Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -6,27 +6,21 @@ import llama_cpp
|
|
6 |
from llama_cpp import Llama
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
|
9 |
-
|
10 |
model_path=hf_hub_download(
|
11 |
-
repo_id="
|
12 |
-
filename="
|
|
|
13 |
),
|
14 |
n_ctx=4086,
|
15 |
)
|
16 |
|
17 |
-
dante = Llama(
|
18 |
-
model_path=hf_hub_download(
|
19 |
-
repo_id="FinancialSupport/saiga-7b-gguf",
|
20 |
-
filename="saiga-7b-dante-qlora.Q4_K_M.gguf",
|
21 |
-
),
|
22 |
-
n_ctx=4086,
|
23 |
-
)
|
24 |
|
25 |
history = []
|
26 |
|
27 |
def generate_text(message, history):
|
28 |
temp = ""
|
29 |
-
input_prompt = "
|
30 |
for interaction in history:
|
31 |
input_prompt += "[|Umano|] " + interaction[0] + "\n"
|
32 |
input_prompt += "[|Assistente|]" + interaction[1]
|
@@ -35,7 +29,7 @@ def generate_text(message, history):
|
|
35 |
|
36 |
print(input_prompt)
|
37 |
|
38 |
-
output =
|
39 |
temperature= 0.15,
|
40 |
top_p= 0.1,
|
41 |
top_k= 40,
|
@@ -54,68 +48,24 @@ def generate_text(message, history):
|
|
54 |
|
55 |
history = ["init", input_prompt]
|
56 |
|
57 |
-
def generate_text_Dante(message, history):
|
58 |
-
temp = ""
|
59 |
-
input_prompt = ""
|
60 |
-
for interaction in history:
|
61 |
-
input_prompt += "[|Umano|] " + interaction[0] + "\n"
|
62 |
-
input_prompt += "[|Assistente|]" + interaction[1]
|
63 |
-
|
64 |
-
input_prompt += "[|Umano|] " + message + "\n[|Assistente|]"
|
65 |
-
|
66 |
-
print(input_prompt)
|
67 |
-
|
68 |
-
output = dante(input_prompt,
|
69 |
-
temperature= 0.15,
|
70 |
-
top_p= 0.1,
|
71 |
-
top_k= 40,
|
72 |
-
repeat_penalty= 1.1,
|
73 |
-
max_tokens= 1024,
|
74 |
-
stop= [
|
75 |
-
"[|Umano|]",
|
76 |
-
"[|Assistente|]",
|
77 |
-
],
|
78 |
-
stream= True)
|
79 |
-
|
80 |
-
for out in output:
|
81 |
-
stream = copy.deepcopy(out)
|
82 |
-
temp += stream["choices"][0]["text"]
|
83 |
-
yield temp
|
84 |
-
|
85 |
-
history = ["init", input_prompt]
|
86 |
|
87 |
|
88 |
with gr.Blocks() as demo:
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
with gr.Tab('Dante'):
|
105 |
-
gr.ChatInterface(
|
106 |
-
generate_text_Dante,
|
107 |
-
title="saigaDante-7b running on CPU (quantized Q4_K)",
|
108 |
-
description="This is a quantized version of saiga-7b with Dante LoRA attached running on CPU (very slow).",
|
109 |
-
examples=[
|
110 |
-
"Traduci in volgare fiorentino: tanto va la gatta al lardo che ci lascia lo zampino",
|
111 |
-
"Traduci in volgare fiorentino: narrami come cucinare la pasta alla carbonara vegana.",
|
112 |
-
"Traduci in volgare fiorentino: raccontami una fiaba su Firenze"
|
113 |
-
],
|
114 |
-
cache_examples=False,
|
115 |
-
retry_btn=None,
|
116 |
-
undo_btn="Delete Previous",
|
117 |
-
clear_btn="Clear",
|
118 |
-
)
|
119 |
|
120 |
demo.queue(concurrency_count=1, max_size=5)
|
121 |
demo.launch()
|
|
|
6 |
from llama_cpp import Llama
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
|
9 |
+
zefiro = Llama(
|
10 |
model_path=hf_hub_download(
|
11 |
+
repo_id="giux78/zefiro-7b-beta-ITA-v0.1-GGUF",
|
12 |
+
filename="zefiro-7b-beta-ITA-v0.1-q4_0.gguf
|
13 |
+
",
|
14 |
),
|
15 |
n_ctx=4086,
|
16 |
)
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
history = []
|
20 |
|
21 |
def generate_text(message, history):
|
22 |
temp = ""
|
23 |
+
input_prompt = "Chiedi a zefiro"
|
24 |
for interaction in history:
|
25 |
input_prompt += "[|Umano|] " + interaction[0] + "\n"
|
26 |
input_prompt += "[|Assistente|]" + interaction[1]
|
|
|
29 |
|
30 |
print(input_prompt)
|
31 |
|
32 |
+
output = zefiro(input_prompt,
|
33 |
temperature= 0.15,
|
34 |
top_p= 0.1,
|
35 |
top_k= 40,
|
|
|
48 |
|
49 |
history = ["init", input_prompt]
|
50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
|
53 |
with gr.Blocks() as demo:
|
54 |
+
with gr.Tab('zefiro'):
|
55 |
+
gr.ChatInterface(
|
56 |
+
generate_text,
|
57 |
+
title="zefiro-7b-v01 running on CPU (quantized Q4_K)",
|
58 |
+
description="This is a quantized version of zefiro-7b-v01 running on CPU (very slow). It is less powerful than the original version, but it can even run on the free tier of huggingface.",
|
59 |
+
examples=[
|
60 |
+
"Dammi 3 idee di ricette che posso fare con i pistacchi",
|
61 |
+
"Prepara un piano di esercizi da poter fare a casa",
|
62 |
+
"Scrivi una poesia su una giornato di pioggia"
|
63 |
+
],
|
64 |
+
cache_examples=False,
|
65 |
+
retry_btn=None,
|
66 |
+
undo_btn="Delete Previous",
|
67 |
+
clear_btn="Clear",
|
68 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
demo.queue(concurrency_count=1, max_size=5)
|
71 |
demo.launch()
|