try distant m7b01

app.py

@@ -21,6 +21,13 @@ def run_llama(_, msg, *__):
         message=msg,
         api_name="/chat"
     )
+
+m7b01 = Client("https://zhuraavl-mistralai-mistral-7b-v0-1.hf.space/")
+def run_m7b01(_, msg, *__):
+    yield m7b01.predict(
+        msg,
+        api_name="/predict"
+    )
 inf_models = list(hf_models.keys()) + list(openai_models)
 
 emb_models = ["bge", "minilm"]
@@ -92,6 +99,8 @@ def bot(history, model_name, oepnai_api_key,
 
     if model_name == "llama 3":
         generate_fn = run_llama
+    if model_name == "mistral-7B 0.1":
+        generate_fn = run_m7b01
     elif model_name in hf_models:
          generate_fn = generate_hf
     elif model_name in openai_models:

backend/query_llm.py

@@ -13,8 +13,8 @@ HF_TOKEN = os.getenv("HF_TOKEN")
 hf_models = {
     "mistral-7B": "mistralai/Mistral-7B-Instruct-v0.2",
     "mistral-7B 0.1": "mistralai/Mistral-7B-v0.1",
-    "vicuna-13b":"lmsys/vicuna-13b-v1.5",
-    "WizardLM-30B": "cognitivecomputations/WizardLM-30B-Uncensored",
+    # "vicuna-13b":"lmsys/vicuna-13b-v1.5",
+    # "WizardLM-30B": "cognitivecomputations/WizardLM-30B-Uncensored",
     "llama 3": "meta-llama/Meta-Llama-3-70B-Instruct",
 }
 openai_models = {"gpt-4o","gpt-3.5-turbo-0125"}