Add DeepSeek-R1 8B (Llama)

app.py

@@ -10,6 +10,7 @@ from huggingface_hub import InferenceClient
 
 HF_MODEL_MISTRAL = "mistralai/Mistral-7B-Instruct-v0.3"
 HF_MODEL_LLAMA = "meta-llama/Llama-3.3-70B-Instruct"
+HF_MODEL_DEEPSEEK = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
 
 UML_PROMPTS_DOC_URL = os.environ['UML_PROMPTS_DOC_URL']
 ERD_PROMPTS_DOC_URL = os.environ['ERD_PROMPTS_DOC_URL']
@@ -77,7 +78,7 @@ def extract_plantuml_code(client_openai, uploaded_file, model_choice, prompts):
 
 
 # Step 2: Compare PlantUML Code
-def compare_plantuml(client_openai, client_hf_mistral, client_hf_llama, plantuml_instructor, plantuml_student, model_choice, prompts, diagram_type="UML"):
+def compare_plantuml(client_openai, client_hf_mistral, client_hf_llama, client_hf_deepseek, plantuml_instructor, plantuml_student, model_choice, prompts, diagram_type="UML"):
 
     st.write("Model: ", model_choice)
 
@@ -118,6 +119,20 @@ def compare_plantuml(client_openai, client_hf_mistral, client_hf_llama, plantuml
             temperature=0.2,
         )
         return response["choices"][0]["message"]["content"]
+
+    elif model_choice in [HF_MODEL_DEEPSEEK]:
+        response = client_hf_deepseek.chat_completion(
+            [
+                {
+                    "role": "system",
+                    "content": prompts[STEP2_SYSTEM_PROMPT],
+                },
+                {"role": "user", "content": user_prompt},
+            ],
+            max_tokens=1024,
+            temperature=0.2,
+        )
+        return response["choices"][0]["message"]["content"]
     
     else:
         response = client_openai.chat.completions.create(
@@ -139,7 +154,7 @@ def compare_plantuml(client_openai, client_hf_mistral, client_hf_llama, plantuml
         return response.choices[0].message.content
 
 # Step 3A: Generate Student Feedback
-def generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama, differences, model_choice, prompts):
+def generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama, client_hf_deepseek, differences, model_choice, prompts):
 
     st.write("Model (Student Feedback):", model_choice)  
 
@@ -177,6 +192,20 @@ def generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama,
         )
 
         return response["choices"][0]["message"]["content"]
+
+    elif model_choice in [HF_MODEL_DEEPSEEK]:
+        response = client_hf_deepseek.chat_completion(
+            [
+                {
+                    "role": "system",
+                    "content": prompts[STEP2_SYSTEM_PROMPT],
+                },
+                {"role": "user", "content": user_prompt},
+            ],
+            max_tokens=1024,
+            temperature=0.2,
+        )
+        return response["choices"][0]["message"]["content"]
     
     else:
         response = client_openai.chat.completions.create(
@@ -200,7 +229,7 @@ def generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama,
 
 
 # Step 3B: Generate Educator Feedback
-def generate_educator_feedback(client_openai, client_hf_mistral, client_hf_llama, differences, model_choice, prompts):
+def generate_educator_feedback(client_openai, client_hf_mistral, client_hf_llama, client_hf_deepseek, differences, model_choice, prompts):
   
     st.write("Model (Educator Feedback): ", model_choice)
 
@@ -238,6 +267,20 @@ def generate_educator_feedback(client_openai, client_hf_mistral, client_hf_llama
         )
 
         return response["choices"][0]["message"]["content"]
+
+    elif model_choice in [HF_MODEL_DEEPSEEK]:
+        response = client_hf_deepseek.chat_completion(
+            [
+                {
+                    "role": "system",
+                    "content": prompts[STEP2_SYSTEM_PROMPT],
+                },
+                {"role": "user", "content": user_prompt},
+            ],
+            max_tokens=1024,
+            temperature=0.2,
+        )
+        return response["choices"][0]["message"]["content"]
     
     else:
         response = client_openai.chat.completions.create(
@@ -283,10 +326,11 @@ if openai_api_key and hf_api_key:
     client_openai = OpenAI(api_key=openai_api_key)
     client_hf_mistral = InferenceClient(model=HF_MODEL_MISTRAL, token=hf_api_key)
     client_hf_llama = InferenceClient(model=HF_MODEL_LLAMA, token=hf_api_key)
+    client_hf_deepseek = InferenceClient(model=HF_MODEL_DEEPSEEK, token=hf_api_key)
     
     model_choice_step1 = st.selectbox("Select the model for Step 1", ["gpt-4o", "gpt-4o-mini"])
-    model_choice_step2 = st.selectbox("Select the model for Step 2", [HF_MODEL_MISTRAL, HF_MODEL_LLAMA, "gpt-4o", "gpt-4o-mini"])
-    model_choice_step3 = st.selectbox("Select the model for Step 3", [HF_MODEL_MISTRAL, HF_MODEL_LLAMA, "gpt-4o", "gpt-4o-mini"])
+    model_choice_step2 = st.selectbox("Select the model for Step 2", [HF_MODEL_MISTRAL, HF_MODEL_LLAMA, HF_MODEL_DEEPSEEK, "gpt-4o", "gpt-4o-mini"])
+    model_choice_step3 = st.selectbox("Select the model for Step 3", [HF_MODEL_MISTRAL, HF_MODEL_LLAMA, HF_MODEL_DEEPSEEK, "gpt-4o", "gpt-4o-mini"])
 
     st.subheader("Step 1: PlantUML Code Extraction using GPT-4o or GPT-4o Mini")
 
@@ -348,6 +392,7 @@ if openai_api_key and hf_api_key:
                     client_openai,
                     client_hf_mistral,
                     client_hf_llama,
+                    client_hf_deepseek
                     plantuml_instructor_solution,
                     plantuml_student_solution,
                     model_choice_step2,
@@ -360,8 +405,8 @@ if openai_api_key and hf_api_key:
 
             st.subheader("Step 3: Structured Feedback")
             with st.spinner("Preparing structured feedback..."):
-                student_feedback = generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama, differences, model_choice_step3, prompts)
-                educator_feedback = generate_educator_feedback(client_openai, client_hf_mistral, client_hf_llama, differences, model_choice_step3, prompts)
+                student_feedback = generate_student_feedback(client_openai, client_hf_mistral, client_hf_llama, client_hf_deepseek, differences, model_choice_step3, prompts)
+                educator_feedback = generate_educator_feedback(client_openai, client_hf_mistral, client_hf_llama, client_hf_deepseek, differences, model_choice_step3, prompts)
 
                 col1, col2 = st.columns(2)
                 with col1: