update: added model/provider in expert mode + fixed visuals

app.py

@@ -55,7 +55,7 @@ with tab_method:
 
 with tab_about:
 
-    st.write(ABOUT_TEXT)
+    st.markdown(ABOUT_TEXT, unsafe_allow_html=True)
 
 with st.expander('📚 Citation'):
     st.html(CITATION_LABEL)

src/calculator.py

@@ -1,12 +1,10 @@
 import streamlit as st
-import pandas as pd
 
 from ecologits.tracers.utils import llm_impacts
 from src.impacts import get_impacts, display_impacts, display_equivalent
 from src.utils import format_impacts
 from src.content import WARNING_CLOSED_SOURCE, WARNING_MULTI_MODAL, WARNING_BOTH
-from src.models import load_models, clean_models_data
-from src.constants import MAIN_MODELS
+from src.models import load_models
 
 from src.constants import PROMPTS
 

src/content.py

@@ -4,13 +4,18 @@ HERO_TEXT = """
     <img style="max-height: 80px" alt="EcoLogits" src="https://raw.githubusercontent.com/genai-impact/ecologits/main/docs/assets/logo_light.png">
   </a>
 </div>
-<h1 align="center">🧮 EcoLogits Calculator</h1>
 <div align="center">
   <p style="max-width: 500px; text-align: center">
     <i><b>EcoLogits</b> is a python library that tracks the <b>energy consumption</b> and <b>environmental 
     footprint</b> of using <b>generative AI</b> models through APIs.</i>
   </p>
 </div>
+<h1 align="center">🧮 EcoLogits Calculator</h1>
+<div align="center">
+  <p style="max-width: 500px; text-align: center">
+    <i>allows a broader access to <b>EcoLogits</b> through a visual application.</i>
+  </p>
+</div>
 <br>
 """
 
@@ -111,12 +116,14 @@ EcoLogits is focused on estimating the environmental impacts of generative AI (o
 providers (such as OpenAI, Anthropic, Cloud APIs...)** whereas  CodeCarbon is more general tool to measure energy 
 consumption and estimate GHG emissions measurement. If you deploy LLMs locally we encourage you to use CodeCarbon to 
 get real numbers of your energy consumption.
+
 ## 🤗 Contributing 
 We are eager to get feedback from the community, don't hesitate to engage the discussion with us on this 
 [GitHub thread](https://github.com/genai-impact/ecologits/discussions/45) or message us on 
 [LinkedIn](https://www.linkedin.com/company/genai-impact/).
 We also welcome any open-source contributions on 🌱 **[EcoLogits](https://github.com/genai-impact/ecologits)** or on 
 🧮 **EcoLogits Calculator**.
+
 ## ⚖️ License
 <p xmlns:cc="http://creativecommons.org/ns#" >
   This work is licensed under 
@@ -127,12 +134,14 @@ We also welcome any open-source contributions on 🌱 **[EcoLogits](https://gith
   <img style="display:inline-block;height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/by.svg?ref=chooser-v1" alt="">
   <img style="display:inline-block;height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/sa.svg?ref=chooser-v1" alt="">
 </p>
+
 ## 🙌 Acknowledgement
 We thank [Data For Good](https://dataforgood.fr/) and [Boavizta](https://boavizta.org/en) for supporting the 
 development of this project. Their contributions of tools, best practices, and expertise in environmental impact 
 assessment have been invaluable.
 We also extend our gratitude to the open-source contributions of 🤗 [Hugging Face](huggingface.com) on the LLM-Perf 
 Leaderboard.
+
 ## 🤝 Contact
 For general question on the project, please use the [GitHub thread](https://github.com/genai-impact/ecologits/discussions/45). 
 Otherwise use our contact form on [genai-impact.org/contact](https://genai-impact.org/contact/).

src/expert.py

@@ -1,12 +1,11 @@
 import streamlit as st
-import pandas as pd
 from ecologits.impacts.llm import compute_llm_impacts
 
-from src.utils import format_impacts, average_range_impacts, format_impacts_expert, model_active_params_fn, model_total_params_fn
+from src.utils import format_impacts, average_range_impacts
 from src.impacts import display_impacts
-#from src.constants import PROVIDERS, MODELS
 from src.electricity_mix import COUNTRY_CODES, find_electricity_mix, dataframe_electricity_mix
-from ecologits.model_repository import models
+from src.models import load_models
+from src.constants import PROMPTS
 
 import plotly.express as px
 
@@ -21,35 +20,56 @@ def expert_mode():
 
         ########## Model info ##########
 
-        # col1, col2, col3 = st.columns(3)
+        col1, col2, col3 = st.columns(3)
+        
+        df = load_models(filter_main=True)
+
+        with col1:
+            provider_exp = st.selectbox(
+                label = 'Provider',
+                options = [x for x in df['provider_clean'].unique()],
+                index = 7,
+                key = 1
+            )
 
