update: filter models + clean files

__init__.py

@@ -1,8 +0,0 @@
-from .content import *
-from .constants import *
-from .expert import expert_mode
-from .utils import *
-from .calculator import calculator_mode
-from .impacts import get_impacts, display_impacts
-from .models import load_models
-from .electricity_mix import *

calculator.py

@@ -1,67 +0,0 @@
-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 PROMPTS
-
-def calculator_mode():
-
-    with st.container(border=True):
-        
-        df = load_models()
-        
-        col1, col2, col3 = st.columns(3)
-
-        with col1:
-            provider = st.selectbox(label = 'Provider',
-                                    options = [x for x in df['provider_clean'].unique()],
-                                    index = 9)
-            provider_raw = df[df['provider_clean'] == provider]['provider'].values[0]
-
-        with col2:
-            model = st.selectbox('Model', [x for x in df['name_clean'].unique() if x in df[df['provider_clean'] == provider]['name_clean'].unique()])
-            model_raw = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]['name'].values[0]
-
-        with col3:
-            output_tokens = st.selectbox('Example prompt', [x[0] for x in PROMPTS])
-            
-        # WARNING DISPLAY
-        
-        df_filtered = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]
-
-        if df_filtered['warning_arch'].values[0] and not df_filtered['warning_multi_modal'].values[0]:
-            st.warning(WARNING_CLOSED_SOURCE)
-        if df_filtered['warning_multi_modal'].values[0] and not df_filtered['warning_arch'].values[0]:
-            st.warning(WARNING_MULTI_MODAL)
-        if df_filtered['warning_arch'].values[0] and df_filtered['warning_multi_modal'].values[0]:
-            st.warning(WARNING_BOTH)
-            
-    try:
-        impacts = llm_impacts(
-                        provider=provider_raw,
-                        model_name=model_raw,
-                        output_token_count=[x[1] for x in PROMPTS if x[0] == output_tokens][0],
-                        request_latency=100000
-                    )
-
-        impacts, _, _ = format_impacts(impacts)
-        
-        with st.container(border=True):
-
-            st.markdown('<h3 align = "center">Environmental impacts</h3>', unsafe_allow_html=True)
-            st.markdown('<p align = "center">To understand how the environmental impacts are computed go to the 📖 Methodology tab.</p>', unsafe_allow_html=True)
-            display_impacts(impacts)
-        
-        with st.container(border=True):
-            
-            st.markdown('<h3 align = "center">That\'s equivalent to ...</h3>', unsafe_allow_html=True)
-            st.markdown('<p align = "center">Making this request to the LLM is equivalent to the following actions :</p>', unsafe_allow_html=True)
-            display_equivalent(impacts)
-            
-    except Exception as e:
-        st.error('Could not find the model in the repository. Please try another model.')

constants.py

@@ -1,106 +0,0 @@
-PROVIDERS = [
-    ("OpenAI", "openai"),
-    ("Anthropic", "anthropic"),
-    ("Cohere", "cohere"),
-    ("Meta", "huggingface_hub/meta"),
-    ("Mistral AI", "mistralai"),
-]
-
-OPENAI_MODELS = [
-    ("GPT-4o", "gpt-4o"),
-    ("GPT-4-Turbo", "gpt-4-turbo"),
-    ("GPT-4", "gpt-4"),
-    ("GPT-3.5-Turbo", "gpt-3.5-turbo"),
-]
-
-ANTHROPIC_MODELS = [
-    ("Claude 3 Opus", "claude-3-opus-20240229"),
-    ("Claude 3 Sonnet", "claude-3-sonnet-20240229"),
-    ("Claude 3 Haiku", "claude-3-haiku-20240307"),
-    ("Claude 2.1", "claude-2.1"),
-    ("Claude 2.0", "claude-2.0"),
-    ("Claude Instant 1.2", "claude-instant-1.2"),
-]
-
-COHERE_MODELS = [
-    ("Command Light", "command-light"),
-    ("Command", "command"),
-    ("Command R", "command-r"),
-    ("Command R+", "command-r-plus"),
-]
-
-META_MODELS = [
-    ("Llama 3 8B", "meta-llama/Meta-Llama-3-8B"),
-    ("Llama 3 70B", "meta-llama/Meta-Llama-3-70B"),
-    ("Llama 2 7B", "meta-llama/Llama-2-7b-hf"),
-    ("Llama 2 13B", "meta-llama/Llama-2-13b-hf"),
-    ("Llama 2 70B", "meta-llama/Llama-2-70b-hf"),
-]
-
-MISTRALAI_MODELS = [
-    ("Mistral 7B", "open-mistral-7b"),
-    ("Mixtral 8x7B", "open-mixtral-8x7b"),
-    ("Mixtral 8x22B", "open-mixtral-8x22b"),
-    ("Tiny", "mistral-tiny-2312"),
-    ("Small", "mistral-small-2402"),
-    ("Medium", "mistral-medium-2312"),
-    ("Large", "mistral-large-2402"),
-]
-
-PROMPTS = [
-    ("Write a Tweet", 50),
-    ("Write an email", 170),
-    ("Write an article summary", 250),
-    ("Small conversation with a chatbot", 400),
-    ("Write a report of 5 pages", 5000),
-    ("Write the code for this app", 10000)
-]
-PROMPTS = [(s + f" ({v} output tokens)", v) for (s, v) in PROMPTS]
-
-CLOSED_SOURCE_MODELS = {
-    "openai/gpt-4o",
-    "openai/gpt-4-turbo",
-    "openai/gpt-4",
-    "openai/gpt-3.5-turbo",
-    "anthropic/claude-3-opus-20240229",
-    "anthropic/claude-3-sonnet-20240229",
-    "anthropic/claude-3-haiku-20240307",
-    "anthropic/claude-2.1",
-    "anthropic/claude-2.0",
-    "anthropic/claude-instant-1.2",
-    "mistralai/mistral-tiny-2312",
-    "mistralai/mistral-small-2402",
-    "mistralai/mistral-medium-2312",
-    "mistralai/mistral-large-2402",
-}
-
-MODELS = [
-    ("OpenAI / GPT-4o", "openai/gpt-4o"),
-    ("OpenAI / GPT-4-Turbo", "openai/gpt-4-turbo"),
-    ("OpenAI / GPT-4", "openai/gpt-4"),
-    ("OpenAI / GPT-3.5-Turbo", "openai/gpt-3.5-turbo"),
-    ("Anthropic / Claude 3 Opus", "anthropic/claude-3-opus-20240229"),
-    ("Anthropic / Claude 3 Sonnet", "anthropic/claude-3-sonnet-20240229"),
-    ("Anthropic / Claude 3 Haiku", "anthropic/claude-3-haiku-20240307"),
-    ("Anthropic / Claude 2.1", "anthropic/claude-2.1"),
-    ("Anthropic / Claude 2.0", "anthropic/claude-2.0"),
-    ("Anthropic / Claude Instant 1.2", "anthropic/claude-instant-1.2"),
-    ("Mistral AI / Mistral 7B", "mistralai/open-mistral-7b"),
-    ("Mistral AI / Mixtral 8x7B", "mistralai/open-mixtral-8x7b"),
-    ("Mistral AI / Mixtral 8x22B", "mistralai/open-mixtral-8x22b"),
-    ("Mistral AI / Tiny", "mistralai/mistral-tiny-2312"),
-    ("Mistral AI / Small", "mistralai/mistral-small-2402"),
-    ("Mistral AI / Medium", "mistralai/mistral-medium-2312"),
-    ("Mistral AI / Large", "mistralai/mistral-large-2402"),
-    ("Meta / Llama 3 8B", "huggingface_hub/meta-llama/Meta-Llama-3-8B"),
-    ("Meta / Llama 3 70B", "huggingface_hub/meta-llama/Meta-Llama-3-70B"),
-    ("Meta / Llama 2 7B", "huggingface_hub/meta-llama/Llama-2-7b-hf"),
-    ("Meta / Llama 2 13B", "huggingface_hub/meta-llama/Llama-2-13b-hf"),
-    ("Meta / Llama 2 70B", "huggingface_hub/meta-llama/Llama-2-70b-hf"),
-    ("Cohere / Command Light", "cohere/command-light"),
-    ("Cohere / Command", "cohere/command"),
-    ("Cohere / Command R", "cohere/command-r"),
-    ("Cohere / Command R+", "cohere/command-r-plus"),
-]
-
-MODEL_REPOSITORY_URL = "https://raw.githubusercontent.com/genai-impact/ecologits/refs/heads/main/ecologits/data/models.json"

