prompt, max_tokens=512, stop=["<|eot_id|>"], echo=False)
|
| 25 |
return response["choices"][0]["text"].strip()
|
| 26 |
|
| 27 |
-
# 3. Interfaccia
|
| 28 |
demo = gr.ChatInterface(
|
| 29 |
respond,
|
| 30 |
title="Limba 1.0 - Progetto Margherita",
|
| 31 |
description="Su primu protòtipu de IA pro s'iscola sarda. Parla-mi in sardu o de fìsica e matemàtica!",
|
| 32 |
-
examples=["Iscrie sa fòrmula de s'energia tzinetica", "Chie est istadu Enricu Fermi?", "A ite zerbit sa derivata?"]
|
| 33 |
-
theme="soft"
|
| 34 |
)
|
| 35 |
|
| 36 |
if __name__ == "__main__":
|
|
|
|
| 3 |
from llama_cpp import Llama
|
| 4 |
import os
|
| 5 |
|
| 6 |
+
# 1. Recupero e pulizia del Token
|
| 7 |
+
token = os.getenv("HF_TOKEN")
|
| 8 |
+
if token:
|
| 9 |
+
token = token.strip()
|
| 10 |
|
| 11 |
+
print(f"⏬ Scaricamento del modello Limba...")
|
| 12 |
+
|
| 13 |
+
# Scarichiamo il modello
|
| 14 |
model_path = hf_hub_download(
|
| 15 |
repo_id="FPll/Limba-1.0",
|
| 16 |
filename="Meta-Llama-3.1-8B.Q4_K_M.gguf",
|
| 17 |
token=token
|
| 18 |
)
|
| 19 |
|
| 20 |
+
print("✅ Modello caricato! Accendo i motori...")
|
| 21 |
+
|
| 22 |
+
# 2. Inizializzazione del modello (Ottimizzato per 4 thread)
|
| 23 |
llm = Llama(model_path=model_path, n_ctx=2048, n_threads=4)
|
| 24 |
|
| 25 |
def respond(message, history):
|
| 26 |
+
# Formato prompt per Llama 3.1
|
| 27 |
prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{message}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
|
| 28 |
|
| 29 |
response = llm(prompt, max_tokens=512, stop=["<|eot_id|>"], echo=False)
|
| 30 |
return response["choices"][0]["text"].strip()
|
| 31 |
|
| 32 |
+
# 3. Interfaccia Chat (Ho rimosso 'theme' che dava errore)
|
| 33 |
demo = gr.ChatInterface(
|
| 34 |
respond,
|
| 35 |
title="Limba 1.0 - Progetto Margherita",
|
| 36 |
description="Su primu protòtipu de IA pro s'iscola sarda. Parla-mi in sardu o de fìsica e matemàtica!",
|
| 37 |
+
examples=["Iscrie sa fòrmula de s'energia tzinetica", "Chie est istadu Enricu Fermi?", "A ite zerbit sa derivata?"]
|
|
|
|
| 38 |
)
|
| 39 |
|
| 40 |
if __name__ == "__main__":
|