Loubna ben allal
commited on
Commit
β’
7cf1a13
1
Parent(s):
cc79f05
update app
Browse files
app.py
CHANGED
@@ -29,49 +29,74 @@ selected_models = st.sidebar.multiselect('Select code generation models to compa
|
|
29 |
models,
|
30 |
default=["CodeParrot"])
|
31 |
st.sidebar.header("Tasks")
|
32 |
-
tasks = [" ","Model architecture", "Model evaluation", "Pretraining dataset", "
|
33 |
selected_task = st.sidebar.selectbox("Select a task:", tasks)
|
34 |
|
35 |
-
|
36 |
-
|
|
|
|
|
|
|
|
|
|
|
37 |
pipelines = {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
if selected_task == " ":
|
40 |
st.title("Code Generation Models comparison π»")
|
41 |
with open("intro.txt", "r") as f:
|
42 |
intro = f.read()
|
43 |
st.markdown(intro)
|
44 |
-
|
45 |
elif selected_task == "Pretraining dataset":
|
46 |
st.title("Pretraining datasets π")
|
47 |
for model in selected_models:
|
48 |
with open(f"datasets/{model.lower()}.txt", "r") as f:
|
49 |
text = f.read()
|
50 |
st.markdown(f"## {model}:")
|
51 |
-
st.markdown(text)
|
52 |
-
|
53 |
elif selected_task == "Model architecture":
|
54 |
st.title("Model architecture π¨")
|
55 |
for model in selected_models:
|
56 |
with open(f"architectures/{model.lower()}.txt", "r") as f:
|
57 |
text = f.read()
|
58 |
st.markdown(f"## {model}:")
|
59 |
-
st.markdown(text)
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
|
|
|
|
|
|
|
29 |
models,
|
30 |
default=["CodeParrot"])
|
31 |
st.sidebar.header("Tasks")
|
32 |
+
tasks = [" ","Model architecture", "Model evaluation", "Pretraining dataset", "Code generation"]
|
33 |
selected_task = st.sidebar.selectbox("Select a task:", tasks)
|
34 |
|
35 |
+
|
36 |
+
tokenizer = load_tokenizer("lvwerra/codeparrot")
|
37 |
+
model = load_model("lvwerra/codeparrot")
|
38 |
+
tokenizer2 = load_tokenizer("facebook/incoder-1B")
|
39 |
+
model2 = load_model("facebook/incoder-1B")
|
40 |
+
tokenizer3 = load_tokenizer("facebook/opt-1.3b")
|
41 |
+
model3 = load_model("facebook/opt-1.3b")
|
42 |
pipelines = {}
|
43 |
+
for model in models:
|
44 |
+
if model == "CodeParrot":
|
45 |
+
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
46 |
+
pipelines[model] = pipe
|
47 |
+
elif model == "InCoder":
|
48 |
+
tokenizer = load_tokenizer("facebook/incoder-1B")
|
49 |
+
model = load_model("facebook/incoder-1B")
|
50 |
+
pipe = pipeline("text-generation", model=model2, tokenizer=tokenizer2)
|
51 |
+
pipelines[model] = pipe
|
52 |
+
else:
|
53 |
+
tokenizer = load_tokenizer("facebook/opt-1.3b")
|
54 |
+
model = load_model("facebook/opt-1.3b")
|
55 |
+
pipe = pipeline("text-generation", model=model3, tokenizer=tokenizer3)
|
56 |
+
pipelines[model] = pipe
|
57 |
+
|
58 |
+
example_names = [example["name"] for example in examples]
|
59 |
+
name2id = dict([(name, i) for i, name in enumerate(example_names)])
|
60 |
+
set_seed(42)
|
61 |
+
gen_kwargs = {}
|
62 |
|
63 |
if selected_task == " ":
|
64 |
st.title("Code Generation Models comparison π»")
|
65 |
with open("intro.txt", "r") as f:
|
66 |
intro = f.read()
|
67 |
st.markdown(intro)
|
|
|
68 |
elif selected_task == "Pretraining dataset":
|
69 |
st.title("Pretraining datasets π")
|
70 |
for model in selected_models:
|
71 |
with open(f"datasets/{model.lower()}.txt", "r") as f:
|
72 |
text = f.read()
|
73 |
st.markdown(f"## {model}:")
|
74 |
+
st.markdown(text)
|
|
|
75 |
elif selected_task == "Model architecture":
|
76 |
st.title("Model architecture π¨")
|
77 |
for model in selected_models:
|
78 |
with open(f"architectures/{model.lower()}.txt", "r") as f:
|
79 |
text = f.read()
|
80 |
st.markdown(f"## {model}:")
|
81 |
+
st.markdown(text)
|
82 |
+
elif selected_task == "Code generation":
|
83 |
+
st.title("Code generation π»")
|
84 |
+
st.sidebar.header("Examples")
|
85 |
+
selected_example = st.sidebar.selectbox("Select one of the following examples:", example_names)
|
86 |
+
example_text = examples[name2id[selected_example]]["value"]
|
87 |
+
default_length = examples[name2id[selected_example]]["length"]
|
88 |
+
st.sidebar.header("Generation settings")
|
89 |
+
gen_kwargs["do_sample"] = st.sidebar.radio("Decoding strategy:", ["Greedy", "Sample"]) == "Sample"
|
90 |
+
gen_kwargs["max_new_tokens"] = st.sidebar.slider("Number of tokens to generate:", value=default_length, min_value=8, step=8, max_value=256)
|
91 |
+
if gen_kwargs["do_sample"]:
|
92 |
+
gen_kwargs["temperature"] = 0.2
|
93 |
+
gen_kwargs["top_k"] = 0
|
94 |
+
gen_kwargs["top_p"] = 0.95
|
95 |
+
gen_prompt = st.text_area("Generate code with prompt:", value=example_text, height=220,).strip()
|
96 |
+
if st.button("Generate code!"):
|
97 |
+
with st.spinner("Generating code..."):
|
98 |
+
for model in selected_models:
|
99 |
+
pipe = pipelines[model]
|
100 |
+
generated_text = pipe(gen_prompt, **gen_kwargs)[0]['generated_text']
|
101 |
+
st.markdown(f"### {model}:")
|
102 |
+
st.code(generated_text)
|