content.py

@@ -1,232 +0,0 @@
-HERO_TEXT = """
-<div align="center">
-  <a href="https://ecologits.ai/">
-    <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>
-<br>
-"""
-
-INTRO_TEXT = """
-This tool is developed and maintained by [GenAI Impact](https://genai-impact.org/) non-profit. Learn more about 
-🌱 EcoLogits by reading the documentation on [ecologits.ai](https://ecologits.ai).
- 
-🩷 Support us by giving a ⭐️ on our [GitHub repository](https://github.com/genai-impact/ecologits) and by following our [LinkedIn page](https://www.linkedin.com/company/genai-impact/).
-"""
-
-WARNING_CLOSED_SOURCE = """
-⚠️ The model architecture has not been publicly released, expect lower precision of estimations.
-"""
-
-WARNING_MULTI_MODAL = """
-⚠️ The model architecture is multimodal, expect lower precision of estimations.
-"""
-
-WARNING_BOTH = """
-⚠️ The model architecture has not been publicly released and is multimodal, expect lower precision of estimations.
-"""
-
-ABOUT_TEXT = r"""
-## 🎯 Our goal
-**The main goal of the EcoLogits Calculator is to raise awareness on the environmental impacts of LLM inference.**
-The rapid evolution of generative AI is reshaping numerous industries and aspects of our daily lives. While these 
-advancements offer some benefits, they also **pose substantial environmental challenges that cannot be overlooked**. 
-Plus the issue of AI's environmental footprint has been mainly discussed at training stage but rarely at the inference 
-stage. That is an issue because **inference impacts for LLMs can largely overcome the training impacts when deployed 
-at large scales**.
-At **[GenAI Impact](https://genai-impact.org/) we are dedicated to understanding and mitigating the environmental 
-impacts of generative AI** through rigorous research, innovative tools, and community engagement. Especially, in early 
-2024 we have launched an new open-source tool called [EcoLogits](https://github.com/genai-impact/ecologits) that tracks
-the energy consumption and environmental footprint of using generative AI models through APIs.
-## 🙋 FAQ 
-**How we assess the impacts of closed-source models?**
-Environmental impacts are calculated based on model architecture and parameter count. For closed-source models, we 
-lack transparency from providers, so we estimate parameter counts using available information. For GPT models, we 
-based our estimates on leaked GPT-4 architecture and scaled parameters count for GPT-4-Turbo and GPT-4o based on 
-pricing differences. For other proprietary models like Anthropic's Claude, we assume similar impacts for models 
-released around the same time with similar performance on public benchmarks. Please note that these estimates are 
-based on assumptions and may not be exact. Our methods are open-source and transparent so you can always see the 
-hypotheses we use.
-**Which generative AI models or providers are supported?**
-To see the full list of **generative AI providers** currently supported by EcoLogits, see the following 
-[documentation page](https://ecologits.ai/providers/). As of today we only support LLMs but we plan to add support for 
-embeddings, image generation, multi-modal models and more. If you are interested don't hesitate to 
-[join us](https://genai-impact.org/contact/) and accelerate our work!
-**How to reduce AI environmental impacts?**
-* Look at **indirect impacts** of your project. Does the finality of your project is impacting negatively the 
-environment?
-* **Be frugal** and question your usage or need of AI
-    * Do you really need AI to solve your problem?
-    * Do you really need GenAI to solve your problem? (you can read this [paper](https://aclanthology.org/2023.emnlp-industry.39.pdf))
-    * Use small and specialized models to solve your problem.
-    * Evaluate before, during and after the development of your project the environmental impacts with tools like 
-    🌱 [EcoLogits](https://github.com/genai-impact/ecologits) or [CodeCarbon](https://github.com/mlco2/codecarbon) 
-    (see [more tools](https://github.com/samuelrince/awesome-green-ai))
-    * Restrict the use case and limit the usage of your tool or feature to the desired purpose.
-* Do NOT buy new GPUs / hardware
-    * Hardware manufacturing for data centers is around 50% of the impact.
-* Use cloud instances that are located in low emissions / high energy efficiency data centers 
-(see [electricitymaps.com](https://app.electricitymaps.com/map))
-* Optimize your models for production
-    * Quantize your models.
-    * Use inference optimization tricks.
-    * Prefer fine-tuning of small and existing models over generalist models.
-**What is the difference between **EcoLogits** and [CodeCarbon](https://github.com/mlco2/codecarbon)?**
-EcoLogits is focused on estimating the environmental impacts of generative AI (only LLMs for now) used **through API 
-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 
-  <a href="https://creativecommons.org/licenses/by-sa/4.0/?ref=chooser-v1" target="_blank" rel="license noopener noreferrer" style="display:inline-block;">
-    CC BY-SA 4.0
-  </a>
-  <img style="display:inline-block;height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/cc.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/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/).
-"""
-
-
-METHODOLOGY_TEXT = r"""
-## 📖 Methodology
-We have developed a methodology to **estimate the energy consumption and environmental impacts for an LLM inference** 
-based on request parameters and hypotheses on the data center location, the hardware used, the model architecture and
-more.
-In this section we will only cover the principles of the methodology related to the 🧮 **EcoLogits Calculator**. If 
-you wish to learn more on the environmental impacts modeling of an LLM request checkout the 
-🌱 [EcoLogits documentation page](https://ecologits.ai/methodology/).
-### Modeling impacts of an LLM request
-The environmental impacts of an LLM inference are split into the **usage impacts** $I_{request}^u$ to account for 
-electricity consumption and the **embodied impacts** $I_{request}^e$ that relates to resource extraction, hardware 
-manufacturing and transportation. In general terms it can be expressed as follow:
-$$ I_{request} = I_{request}^u  + I_{request}^e $$
-$$ I_{request} = E_{request}*F_{em}+\frac{\Delta T}{\Delta L}*I_{server}^e $$
-With,
-* $E_{request}$ the estimated energy consumption of the server and its cooling system.
-* $F_{em}$ the electricity mix that depends on the country and time.
-* $\frac{\Delta T}{\Delta L}$ the hardware usage ratio i.e. the computation time over the lifetime of the hardware.
-* $I_{server}^e$ the embodied impacts of the server.
-Additionally, to ⚡️ **direct energy consumption** the environmental impacts are expressed in **three dimensions 
-(multi-criteria impacts)** that are:
-* 🌍 **Global Warming Potential** (GWP): Potential impact on global warming in kgCO2eq (commonly known as GHG/carbon 
-emissions).
-* 🪨 **Abiotic Depletion Potential for Elements** (ADPe): Impact on the depletion of non-living resources such as 
