app.py
CHANGED
@@ -40,7 +40,7 @@ index.add(embeddings)
|
|
40 |
|
41 |
model_name = "TheBloke/zephyr-7B-beta-GPTQ"
|
42 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
|
43 |
-
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="balanced", trust_remote_code=False
|
44 |
|
45 |
|
46 |
# Step 4: Define the Retrieval Function
|
|
|
40 |
|
41 |
model_name = "TheBloke/zephyr-7B-beta-GPTQ"
|
42 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
|
43 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="balanced", trust_remote_code=False)
|
44 |
|
45 |
|
46 |
# Step 4: Define the Retrieval Function
|