Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -79,18 +79,28 @@ def load_pipeline(model_name):
|
|
79 |
Load and cache a transformers pipeline for text generation.
|
80 |
Tries bfloat16, falls back to float16 or float32 if unsupported.
|
81 |
"""
|
82 |
-
|
83 |
if model_name in PIPELINES.keys():
|
84 |
return PIPELINES[model_name]
|
85 |
repo = MODELS[model_name]["repo_id"]
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
PIPELINES[model_name] = {"tokenizer": tokenizer, "model": model}
|
93 |
-
return
|
94 |
|
95 |
|
96 |
def retrieve_context(query, max_results=6, max_chars=600):
|
|
|
79 |
Load and cache a transformers pipeline for text generation.
|
80 |
Tries bfloat16, falls back to float16 or float32 if unsupported.
|
81 |
"""
|
82 |
+
|
83 |
if model_name in PIPELINES.keys():
|
84 |
return PIPELINES[model_name]
|
85 |
repo = MODELS[model_name]["repo_id"]
|
86 |
+
if model_name == "secgpt-mini":
|
87 |
+
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True, subfolder="models")
|
88 |
+
model = AutoModelForCausalLM.from_pretrained(
|
89 |
+
repo,
|
90 |
+
device_map=device,
|
91 |
+
trust_remote_code=True,
|
92 |
+
subfolder="models",
|
93 |
+
)
|
94 |
+
else:
|
95 |
+
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)
|
96 |
+
model = AutoModelForCausalLM.from_pretrained(
|
97 |
+
repo,
|
98 |
+
device_map=device,
|
99 |
+
trust_remote_code=True,
|
100 |
+
)
|
101 |
+
global PIPELINES
|
102 |
PIPELINES[model_name] = {"tokenizer": tokenizer, "model": model}
|
103 |
+
return {"tokenizer": tokenizer, "model": model}
|
104 |
|
105 |
|
106 |
def retrieve_context(query, max_results=6, max_chars=600):
|