-minerals or metals in kgSbeq.
-* ⛽️ **Primary Energy** (PE): Total energy consumed from primary sources in MJ.
-### Principles, Data and Hypotheses
-We use a **bottom-up methodology** to model impacts, meaning that we will estimate the impacts of low-level physical 
-components to then estimate the impacts at software level (in that case an LLM inference). We also rely on **Life 
-Cycle Approach (LCA) proxies and approach** to model both usage and embodied phases with multi-criteria impacts. 
-If you are interested in this approach we recommend you to read the following [Boavizta](https://boavizta.org/) 
-resources.
-* [Digital & environment: How to evaluate server manufacturing footprint, beyond greenhouse gas emissions?](https://boavizta.org/en/blog/empreinte-de-la-fabrication-d-un-serveur) 
-* [Boavizta API automated evaluation of environmental impacts of ICT services and equipments](https://boavizta.org/en/blog/boavizta-api-automated-evaluation-of-ict-impacts-on-the-environment)
-* [Boavizta API documentation](https://doc.api.boavizta.org/)
-We leverage **open data to estimate the environmental impacts**, here is an exhaustive list of our data providers.
-* [LLM-Perf Leaderboard](https://huggingface.co/spaces/optimum/llm-perf-leaderboard) to estimate GPU energy consumption 
-and latency based on the model architecture and number of output tokens.
-* [Boavizta API](https://github.com/Boavizta/boaviztapi) to estimate server embodied impacts and base energy 
-consumption.
-* [ADEME Base Empreinte®](https://base-empreinte.ademe.fr/) for electricity mix impacts per country.
-Finally here are the **main hypotheses** we have made to compute the impacts.
-* ⚠️ **We *"guesstimate"* the model architecture of proprietary LLMs when not disclosed by the provider.** 
-* Production setup: quantized models running on data center grade servers and GPUs such as A100.
-* Electricity mix does not depend on time (help us enhance EcoLogits and work on this [issue](https://github.com/genai-impact/ecologits/issues/42))
-* Ignore the following impacts: unused cloud resources, data center building, network and end-user devices... (for now)
-## Equivalents
-We have integrated impact equivalents to help people better understand the impacts and have reference points for 
-standard use cases and everyday activities.
-### Request impacts
-These equivalents are computed based on the request impacts only.
-#### 🚶‍♂️‍➡️ Walking or 🏃‍♂️‍➡️ running distance
-We compare the ⚡️ direct energy consumption with the energy consumption of someone 🚶‍♂️‍➡️ walking or 🏃‍♂️‍➡️ running. 
-From [runningtools.com](https://www.runningtools.com/energyusage.htm) we consider the following energy values per 
-physical activity (for someone weighing 70kg):
-* 🚶‍♂️‍➡️ walking: $ 196\ kJ/km $ (speed of $ 3\ km/h $)
-* 🏃‍♂️‍➡️ running: $ 294\ kJ/km $ (speed of $ 10\ km/h $)
-We divide the request energy consumption by these values to compute the distance traveled. 
-#### 🔋 Electric Vehicle distance
-We compare the ⚡️ direct energy consumption with the energy consumer by a EV car. From 
-[selectra.info](https://selectra.info/energie/actualites/insolite/consommation-vehicules-electriques-france-2040) or 
-[tesla.com](https://www.tesla.com/fr_fr/support/power-consumption) we consider an average value of energy consumed per 
-kilometer of: $ 0.17\ kWh/km $.
-We divide the request energy consumption by this value to compute the distance driven by an EV. 
-#### ⏯️ Streaming time
-We compare the 🌍 GHG emissions of the request and of streaming a video. From 
-[impactco2.fr](https://impactco2.fr/outils/comparateur?value=1&comparisons=streamingvideo), we consider that 
-$ 1\ kgCO2eq $ is equivalent to $ 15.6\ h $ of streaming.
-We multiply that value by the GHG emissions of the request to get an equivalent in hours of video streaming.
-### Scaled impacts
-These equivalents are computed based on the request impacts scaled to a worldwide adoption use case. We imply that the
-same request is done 1% of the planet everyday for 1 year, and then compute impact equivalents.
-$$
-I_{scaled} = I_{request} * [1 \\% \ \text{of}\ 8B\ \text{people on earth}] * 365\ \text{days}
-$$
-#### Number of 💨 wind turbines or ☢️ nuclear plants
-We compare the ⚡️ direct energy consumption (scaled) by the energy production of wind turbines and nuclear power 
-plants. From [ecologie.gouv.fr](https://www.ecologie.gouv.fr/eolien-terrestre) we consider that a $ 2\ MW $ wind 
-turbine produces $ 4.2\ GWh $ a year. And from [edf.fr](https://www.edf.fr/groupe-edf/espaces-dedies/jeunes-enseignants/pour-les-jeunes/lenergie-de-a-a-z/produire-de-lelectricite/le-nucleaire-en-chiffres) 
-we learn that a $ 900\ MW $ nuclear power plant produces $ 6\ TWh $ a year.
-We divide the scaled energy consumption by these values to get the number of wind turbines or nuclear power plants 
-needed.
-#### Multiplier of 🇮🇪 Ireland electricity consumption
-We compare the ⚡️ direct energy consumption (scaled) by the electricity consumption of Ireland per year. From 
-[wikipedia.org](https://en.wikipedia.org/wiki/List_of_countries_by_electricity_consumption) we consider the Ireland 
-electricity consumption to be $ 33\ TWh $ a year for a population of 5M.
-We divide the scaled energy consumption by this value to get the equivalent number of "Ireland countries".
-#### Number of ✈️ Paris ↔ New York City flights
-We compare the 🌍 GHG emissions (scaled) of the request and of a return flight Paris ↔ New York City. From 
-[impactco2.fr](https://impactco2.fr/outils/comparateur?value=1&comparisons=&equivalent=avion-pny) we consider that a 
-return flight Paris → New York City → Paris for one passenger emits $ 1,770\ kgCO2eq $ and we consider an overall 
-average load of 100 passengers per flight.
-We divide the scaled GHG emissions by this value to get the equivalent number of return flights.
-
-#### If you are motivated to help us test and enhance this methodology [contact us](https://genai-impact.org/contact/)! 💪
-"""
-
-CITATION_LABEL = "BibTeX citation for EcoLogits Calculator and the EcoLogits library:"
-CITATION_TEXT = r"""@misc{ecologits-calculator,
-  author={Samuel Rincé, Adrien Banse and Valentin Defour},
-  title={EcoLogits Calculator},
-  year={2025},
-  howpublished= {\url{https://huggingface.co/spaces/genai-impact/ecologits-calculator}},
-}
-@software{ecologits,
-  author = {Samuel Rincé, Adrien Banse, Vinh Nguyen and Luc Berton},
-  publisher = {GenAI Impact},
-  title = {EcoLogits: track the energy consumption and environmental footprint of using generative AI models through APIs.},
-}"""
-
-LICENCE_TEXT = """<p xmlns:cc="http://creativecommons.org/ns#" >
-  This work is licensed under 
-  <a href="https://creativecommons.org/licenses/by-sa/4.0/?ref=chooser-v1" target="_blank" rel="license noopener noreferrer" style="display:inline-block;">
-    CC BY-SA 4.0
-  </a>
-  <img style="display:inline-block;height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/cc.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/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>"""