-        # with col1:
-        #     provider = st.selectbox(label = 'Provider expert',
-        #                             options = [x[0] for x in PROVIDERS],
-        #                             index = 0)
-        #     provider = [x[1] for x in PROVIDERS if x[0] == provider][0]
-        #     if 'huggingface_hub' in provider:
-        #         provider = 'huggingface_hub'
+        with col2:
+            model_exp = st.selectbox(
+                label = 'Model',
+                options = [x for x in df['name_clean'].unique() if x in df[df['provider_clean'] == provider_exp]['name_clean'].unique()],
+                key = 2
+            )
+
+        with col3:
+            output_tokens_exp = st.selectbox(
+                label = 'Example prompt',
+                options = [x[0] for x in PROMPTS],
+                key = 3
+            )
         
-        # with col2:
-        #     model = st.selectbox('Model expert', [x[0] for x in MODELS if provider in x[1]])
-        #     model = [x[1] for x in MODELS if x[0] == model][0].split('/', 1)[1]
+        df_filtered = df[(df['provider_clean'] == provider_exp) & (df['name_clean'] == model_exp)]
+
+        try:
+            total_params = int(df_filtered['total_parameters'].iloc[0])
+        except:
+            total_params = int((df_filtered['total_parameters'].values[0]['min'] + df_filtered['total_parameters'].values[0]['max'])/2)
+            
+        try:
+            active_params = int(df_filtered['active_parameters'].iloc[0])
+        except:
+            active_params = int((df_filtered['active_parameters'].values[0]['min'] + df_filtered['active_parameters'].values[0]['max'])/2)
 
         ########## Model parameters ##########   
 
         col11, col22, col33 = st.columns(3)
 
         with col11:
-            # st.write(provider, model)
-            # st.write(models.find_model(provider, model))
-            # st.write(model_active_params_fn(provider, model, 45))
-            active_params = st.number_input('Active parameters (B)', 0, None, 45)
+            active_params = st.number_input('Active parameters (B)', 0, None, active_params)
 
         with col22:
-            total_params = st.number_input('Total parameters (B)', 0, None, 45)
+            total_params = st.number_input('Total parameters (B)', 0, None, total_params)
 
         with col33:
-            output_tokens = st.number_input('Output completion tokens', 100)
+            output_tokens = st.number_input('Output completion tokens', [x[1] for x in PROMPTS if x[0] == output_tokens_exp][0])
 
         ########## Electricity mix ##########
 
@@ -89,15 +109,16 @@ def expert_mode():
         st.markdown('The usage impacts account for the electricity consumption of the model while the embodied impacts account for resource extraction (e.g., minerals and metals), manufacturing, and transportation of the hardware.')
         
         col_ghg_comparison, col_adpe_comparison, col_pe_comparison = st.columns(3)
-
+        
         with col_ghg_comparison:
+
             fig_gwp = px.pie(
-                values = [average_range_impacts(usage.gwp.value), average_range_impacts(embodied.gwp.value)],
-                names = ['usage', 'embodied'],
-                title = 'GHG emissions',
-                color_discrete_sequence=["#636EFA", "#00CC96"],
-                width = 100
-                )
+            values = [average_range_impacts(usage.gwp.value), average_range_impacts(embodied.gwp.value)],
+            names = ['usage', 'embodied'],
+            title = 'GHG emissions',
+            color_discrete_sequence=["#00BF63", "#0B3B36"],
+            width = 100
+            )
             fig_gwp.update_layout(showlegend=False, title_x=0.5)
 
             st.plotly_chart(fig_gwp)
@@ -107,11 +128,10 @@ def expert_mode():
                 values = [average_range_impacts(usage.adpe.value), average_range_impacts(embodied.adpe.value)],
                 names = ['usage', 'embodied'],
                 title = 'Abiotic depletion',
-                color_discrete_sequence=["#00CC96","#636EFA"],
+                color_discrete_sequence=["#0B3B36","#00BF63"],
                 width = 100)
             fig_adpe.update_layout(
-                showlegend=True,
-                legend=dict(yanchor="bottom", x = 0.35, y = -0.1),
+                showlegend=False,
                 title_x=0.5)
             
             st.plotly_chart(fig_adpe)
@@ -121,7 +141,7 @@ def expert_mode():
                 values = [average_range_impacts(usage.pe.value), average_range_impacts(embodied.pe.value)],
                 names = ['usage', 'embodied'],
                 title = 'Primary energy',
-                color_discrete_sequence=["#636EFA", "#00CC96"],
+                color_discrete_sequence=["#00BF63", "#0B3B36"],
                 width = 100)
             fig_pe.update_layout(showlegend=False, title_x=0.5)
 
@@ -139,16 +159,23 @@ def expert_mode():
 
         try:
 
-            df = dataframe_electricity_mix(countries_to_compare)
+            df_comp = dataframe_electricity_mix(countries_to_compare)
 
             impact_type = st.selectbox(
                 label='Select an impact type to compare',
-                options=[x for x in df.columns if x!='country'],
+                options=[x for x in df_comp.columns if x!='country'],
                 index=1)
 
