:wrench: drop reflection. add Nemotron. make default model.

README.md

@@ -1977,7 +1977,7 @@ models:
   - ziniuli/Mistral-7B-ReMax-v0.1
   - zmzmxz/NeuralPipe-7B-slerp
   - Qwen/Qwen2.5-72B
-  - mattshumer/Reflection-Llama-3.1-70B
+  - nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
 ---
 
 # Overview

app.py

@@ -40,12 +40,14 @@ model_class_filter = {
 # we run a few other models here as well
 REFLECTION="mattshumer/Reflection-Llama-3.1-70B"
 QWEN25_72B="Qwen/Qwen2.5-72B"
+NEMOTRON="nvidia/Llama-3.1-Nemotron-70B-Instruct-HF"
 bigger_whitelisted_models = [
-    REFLECTION,
-    QWEN25_72B
+    QWEN25_72B,
+    NEMOTRON
 ]
 # REFLECTION is in backup hosting
 model_class_from_model_id[REFLECTION] = 'llama31-70b-16k'
+model_class_from_model_id[NEMOTRON] = 'llama31-70b-16k'
 def build_model_choices():
     all_choices = []
     for model_class in model_cache:
@@ -72,6 +74,7 @@ def model_in_list(model):
 key=os.environ.get('RANDOM_SEED', 'kcOtfNHA+e')
 o = random.Random(f"{key}-{datetime.date.today().strftime('%Y-%m-%d')}")
 initial_model = o.choice(model_choices)[1]
+initial_model = NEMOTRON
 # this doesn't work in HF spaces because we're iframed :(
 # def initial_model(referer=None):
 #     return REFLECTION
@@ -105,7 +108,7 @@ def respond(message, history, model):
         history_openai_format.append({"role": "assistant", "content":assistant})
     history_openai_format.append({"role": "user", "content": message})
 
-    if model == "mattshumer/Reflection-Llama-3.1-70B":
+    if model == REFLECTION:
         history_openai_format = [
             {"role": "system", "content": REFLECTION_SYSTEM_PROMPT},
             *history_openai_format

readme-generator/generate.sh

@@ -20,7 +20,7 @@ $(cat ../model-cache.json \
     console.log(yamlStr);"
 )
   - Qwen/Qwen2.5-72B
-  - mattshumer/Reflection-Llama-3.1-70B
+  - nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
 ---
 
 $(cat body.md)