electricity_mix.py

@@ -1,175 +0,0 @@
-from csv import DictReader
-import pandas as pd
-
-PATH = "src/data/electricity_mix.csv"
-
-COUNTRY_CODES = [
-    ("🌎 World", "WOR"),
-    ("🇪🇺 Europe", "EEE"),
-    ("🇿🇼 Zimbabwe", "ZWE"),
-    ("🇿🇲 Zambia", "ZMB"),
-    ("🇿🇦 South Africa", "ZAF"),
-    ("🇾🇪 Yemen", "YEM"),
-    ("🇻🇳 Vietnam", "VNM"),
-    ("🇻🇪 Venezuela", "VEN"),
-    ("🇺🇿 Uzbekistan", "UZB"),
-    ("🇺🇾 Uruguay", "URY"),
-    ("🇺🇸 United States", "USA"),
-    ("🇺🇦 Ukraine", "UKR"),
-    ("🇹🇿 Tanzania", "TZA"),
-    ("🇹🇼 Taiwan", "TWN"),
-    ("🇹🇹 Trinidad and Tobago", "TTO"),
-    ("🇹🇷 Turkey", "TUR"),
-    ("🇹🇳 Tunisia", "TUN"),
-    ("🇹🇲 Turkmenistan", "TKM"),
-    ("🇹🇯 Tajikistan", "TJK"),
-    ("🇹🇭 Thailand", "THA"),
-    ("🇹🇬 Togo", "TGO"),
-    ("🇸🇾 Syrian Arab Republic", "SYR"),
-    ("🇸🇻 El Salvador", "SLV"),
-    ("🇸🇳 Senegal", "SEN"),
-    ("🇸🇰 Slovak Republic", "SVK"),
-    ("🇸🇮 Slovenia", "SVN"),
-    ("🇸🇬 Singapore", "SGP"),
-    ("🇸🇪 Sweden", "SWE"),
-    ("🇸🇩 Sudan", "SDN"),
-    ("🇸🇦 Saudi Arabia", "SAU"),
-    ("🇷🇺 Russian Federation", "RUS"),
-    ("🇷🇸 Serbia and Montenegro", "SCG"),
-    ("🇷🇴 Romania", "ROU"),
-    ("🇶🇦 Qatar", "QAT"),
-    ("🇵🇾 Paraguay", "PRY"),
-    ("🇵🇹 Portugal", "PRT"),
-    ("🇵🇱 Poland", "POL"),
-    ("🇵🇰 Pakistan", "PAK"),
-    ("🇵🇭 Philippines", "PHL"),
-    ("🇵🇪 Peru", "PER"),
-    ("🇵🇦 Panama", "PAN"),
-    ("🇴🇲 Oman", "OMN"),
-    ("🇳🇿 New Zealand", "NZL"),
-    ("🇳🇵 Nepal", "NPL"),
-    ("🇳🇴 Norway", "NOR"),
-    ("🇳🇱 Netherlands", "NLD"),
-    ("🇳🇮 Nicaragua", "NIC"),
-    ("🇳🇬 Nigeria", "NGA"),
-    ("🇳🇦 Namibia", "NAM"),
-    ("🇲🇿 Mozambique", "MOZ"),
-    ("🇲🇾 Malaysia", "MYS"),
-    ("🇲🇽 Mexico", "MEX"),
-    ("🇲🇹 Malta", "MLT"),
-    ("🇲🇳 Mongolia", "MNG"),
-    ("🇲🇲 Myanmar", "MMR"),
-    ("🇲🇰 North Macedonia", "MKD"),
-    ("🇲🇩 Moldova", "MDA"),
-    ("🇲🇦 Morocco", "MAR"),
-    ("🇱🇾 Libya", "LBY"),
-    ("🇱🇻 Latvia", "LVA"),
-    ("🇱🇺 Luxembourg", "LUX"),
-    ("🇱🇹 Lithuania", "LTU"),
-    ("🇱🇰 Sri Lanka", "LKA"),
-    ("🇱🇧 Lebanon", "LBN"),
-    ("🇰🇿 Kazakhstan", "KAZ"),
-    ("🇰🇼 Kuwait", "KWT"),
-    ("🇰🇷 South Korea", "KOR"),
-    ("🇰🇵 North Korea", "PRK"),
-    ("🇰🇭 Cambodia", "KHM"),
-    ("🇰🇬 Kyrgyz Republic", "KGZ"),
-    ("🇰🇪 Kenya", "KEN"),
-    ("🇯🇵 Japan", "JPN"),
-    ("🇯🇴 Jordan", "JOR"),
-    ("🇯🇲 Jamaica", "JAM"),
-    ("🇮🇹 Italy", "ITA"),
-    ("🇮🇸 Iceland", "ISL"),
-    ("🇮🇷 Iran", "IRN"),
-    ("🇮🇶 Iraq", "IRQ"),
-    ("🇮🇳 India", "IND"),
-    ("🇮🇱 Israel", "ISR"),
-    ("🇮🇪 Ireland", "IRL"),
-    ("🇮🇩 Indonesia", "IDN"),
-    ("🇭🇺 Hungary", "HUN"),
-    ("🇭🇹 Haiti", "HTI"),
-    ("🇭🇷 Croatia", "HRV"),
-    ("🇭🇳 Honduras", "HND"),
-    ("🇭🇰 Hong Kong", "HKG"),
-    ("🇬🇹 Guatemala", "GTM"),
-    ("🇬🇷 Greece", "GRC"),
-    ("🇬🇮 Gibraltar", "GIB"),
-    ("🇬🇭 Ghana", "GHA"),
-    ("🇬🇪 Georgia", "GEO"),
-    ("🇬🇧 United Kingdom", "GBR"),
-    ("🇬🇦 Gabon", "GAB"),
-    ("🇫🇷 France", "FRA"),
-    ("🇫🇮 Finland", "FIN"),
-    ("🇪🇹 Ethiopia", "ETH"),
-    ("🇪🇸 Spain", "ESP"),
-    ("🇪🇷 Eritrea", "ERI"),
-    ("🇪🇬 Egypt", "EGY"),
-    ("🇪🇪 Estonia", "EST"),
-    ("🇪🇨 Ecuador", "ECU"),
-    ("🇩🇿 Algeria", "DZA"),
-    ("🇩🇴 Dominican Republic", "DOM"),
-    ("🇩🇰 Denmark", "DNK"),
-    ("🇩🇪 Germany", "DEU"),
-    ("🇨🇿 Czech Republic", "CZE"),
-    ("🇨🇾 Cyprus", "CYP"),
-    ("🇨🇺 Cuba", "CUB"),
-    ("🇨🇷 Costa Rica", "CRI"),
-    ("🇨🇴 Colombia", "COL"),
-    ("🇨🇳 China", "CHN"),
-    ("🇨🇲 Cameroon", "CMR"),
-    ("🇨🇱 Chile", "CHL"),
-    ("🇨🇮 Cote d'Ivoire", "CIV"),
-    ("🇨🇭 Switzerland", "CHE"),
-    ("🇨🇬 Congo", "COG"),
-    ("🇨🇩 Democratic Republic of the Congo", "COD"),
-    ("🇨🇦 Canada", "CAN"),
-    ("🇧🇾 Belarus", "BLR"),
-    ("🇧🇼 Botswana", "BWA"),
-    ("🇧🇷 Brazil", "BRA"),
-    ("🇧🇴 Bolivia", "BOL"),
-    ("🇧🇳 Brunei", "BRN"),
-    ("🇧🇯 Benin", "BEN"),
-    ("🇧🇭 Bahrain", "BHR"),
-    ("🇧🇬 Bulgaria", "BGR"),
-    ("🇧🇪 Belgium", "BEL"),
-    ("🇧🇩 Bangladesh", "BGD"),
-    ("🇧🇦 Bosnia and Herzegovina", "BIH"),
-    ("🇦🇿 Azerbaijan", "AZE"),
-    ("🇦🇺 Australia", "AUS"),
-    ("🇦🇹 Austria", "AUT"),
-    ("🇦🇷 Argentina", "ARG"),
-    ("🇦🇴 Angola", "AGO"),
-    ("🇦 Netherlands Antilles", "ANT"),
-    ("🇦🇲 Armenia", "ARM"),
-    ("🇦🇱 Albania", "ALB"),
-    ("🇦🇪 United Arab Emirates", "ARE")
-]
-
-
-def find_electricity_mix(code: str):
-    # TODO: Maybe more optimal to construct database at the beginning of the app
-    #       in the same fashion as find_model
-    res = []
-    with open(PATH) as fd:
-        csv = DictReader(fd)
-        for row in csv:
-            res += [float(row[code])]
-    return res
-
-def dataframe_electricity_mix(countries: list):
-
-    df = pd.read_csv('src/data/electricity_mix.csv')
-    df['name_unit'] = df['name'] + ' (' + df['unit'] + ')'
-    df = df[['name_unit'] + [x[1] for x in COUNTRY_CODES if x[0] in countries]]
-
-    df_melted = df.melt(
-        id_vars=['name_unit'],
-        value_vars=[x[1] for x in COUNTRY_CODES if x[0] in countries],
-        var_name='country',
-        value_name='value')
-
-    df = df_melted.pivot(columns='name_unit', 
-                        index='country',
-                        values='value')
-
-    return df