-            df.sort_values(by = impact_type, inplace = True)
-
-            fig_2 = px.bar(df, x = df.index, y = impact_type, text = impact_type, color = impact_type)
+            df_comp.sort_values(by = impact_type, inplace = True)
+            
+            fig_2 = px.bar(
+                df_comp,
+                x = df_comp.index,
+                y = impact_type,
+                text = impact_type,
+                color = impact_type
+            )
+            
             st.plotly_chart(fig_2)
 
         except:

src/models.py

@@ -32,6 +32,8 @@ def clean_models_data(df, with_filter = True):
     
     df['architecture_type'] = df['architecture'].apply(lambda x: x['type'])
     df['architecture_parameters'] = df['architecture'].apply(lambda x: x['parameters'])
+    df['total_parameters'] = df['architecture_parameters'].apply(lambda x: x['total'] if isinstance(x, dict) and 'total' in x.keys() else x)
+    df['active_parameters'] = df['architecture_parameters'].apply(lambda x: x['active'] if isinstance(x, dict) and 'active' in x.keys() else x)
     
     df['warnings'] = df['warnings'].apply(lambda x: ', '.join(x) if x else None).fillna('none')
     df['warning_arch'] = df['warnings'].apply(lambda x: 'model-arch-not-released' in x)
@@ -39,8 +41,8 @@ def clean_models_data(df, with_filter = True):
 
     if with_filter == True:
         df = df[df['name'].isin(models_to_keep)]
-    
-    return df[['provider', 'provider_clean', 'name', 'name_clean', 'architecture_type', 'architecture_parameters', 'warning_arch', 'warning_multi_modal']]
+        
+    return df[['provider', 'provider_clean', 'name', 'name_clean', 'architecture_type', 'architecture_parameters', 'total_parameters', 'active_parameters', 'warning_arch', 'warning_multi_modal']]
 
 @st.cache_data
 def load_models(filter_main = True):

src/utils.py

@@ -1,7 +1,6 @@
 from dataclasses import dataclass
 from enum import Enum
 
-import pandas as pd
 from ecologits.model_repository import models
 from ecologits.impacts.modeling import Impacts, Energy, GWP, ADPe, PE
 #from ecologits.tracers.utils import llm_impacts
@@ -91,12 +90,6 @@ IRELAND_POPULATION_MILLION = 5
 # From https://impactco2.fr/outils/comparateur?value=1&comparisons=&equivalent=avion-pny
 AIRPLANE_PARIS_NYC_GWP_EQ = q("177000 kgCO2eq")
 
-def filter_models(provider, list_models):
-
-    model = 1
-
-    return model
-
 #####################################################################################
 ### IMPACTS FORMATING
 #####################################################################################
@@ -146,8 +139,12 @@ def format_impacts(impacts: Impacts) -> QImpacts:
 def split_impacts_u_e(impacts: Impacts) -> QImpacts:
     return impacts.usage, impacts.embodied
 
-def average_range_impacts(RangeValue):
-    return (RangeValue.max + RangeValue.min)/2
+def average_range_impacts(x):
+    
+    if isinstance(x, float):
+        return x 
+    else:
+        return (x.max + x.min)/2
 
 def format_impacts_expert(impacts: Impacts, display_range: bool) -> QImpacts:
     
@@ -219,44 +216,4 @@ def format_energy_eq_electricity_consumption_ireland(energy: Quantity) -> Quanti
 def format_gwp_eq_airplane_paris_nyc(gwp: Quantity) -> Quantity:
     gwp_eq = gwp * ONE_PERCENT_WORLD_POPULATION * DAYS_IN_YEAR
     gwp_eq = gwp_eq.to("kgCO2eq")
-    return gwp_eq / AIRPLANE_PARIS_NYC_GWP_EQ
-
-#####################################################################################
-### MODELS PARAMETERS
-#####################################################################################
-
-def model_active_params_fn(provider_name: str, model_name: str, n_param: float):
-    if model_name == 'CUSTOM':
-        return n_param
-    else:
-        model = models.find_model(provider=provider_name, model_name=model_name)
-
-        if model.architecture == 'moe':
-            try:
-                return model.architecture.parameters.active.max
-            except:
-                try:
-                    return model.architecture.parameters.active
-                except:
-                    return model.architecture.parameters
-        elif model.architecture == 'dense':
-            try: #dense with range
-                return model.architecture.parameters.max
-            except: #dense without range
-                return model.architecture.parameters
-
-def model_total_params_fn(provider_name: str, model_name: str, n_param: float):
-    if model_name == 'CUSTOM':
-        return n_param
-    provider, model_name = model_name.split('/', 1)
-    model = models.find_model(provider=provider, model_name=model_name)
-    try: #moe
-        return model.architecture.parameters.total.max
-    except:
-        try: #dense with range
-            return model.architecture.parameters.max
-        except: #dense without range
-            try:
-                return model.architecture.parameters.total
-            except:
-                return model.architecture.parameters
+    return gwp_eq / AIRPLANE_PARIS_NYC_GWP_EQ####################################################################################### MODELS PARAMETER####################################################################################