:sparkles: adding meta-llama/Llama-3.3-70B-Instruct

README.md

@@ -2034,6 +2034,7 @@ models:
   - Qwen/Qwen2.5-72B
   - nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
   - Qwen/QwQ-32B-Preview
+  - meta-llama/Llama-3.3-70B-Instruct
 ---
 
 # Overview

app.py

@@ -41,13 +41,16 @@ model_class_filter = {
 REFLECTION="mattshumer/Reflection-Llama-3.1-70B"
 QWEN25_72B="Qwen/Qwen2.5-72B"
 NEMOTRON="nvidia/Llama-3.1-Nemotron-70B-Instruct-HF"
+LLAMA3="meta-llama/Llama-3.3-70B-Instruct"
 bigger_whitelisted_models = [
     QWEN25_72B,
-    NEMOTRON
+    NEMOTRON,
+    LLAMA3
 ]
 # REFLECTION is in backup hosting
 model_class_from_model_id[REFLECTION] = 'llama31-70b-16k'
 model_class_from_model_id[NEMOTRON] = 'llama31-70b-16k'
+model_class_from_model_id[LLAMA3] = 'llama31-70b-16k'
 def build_model_choices():
     all_choices = []
     for model_class in model_cache:
@@ -74,7 +77,7 @@ 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
+initial_model = LLAMA3
 # this doesn't work in HF spaces because we're iframed :(
 # def initial_model(referer=None):
 #     return REFLECTION

readme-generator/generate.sh

@@ -21,6 +21,8 @@ $(cat ../model-cache.json \
 )
   - Qwen/Qwen2.5-72B
   - nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
+  - Qwen/QwQ-32B-Preview
+  - meta-llama/Llama-3.3-70B-Instruct
 ---
 
 $(cat body.md)