expert.py

@@ -1,156 +0,0 @@
-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.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
-
-import plotly.express as px
-
-def reset_model():
-    model = 'CUSTOM'
-
-def expert_mode():
-
-    st.markdown("### 🤓 Expert mode")
-
-    with st.container(border = True):
-
-        ########## Model info ##########
-
-        # col1, col2, col3 = st.columns(3)
-
-        # 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 = 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]
-
-        ########## 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)
-
-        with col22:
-            total_params = st.number_input('Total parameters (B)', 0, None, 45)
-
-        with col33:
-            output_tokens = st.number_input('Output completion tokens', 100)
-
-        ########## Electricity mix ##########
-
-        location = st.selectbox('Location', [x[0] for x in COUNTRY_CODES])
-
-        col4, col5, col6 = st.columns(3)
-
-        with col4:
-            mix_gwp = st.number_input('Electricity mix - GHG emissions [kgCO2eq / kWh]', find_electricity_mix([x[1] for x in COUNTRY_CODES if x[0] ==location][0])[2], format="%0.6f")
-            #disp_ranges = st.toggle('Display impact ranges', False)
-        with col5:
-            mix_adpe = st.number_input('Electricity mix - Abiotic resources [kgSbeq / kWh]', find_electricity_mix([x[1] for x in COUNTRY_CODES if x[0] ==location][0])[0], format="%0.13f")
-        with col6:
-            mix_pe = st.number_input('Electricity mix - Primary energy [MJ / kWh]', find_electricity_mix([x[1] for x in COUNTRY_CODES if x[0] ==location][0])[1], format="%0.3f")
-
-    impacts = compute_llm_impacts(model_active_parameter_count=active_params,
-                model_total_parameter_count=total_params,
-                output_token_count=output_tokens,
-                request_latency=100000,
-                if_electricity_mix_gwp=mix_gwp,
-                if_electricity_mix_adpe=mix_adpe,
-                if_electricity_mix_pe=mix_pe
-            )
-    
-    impacts, usage, embodied = format_impacts(impacts)
-    
-    with st.container(border = True):
-
-        st.markdown('<h3 align="center">Environmental Impacts</h2>', unsafe_allow_html = True)
-
-        display_impacts(impacts)
-
-    with st.expander('⚖️ Usage vs Embodied'):
-
-        st.markdown('<h3 align="center">Embodied vs Usage comparison</h2>', unsafe_allow_html = True)
-
-        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
-                )
-            fig_gwp.update_layout(showlegend=False, title_x=0.5)
-
-            st.plotly_chart(fig_gwp)
-
-        with col_adpe_comparison:
-            fig_adpe = px.pie(
-                values = [average_range_impacts(usage.adpe.value), average_range_impacts(embodied.adpe.value)],
-                names = ['usage', 'embodied'],
-                title = 'Abiotic depletion',
-                color_discrete_sequence=["#00CC96","#636EFA"],
-                width = 100)
-            fig_adpe.update_layout(
-                showlegend=True,
-                legend=dict(yanchor="bottom", x = 0.35, y = -0.1),
-                title_x=0.5)
-            
-            st.plotly_chart(fig_adpe)
-
-        with col_pe_comparison:
-            fig_pe = px.pie(
-                values = [average_range_impacts(usage.pe.value), average_range_impacts(embodied.pe.value)],
-                names = ['usage', 'embodied'],
-                title = 'Primary energy',
-                color_discrete_sequence=["#636EFA", "#00CC96"],
-                width = 100)
-            fig_pe.update_layout(showlegend=False, title_x=0.5)
-
-            st.plotly_chart(fig_pe)
-
-    with st.expander('🌍️ Location impact'):
-
-        st.markdown('<h4 align="center">How can location impact the footprint ?</h4>', unsafe_allow_html = True)
-
-        countries_to_compare = st.multiselect(
-            label = 'Countries to compare',
-            options = [x[0] for x in COUNTRY_CODES],
-            default = ["🇫🇷 France", "🇺🇸 United States", "🇨🇳 China"]
-            )
-
-        try:
-
-            df = 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'],
-                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)
-            st.plotly_chart(fig_2)
-
-        except:
-
-            st.warning("Can't display chart with no values.")

impacts.py

@@ -1,116 +0,0 @@
-import streamlit as st
-import ecologits
-from src.utils import (
-    format_energy_eq_electric_vehicle,
-    format_energy_eq_electricity_consumption_ireland,
-    format_energy_eq_electricity_production,
-    format_energy_eq_physical_activity,
-    format_gwp_eq_airplane_paris_nyc,
-    format_gwp_eq_streaming,
-    PhysicalActivity,
-    EnergyProduction
-)
-
-############################################################################################################
-
-def get_impacts(model, active_params, total_params, mix_ghg, mix_adpe, mix_pe):
-
-    return 1
-
-############################################################################################################
-
-
-def display_impacts(impacts):
-
-    st.divider()
-
-    col_energy, col_ghg, col_adpe, col_pe, col_water = st.columns(5)
-
-    with col_energy:
-        st.markdown('<h4 align="center">⚡️ Energy</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {impacts.energy.magnitude:.3g} \ \large {impacts.energy.units}')
-        st.markdown(f'<p align="center"><i>Evaluates the electricity consumption<i></p>', unsafe_allow_html = True)
-
-    with col_ghg:
-        st.markdown('<h4 align="center">🌍️ GHG Emissions</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {impacts.gwp.magnitude:.3g} \ \large {impacts.gwp.units}')
-        st.markdown(f'<p align="center"><i>Evaluates the effect on global warming<i></p>', unsafe_allow_html = True)
-
-    with col_adpe:
-        st.markdown('<h4 align="center">🪨 Abiotic Resources</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {impacts.adpe.magnitude:.3g} \ \large {impacts.adpe.units}')
-        st.markdown(f'<p align="center"><i>Evaluates the use of metals and minerals<i></p>', unsafe_allow_html = True)
-
-    with col_pe:
-        st.markdown('<h4 align="center">⛽️ Primary Energy</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {impacts.pe.magnitude:.3g} \ \large {impacts.pe.units}')
-        st.markdown(f'<p align="center"><i>Evaluates the use of energy resources<i></p>', unsafe_allow_html = True)
-
-    with col_water:
-        st.markdown('<h4 align="center">🚰 Water</h4>', unsafe_allow_html = True)
-        st.latex('\Large Upcoming...')
-        st.markdown(f'<p align="center"><i>Evaluates the use of water<i></p>', unsafe_allow_html = True)
-
-############################################################################################################
-
-def display_equivalent(impacts):
-
-    st.divider()
-
-    ev_eq = format_energy_eq_electric_vehicle(impacts.energy)
-    
-    streaming_eq = format_gwp_eq_streaming(impacts.gwp)
-
-    col1, col2, col3 = st.columns(3)
-
-    with col1:
-        physical_activity, distance = format_energy_eq_physical_activity(impacts.energy)
-        if physical_activity == PhysicalActivity.WALKING:
-            physical_activity = "🚶 " + physical_activity.capitalize()
-        if physical_activity == PhysicalActivity.RUNNING:
-            physical_activity = "🏃 " + physical_activity.capitalize()
-
-        st.markdown(f'<h4 align="center">{physical_activity}</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {distance.magnitude:.3g} \ \large {distance.units}')
-        st.markdown(f'<p align="center"><i>Based on energy consumption<i></p>', unsafe_allow_html = True)
-
-    with col2:
-        ev_eq = format_energy_eq_electric_vehicle(impacts.energy)
-        st.markdown(f'<h4 align="center">🔋 Electric Vehicle</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {ev_eq.magnitude:.3g} \ \large {ev_eq.units}')
-        st.markdown(f'<p align="center"><i>Based on energy consumption<i></p>', unsafe_allow_html = True)
-
-    with col3:
-        streaming_eq = format_gwp_eq_streaming(impacts.gwp)
-        st.markdown(f'<h4 align="center">⏯️ Streaming</h4>', unsafe_allow_html = True)
-        st.latex(f'\Large {streaming_eq.magnitude:.3g} \ \large {streaming_eq.units}')
-        st.markdown(f'<p align="center"><i>Based on GHG emissions<i></p>', unsafe_allow_html = True)
-    
-    st.divider()
-    
-    st.markdown('<h3 align="center">What if 1% of the planet does this request everyday for 1 year ?</h3>', unsafe_allow_html = True)
-    st.markdown('<p align="center">If this use case is largely deployed around the world, the equivalent impacts would be the impacts of this request x 1% of 8 billion people x 365 days in a year.</p>', unsafe_allow_html = True)
-
-    col4, col5, col6 = st.columns(3)
-
-    with col4:
-
-        electricity_production, count = format_energy_eq_electricity_production(impacts.energy)
-        if electricity_production == EnergyProduction.NUCLEAR:
-            emoji = "☢️"
-            name = "Nuclear power plants"
-        if electricity_production == EnergyProduction.WIND:
-            emoji = "💨️ "
-            name = "Wind turbines"
-        st.markdown(f'<h4 align="center">{emoji} {count.magnitude:.0f} {name} (yearly)</h4>', unsafe_allow_html = True)
-        st.markdown(f'<p align="center"><i>Based on energy consumption<i></p>', unsafe_allow_html = True)
-        
-    with col5:
-        ireland_count = format_energy_eq_electricity_consumption_ireland(impacts.energy)
-        st.markdown(f'<h4 align="center">🇮🇪 {ireland_count.magnitude:.3f} x Ireland <span style="font-size: 12px">(yearly ⚡️ cons.)</span></h2></h4>', unsafe_allow_html = True)
-        st.markdown(f'<p align="center"><i>Based on energy consumption<i></p>', unsafe_allow_html = True)
-
-    with col6:
-        paris_nyc_airplane = format_gwp_eq_airplane_paris_nyc(impacts.gwp)
-        st.markdown(f'<h4 align="center">✈️ {round(paris_nyc_airplane.magnitude):,} Paris ↔ NYC</h4>', unsafe_allow_html = True)
-        st.markdown(f'<p align="center"><i>Based on GHG emissions<i></p>', unsafe_allow_html = True)

models.py

@@ -1,46 +0,0 @@
-import requests
-import json
-import pandas as pd
-from src.constants import MODEL_REPOSITORY_URL
-
-def clean_models_data(df):
-    
-    dict_providers = {
-        'google': 'Google',
-        'mistralai': 'MistralAI',
-        'meta-llama': 'Meta',
-        'openai': 'OpenAI',
-        'anthropic': 'Anthropic',
-        'cohere': 'Cohere',
-        'microsoft': 'Microsoft',
-        'mistral-community': 'Mistral Community',
-        'databricks': 'Databricks'
-    }
-    
-    df.drop('type', axis=1, inplace=True)
-
-    df.loc[df['name'].str.contains('/'), 'name_clean'] = df.loc[df['name'].str.contains('/'), 'name'].str.split('/').str[1]
-    df['name_clean'] = df['name_clean'].fillna(df['name'])
-    df['name_clean'] = df['name_clean'].replace({'-': ' '}, regex = True)
-    
-    df.loc[df['provider'] == 'huggingface_hub', 'provider_clean'] = df.loc[df['provider'] == 'huggingface_hub', 'name'].str.split('/').str[0]
-    df['provider_clean'] = df['provider_clean'].fillna(df['provider'])
-    df['provider_clean'] = df['provider_clean'].replace(dict_providers, regex = True)
-    
-    df['architecture_type'] = df['architecture'].apply(lambda x: x['type'])
-    df['architecture_parameters'] = df['architecture'].apply(lambda x: x['parameters'])
-    
-    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)
-    df['warning_multi_modal'] = df['warnings'].apply(lambda x: 'model-arch-multimodal' in x)
-    
-    return df[['provider', 'provider_clean', 'name', 'name_clean', 'architecture_type', 'architecture_parameters', 'warning_arch', 'warning_multi_modal']]
-
-def load_models():
-
-    resp = requests.get(MODEL_REPOSITORY_URL)
-    data = json.loads(resp.text)
-    df = pd.DataFrame(data['models'])
-
-    return clean_models_data(df)
-

src/calculator.py

@@ -6,6 +6,7 @@ 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.constants import PROMPTS
 
@@ -13,25 +14,30 @@ def calculator_mode():
 
     with st.container(border=True):
         
-        df = load_models()
+        df = load_models(filter_main=True)
         
         col1, col2, col3 = st.columns(3)
 
         with col1:
-            provider = st.selectbox(label = 'Provider',
-                                    options = [x for x in df['provider_clean'].unique()],
-                                    index = 9)
-            provider_raw = df[df['provider_clean'] == provider]['provider'].values[0]
+            provider = st.selectbox(
+                label = 'Provider',
+                options = [x for x in df['provider_clean'].unique()],
+                index = 7
+            )
 
         with col2:
-            model = st.selectbox('Model', [x for x in df['name_clean'].unique() if x in df[df['provider_clean'] == provider]['name_clean'].unique()])
-            model_raw = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]['name'].values[0]
+            model = st.selectbox(
+                label = 'Model',
+                options = [x for x in df['name_clean'].unique() if x in df[df['provider_clean'] == provider]['name_clean'].unique()]
+            )
 
         with col3:
             output_tokens = st.selectbox('Example prompt', [x[0] for x in PROMPTS])
             
         # WARNING DISPLAY
-        
+        provider_raw = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]['provider'].values[0]
+        model_raw = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]['name'].values[0]
+
         df_filtered = df[(df['provider_clean'] == provider) & (df['name_clean'] == model)]
 
         if df_filtered['warning_arch'].values[0] and not df_filtered['warning_multi_modal'].values[0]:
@@ -42,6 +48,7 @@ def calculator_mode():
             st.warning(WARNING_BOTH)
             
     try:
+
         impacts = llm_impacts(
                         provider=provider_raw,
                         model_name=model_raw,

src/constants.py

@@ -1,106 +1,102 @@
-PROVIDERS = [
-    ("OpenAI", "openai"),
-    ("Anthropic", "anthropic"),
-    ("Cohere", "cohere"),
-    ("Meta", "huggingface_hub/meta"),
-    ("Mistral AI", "mistralai"),
+PROMPTS = [
+    ("Write a Tweet", 50),
+    ("Write an email", 170),
+    ("Write an article summary", 250),
+    ("Small conversation with a chatbot", 400),
+    ("Write a report of 5 pages", 5000),
+    ("Write the code for this app", 15000)
 ]
+PROMPTS = [(s + f" ({v} output tokens)", v) for (s, v) in PROMPTS]
+
+MODEL_REPOSITORY_URL = "https://raw.githubusercontent.com/genai-impact/ecologits/refs/heads/main/ecologits/data/models.json"
 
-OPENAI_MODELS = [
-    ("GPT-4o", "gpt-4o"),
-    ("GPT-4-Turbo", "gpt-4-turbo"),
-    ("GPT-4", "gpt-4"),
-    ("GPT-3.5-Turbo", "gpt-3.5-turbo"),
+main_models_openai = [
+    'chatgpt-4o-latest',
+    'gpt-3.5-turbo',
+    'gpt-4',
+    'gpt-4-turbo',
+    'gpt-4o',
+    'gpt-4o-mini',
+    'o1',
+    'o1-mini'
 ]
 
-ANTHROPIC_MODELS = [
-    ("Claude 3 Opus", "claude-3-opus-20240229"),
-    ("Claude 3 Sonnet", "claude-3-sonnet-20240229"),
-    ("Claude 3 Haiku", "claude-3-haiku-20240307"),
-    ("Claude 2.1", "claude-2.1"),
-    ("Claude 2.0", "claude-2.0"),
-    ("Claude Instant 1.2", "claude-instant-1.2"),
+main_models_meta = [
+    'meta-llama/Meta-Llama-3.1-8B',
+    'meta-llama/Meta-Llama-3.1-70B',
+    'meta-llama/Meta-Llama-3.1-405B',
+    'meta-llama/Meta-Llama-3-8B',
+    'meta-llama/Meta-Llama-3-70B',
+    'meta-llama/Meta-Llama-3-70B',
+    'meta-llama/Llama-2-7b',
+    'meta-llama/Llama-2-13b',
+    'meta-llama/Llama-2-70b',
+    'meta-llama/CodeLlama-7b-hf',
+    'meta-llama/CodeLlama-13b-hf',
+    'meta-llama/CodeLlama-34b-hf',
+    'meta-llama/CodeLlama-70b-hf'
 ]
 
-COHERE_MODELS = [
-    ("Command Light", "command-light"),
-    ("Command", "command"),
-    ("Command R", "command-r"),
-    ("Command R+", "command-r-plus"),
+main_models_msft = [
+    'microsoft/phi-1',
+    'microsoft/phi-1_5',
+    'microsoft/Phi-3-mini-128k-instruct',
+    'microsoft/Phi-3-small-128k-instruct',
+    'microsoft/Phi-3-medium-128k-instruct',
 ]
 
-META_MODELS = [
-    ("Llama 3 8B", "meta-llama/Meta-Llama-3-8B"),
-    ("Llama 3 70B", "meta-llama/Meta-Llama-3-70B"),
-    ("Llama 2 7B", "meta-llama/Llama-2-7b-hf"),
-    ("Llama 2 13B", "meta-llama/Llama-2-13b-hf"),
-    ("Llama 2 70B", "meta-llama/Llama-2-70b-hf"),
+main_models_anthropic = [
+    'claude-2.0',
+    'claude-2.1',
+    'claude-3-5-haiku-latest',
+    'claude-3-5-sonnet-latest',
+    'claude-3-7-sonnet-latest',
+    'claude-3-haiku-20240307',
+    'claude-3-opus-latest',
+    'claude-3-sonnet-20240229'
 ]
 
-MISTRALAI_MODELS = [
-    ("Mistral 7B", "open-mistral-7b"),
-    ("Mixtral 8x7B", "open-mixtral-8x7b"),
-    ("Mixtral 8x22B", "open-mixtral-8x22b"),
-    ("Tiny", "mistral-tiny-2312"),
-    ("Small", "mistral-small-2402"),
-    ("Medium", "mistral-medium-2312"),
-    ("Large", "mistral-large-2402"),
+main_models_cohere = [
+    'c4ai-aya-expanse-8b',
+    'c4ai-aya-expanse-32b',
+    'command',
+    'command-light',
+    'command-r',
+    'command-r-plus',
 ]
 
-PROMPTS = [
-    ("Write a Tweet", 50),
-    ("Write an email", 170),
-    ("Write an article summary", 250),
-    ("Small conversation with a chatbot", 400),
-    ("Write a report of 5 pages", 5000),
-    ("Write the code for this app", 10000)
+main_models_google = [
+    'google/gemma-2-2b',
+    'google/gemma-2-9b',
+    'google/gemma-2-27b',
+    'google/codegemma-2b',
+    'google/codegemma-7b',
+    'gemini-1.0-pro',
+    'gemini-1.5-pro',
+    'gemini-1.5-flash',
+    'gemini-2.0-flash'
 ]
-PROMPTS = [(s + f" ({v} output tokens)", v) for (s, v) in PROMPTS]
 
-CLOSED_SOURCE_MODELS = {
-    "openai/gpt-4o",
-    "openai/gpt-4-turbo",
-    "openai/gpt-4",
-    "openai/gpt-3.5-turbo",
-    "anthropic/claude-3-opus-20240229",
-    "anthropic/claude-3-sonnet-20240229",
-    "anthropic/claude-3-haiku-20240307",
-    "anthropic/claude-2.1",
-    "anthropic/claude-2.0",
-    "anthropic/claude-instant-1.2",
-    "mistralai/mistral-tiny-2312",
-    "mistralai/mistral-small-2402",
-    "mistralai/mistral-medium-2312",
-    "mistralai/mistral-large-2402",
-}
+main_models_databricks = [
+    'databricks/dolly-v1-6b',
+    'databricks/dolly-v2-12b',
+    'databricks/dolly-v2-7b',
+    'databricks/dolly-v2-3b',
+    'databricks/dbrx-base'
+]
 
-MODELS = [
-    ("OpenAI / GPT-4o", "openai/gpt-4o"),
-    ("OpenAI / GPT-4-Turbo", "openai/gpt-4-turbo"),
-    ("OpenAI / GPT-4", "openai/gpt-4"),
-    ("OpenAI / GPT-3.5-Turbo", "openai/gpt-3.5-turbo"),
-    ("Anthropic / Claude 3 Opus", "anthropic/claude-3-opus-20240229"),
-    ("Anthropic / Claude 3 Sonnet", "anthropic/claude-3-sonnet-20240229"),
-    ("Anthropic / Claude 3 Haiku", "anthropic/claude-3-haiku-20240307"),
-    ("Anthropic / Claude 2.1", "anthropic/claude-2.1"),
-    ("Anthropic / Claude 2.0", "anthropic/claude-2.0"),
-    ("Anthropic / Claude Instant 1.2", "anthropic/claude-instant-1.2"),
-    ("Mistral AI / Mistral 7B", "mistralai/open-mistral-7b"),
-    ("Mistral AI / Mixtral 8x7B", "mistralai/open-mixtral-8x7b"),
-    ("Mistral AI / Mixtral 8x22B", "mistralai/open-mixtral-8x22b"),
-    ("Mistral AI / Tiny", "mistralai/mistral-tiny-2312"),
-    ("Mistral AI / Small", "mistralai/mistral-small-2402"),
-    ("Mistral AI / Medium", "mistralai/mistral-medium-2312"),
-    ("Mistral AI / Large", "mistralai/mistral-large-2402"),
-    ("Meta / Llama 3 8B", "huggingface_hub/meta-llama/Meta-Llama-3-8B"),
-    ("Meta / Llama 3 70B", "huggingface_hub/meta-llama/Meta-Llama-3-70B"),
-    ("Meta / Llama 2 7B", "huggingface_hub/meta-llama/Llama-2-7b-hf"),
-    ("Meta / Llama 2 13B", "huggingface_hub/meta-llama/Llama-2-13b-hf"),
-    ("Meta / Llama 2 70B", "huggingface_hub/meta-llama/Llama-2-70b-hf"),
-    ("Cohere / Command Light", "cohere/command-light"),
-    ("Cohere / Command", "cohere/command"),
-    ("Cohere / Command R", "cohere/command-r"),
-    ("Cohere / Command R+", "cohere/command-r-plus"),
+main_models_mistral = [
+    'mistralai/Mistral-7B-v0.3',
+    'mistralai/Mixtral-8x7B-v0.1',
+    'mistralai/Mixtral-8x22B-v0.1',
+    'mistralai/Codestral-22B-v0.1',
+    'mistralai/Mathstral-7B-v0.1',
+    'ministral-3b-latest',
+    'ministral-8b-latest',
+    'mistral-tiny',
+    'mistral-small',
+    'mistral-medium',
+    'mistral-large-latest'
 ]
 
-MODEL_REPOSITORY_URL = "https://raw.githubusercontent.com/genai-impact/ecologits/refs/heads/main/ecologits/data/models.json"
+MAIN_MODELS = main_models_meta + main_models_openai + main_models_anthropic + main_models_cohere + main_models_msft + main_models_mistral + main_models_databricks + main_models_google

src/models.py

@@ -1,9 +1,10 @@
 import requests
 import json
 import pandas as pd
-from src.constants import MODEL_REPOSITORY_URL
+from src.constants import MODEL_REPOSITORY_URL, MAIN_MODELS
+import streamlit as st
 
-def clean_models_data(df):
+def clean_models_data(df, with_filter = True):
     
     dict_providers = {
         'google': 'Google',
@@ -16,12 +17,14 @@ def clean_models_data(df):
         'mistral-community': 'Mistral Community',
         'databricks': 'Databricks'
     }
+
+    models_to_keep = MAIN_MODELS
     
     df.drop('type', axis=1, inplace=True)
 
     df.loc[df['name'].str.contains('/'), 'name_clean'] = df.loc[df['name'].str.contains('/'), 'name'].str.split('/').str[1]
     df['name_clean'] = df['name_clean'].fillna(df['name'])
-    df['name_clean'] = df['name_clean'].replace({'-': ' '}, regex = True)
+    df['name_clean'] = df['name_clean'].replace({'-': ' ', 'latest': ''}, regex = True)
     
     df.loc[df['provider'] == 'huggingface_hub', 'provider_clean'] = df.loc[df['provider'] == 'huggingface_hub', 'name'].str.split('/').str[0]
     df['provider_clean'] = df['provider_clean'].fillna(df['provider'])
@@ -33,14 +36,18 @@ def clean_models_data(df):
     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)
     df['warning_multi_modal'] = df['warnings'].apply(lambda x: 'model-arch-multimodal' in x)
+
+    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']]
 
-def load_models():
+@st.cache_data
+def load_models(filter_main = True):
 
     resp = requests.get(MODEL_REPOSITORY_URL)
     data = json.loads(resp.text)
     df = pd.DataFrame(data['models'])
 
-    return clean_models_data(df)
+    return clean_models_data(df, filter_main)
 

utils.py

@@ -1,262 +0,0 @@
-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
-from pint import UnitRegistry, Quantity
-
-#####################################################################################
-### UNITS DEFINITION
-#####################################################################################
-
-u = UnitRegistry()
-u.define('Wh = watt_hour')
-u.define('kWh = kilowatt_hour')
-u.define('MWh = megawatt_hour')
-u.define('GWh = gigawatt_hour')
-u.define('TWh = terawatt_hour')
-u.define('gCO2eq = gram')
-u.define('kgCO2eq = kilogram')
-u.define('tCO2eq = metricton')
-u.define('kgSbeq = kilogram')
-u.define('kJ = kilojoule')
-u.define('MJ = megajoule')
-u.define('m = meter')
-u.define('km = kilometer')
-u.define('s = second')
-u.define('min = minute')
-u.define('h = hour')
-q = u.Quantity
-
-@dataclass
-class QImpacts:
-    energy: Quantity
-    gwp: Quantity
-    adpe: Quantity
-    pe: Quantity
-
-
-class PhysicalActivity(str, Enum):
-    RUNNING = "running"
-    WALKING = "walking"
-
-
-class EnergyProduction(str, Enum):
-    NUCLEAR = "nuclear"
-    WIND = "wind"
-
-
-COUNTRIES = [
-    ("cook_islands", 38.81, 9_556),
-    ("tonga", 51.15, 104_490),
-    ("comoros", 100, 821_632),
-    ("samoa", 100, 821_632),
-]
-
-#####################################################################################
-### EQUIVALENT RAW DATA
-#####################################################################################
-
-# From https://www.runningtools.com/energyusage.htm
-RUNNING_ENERGY_EQ = q("294 kJ / km")     # running 1 km at 10 km/h with a weight of 70 kg
-WALKING_ENERGY_EQ = q("196 kJ / km")     # walking 1 km at 3 km/h with a weight of 70 kg
-
-# From https://selectra.info/energie/actualites/insolite/consommation-vehicules-electriques-france-2040
-# and https://www.tesla.com/fr_fr/support/power-consumption
-EV_ENERGY_EQ = q("0.17 kWh / km")
-
-# From https://impactco2.fr/outils/comparateur?value=1&comparisons=streamingvideo
-STREAMING_GWP_EQ = q("15.6 h / kgCO2eq")
-
-# From https://ourworldindata.org/population-growth
-ONE_PERCENT_WORLD_POPULATION = 80_000_000
-
-DAYS_IN_YEAR = 365
-
-# For a 900 MW nuclear plant -> 500 000 MWh / month
-# From https://www.edf.fr/groupe-edf/espaces-dedies/jeunes-enseignants/pour-les-jeunes/lenergie-de-a-a-z/produire-de-lelectricite/le-nucleaire-en-chiffres
-YEARLY_NUCLEAR_ENERGY_EQ = q("6 TWh")
-
-# For a 2MW wind turbine
-# https://www.ecologie.gouv.fr/eolien-terrestre
-YEARLY_WIND_ENERGY_EQ = q("4.2 GWh")
-
-# Ireland yearly electricity consumption
-# From https://en.wikipedia.org/wiki/List_of_countries_by_electricity_consumption
-YEARLY_IRELAND_ELECTRICITY_CONSUMPTION = q("33 TWh")
-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
-#####################################################################################
-
-def format_energy(energy: Energy) -> Quantity:
-    val = q(energy.value, energy.unit)
-    if val < q("1 kWh"):
-        val = val.to("Wh")
-    return val
-
-def format_gwp(gwp: GWP) -> Quantity:
-    val = q(gwp.value, gwp.unit)
-    if val < q("1 kgCO2eq"):
-        val = val.to("gCO2eq")
-    return val
-
-def format_adpe(adpe: ADPe) -> Quantity:
-    return q(adpe.value, adpe.unit)
-
-def format_pe(pe: PE) -> Quantity:
-    val = q(pe.value, pe.unit)
-    if val < q("1 MJ"):
-        val = val.to("kJ")
-    return val
-
-def format_impacts(impacts: Impacts) -> QImpacts:
-
-    try:
-        impacts.energy.value = (impacts.energy.value.max + impacts.energy.value.min)/2
-        impacts.gwp.value = (impacts.gwp.value.max + impacts.gwp.value.min)/2
-        impacts.adpe.value = (impacts.adpe.value.max + impacts.adpe.value.min)/2
-        impacts.pe.value = (impacts.pe.value.max + impacts.pe.value.min)/2
-        return QImpacts(
-            energy=format_energy(impacts.energy),
-            gwp=format_gwp(impacts.gwp),
-            adpe=format_adpe(impacts.adpe),
-            pe=format_pe(impacts.pe)
-        ), impacts.usage, impacts.embodied
-    except: #when no range
-        return QImpacts(
-            energy=format_energy(impacts.energy),
-            gwp=format_gwp(impacts.gwp),
-            adpe=format_adpe(impacts.adpe),
-            pe=format_pe(impacts.pe)
-        ), impacts.usage, impacts.embodied
-
-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 format_impacts_expert(impacts: Impacts, display_range: bool) -> QImpacts:
-    
-    if display_range:
-        return QImpacts(
-            energy=format_energy(impacts.energy),
-            gwp=format_gwp(impacts.gwp),
-            adpe=format_adpe(impacts.adpe),
-            pe=format_pe(impacts.pe)
-        ), impacts.usage, impacts.embodied
-    
-    else:
-        energy = {"value":(impacts.energy.value.max + impacts.energy.value.min)/2, "unit":impacts.energy.unit}
-        gwp = (impacts.gwp.value.max + impacts.gwp.value.min)/2
-        adpe = (impacts.adpe.value.max + impacts.adpe.value.min)/2
-        pe = (impacts.pe.value.max + impacts.pe.value.min)/2
-        return QImpacts(
-            energy=format_energy(energy),
-            gwp=format_gwp(gwp),
-            adpe=format_adpe(adpe),
-            pe=format_pe(pe)
-        ), impacts.usage, impacts.embodied
-
-#####################################################################################
-### EQUIVALENT FORMATING
-#####################################################################################
-
-def format_energy_eq_physical_activity(energy: Quantity) -> tuple[PhysicalActivity, Quantity]:
-    energy = energy.to("kJ")
-    running_eq = energy / RUNNING_ENERGY_EQ
-    if running_eq > q("1 km"):
-        return PhysicalActivity.RUNNING, running_eq
-
-    walking_eq = energy / WALKING_ENERGY_EQ
-    if walking_eq < q("1 km"):
-        walking_eq = walking_eq.to("meter")
-    return PhysicalActivity.WALKING, walking_eq
-
-def format_energy_eq_electric_vehicle(energy: Quantity) -> Quantity:
-    energy = energy.to("kWh")
-    ev_eq = energy / EV_ENERGY_EQ
-    if ev_eq < q("1 km"):
-        ev_eq = ev_eq.to("meter")
-    return ev_eq
-
-def format_gwp_eq_streaming(gwp: Quantity) -> Quantity:
-    gwp = gwp.to("kgCO2eq")
-    streaming_eq = gwp * STREAMING_GWP_EQ
-    if streaming_eq < q("1 h"):
-        streaming_eq = streaming_eq.to("min")
-    if streaming_eq < q("1 min"):
-        streaming_eq = streaming_eq.to("s")
-    return streaming_eq
-
-def format_energy_eq_electricity_production(energy: Quantity) -> tuple[EnergyProduction, Quantity]:
-    electricity_eq = energy * ONE_PERCENT_WORLD_POPULATION * DAYS_IN_YEAR
-    electricity_eq = electricity_eq.to("TWh")
-    if electricity_eq > YEARLY_NUCLEAR_ENERGY_EQ:
-        return EnergyProduction.NUCLEAR, electricity_eq / YEARLY_NUCLEAR_ENERGY_EQ
-    electricity_eq = electricity_eq.to("GWh")
-    return EnergyProduction.WIND, electricity_eq / YEARLY_WIND_ENERGY_EQ
-
-
-def format_energy_eq_electricity_consumption_ireland(energy: Quantity) -> Quantity:
-    electricity_eq = energy * ONE_PERCENT_WORLD_POPULATION * DAYS_IN_YEAR
-    electricity_eq = electricity_eq.to("TWh")
-    return electricity_eq / YEARLY_IRELAND_ELECTRICITY_CONSUMPTION
-
-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

uv.lock

@@ -1,5 +1,4 @@
 version = 1
-revision = 1
 requires-python = ">=3.12"
 
 [[package]]