updates

Author: AseelOmer

Commercial_Models/models.md

@@ -0,0 +1,226 @@
+# Model 1: GPT-4 (OpenAI)
+
+## Model Name & Provider
+
+**GPT-4**, developed by **OpenAI**
+
+### Hosting / Deployment
+
+Hosted via **Microsoft Azure OpenAI Service**.
+
+Source: [Azure blog – Introducing GPT-4 in Azure OpenAI Service][azure-blog]
+
+Cloud infrastructure with global data-centers; exact hosting regions not
+publicly specified.
+
+### Estimated Energy (Inference)
+
+Published or estimated per-query energy values vary between studies.
+Representative numbers include:
+**Epoch AI (2024):** $\approx 0.3$ Wh ($0.0003$ kWh) per typical ChatGPT/GPT-4
+query under assumed workloads.
+
+Source: [Epoch AI – How Much Energy Does ChatGPT Use?][epoch-ai]
+
+Other independent analysts: $\sim 0.3 - 1.8$ Wh ($0.0003 - 0.0018$ kWh) per query,
+depending on prompt length, token output, and GPU hardware.
+Sources: [The Carbon Footprint of ChatGPT], media analyses.
+
+**Caveat:** OpenAI does not publish per-query energy data. All estimates
+depend heavily on assumptions such as:
+
+* Hardware type (GPU vs TPU)
+* Power Usage Effectiveness (PUE)
+* Data-center region and carbon intensity
+* Prompt and token length
+
+### Training Energy (GPT-4)
+
+Some analyses extrapolate GPT-4’s training energy from its model size and
+compute budget:
+Example estimate:
+$\approx 51,772,500 – 62,318,750$ kWh ($\approx 51,773 – 62,319$ MWh)
+consumed for full-scale training.
+
+Source: [The Carbon Footprint of ChatGPT][sustainability-numbers]
+These are indirect estimates, not official OpenAI disclosures.
+
+### Water Usage (GPT-4)
+
+Official data are unavailable, but public remarks and media analyses give
+approximate indicators:
+
+A single ChatGPT query may indirectly consume $\sim 0.5$ liters of water,
+depending on data-center cooling.
+
+For instance, generating a 100-word email may consume $\sim 0.14$ kWh energy
+and $0.519$ L of water.
+
+Source: [The Verge – Sam Altman on ChatGPT Energy and Water Use][verge-gpt]
+
+### PUE / CI Context (GPT-4)
+
+To estimate total environmental footprint, studies multiply compute energy by:
+
+* **PUE** (Power Usage Effectiveness) – ratio of total facility power to IT
+  equipment power.
+* **CI** (Carbon Intensity) – kg $\text{CO}_2\text{e}$ emitted per kWh of electricity
+  generated.
+
+Example assumptions from literature:
+
+* **PUE** $\approx 1.1 – 1.3$, typical for **hyperscale** Azure data-centers.
+* **CI** $\approx 0.3–0.4$ kg $\text{CO}_2\text{e}$/kWh,
+ depending on the region’s energy mix.
+
+---
+
+## Model 2: Claude Haiku (Anthropic)
+
+### Model Name & Provider (Claude Haiku)
+
+**Claude 3 Haiku**, developed by **Anthropic**
+
+### Model Description
+
+Claude 3 Haiku is part of Anthropic’s Claude 3 model family, released in
+March 2024. It is the smallest and fastest model in the lineup (Haiku,
+Sonnet, Opus) and is designed for low-latency, energy-efficient inference.
+Haiku is optimized for lightweight commercial use cases, including chat
+applications, summarization, and enterprise automation.
+
+Source: [Anthropic Blog – Claude 3 Technical Overview][anthropic-blog]
+
+### Hosting / Deployment (Claude Haiku)
+
+Claude 3 Haiku is hosted through Anthropic’s own API and via **Amazon Bedrock**
+(**AWS** cloud).
+These centers typically maintain a **Power Usage Effectiveness (PUE) of $\sim 1.2$**.
+
+Sources: [AWS Bedrock Claude Integration], [AWS Sustainability Report 2024][aws-report]
+
+### Estimated Energy (Haiku Inference)
+
+Anthropic does not publicly disclose per-query energy data.
+Independent analysts estimate inference use around:
+$\sim 0.05–0.1$ Wh ($0.00005–0.0001$ kWh) per query, depending on token count and
+GPU efficiency.
+
+Claude 3 Haiku is reported to be **$\sim 5\times$ faster and more efficient** than
+larger Claude 3 models (Sonnet or Opus).
+
+Sources: [Epoch AI – Energy Use of AI Models][epoch-ai], [Anthropic Claude 3 Announcement]
+
+### Training Energy (Claude Haiku)
+
+Claude 3 models are trained on **GPU clusters** (NVIDIA A100/H100) primarily
+hosted on AWS infrastructure.
+For models in the $10–30$B parameter range, training energy is typically
+**$3,000–10,000$ MWh**.
+
+Sources: [Epoch AI – AI Training Compute & Energy Scaling],
+ [Anthropic Responsible Scaling Policy][anthropic-policy]
+
+### Water Usage (Claude Haiku)
+
+Anthropic has not published specific water consumption figures for the
+Claude 3 family.
+Cooling water use is managed under **AWS’s sustainability strategy**.
+AWS data centers in cooler regions use air cooling to reduce water footprint,
+while others recycle water on-site.
+
+Sources: [AWS Water Stewardship Report][aws-water], [Anthropic Sustainability Commitments]
+
+### PUE / CI Context (Claude Haiku)
+
+* **AWS’s average PUE:** $\sim 1.2$ (accounts for cooling and power delivery losses).
+* **Carbon intensity (CI):
+* ** $\sim 0–0.2$ kg $\text{CO}_2\text{e}$/kWh, depending on regional
+  renewable mix.
+AWS aims for **$100\%$ renewable energy by 2025**, lowering emissions over time.
+
+Sources: [AWS Global Infrastructure Efficiency Data],
+ [Anthropic Responsible Scaling Policy][anthropic-policy]
+
+---
+
+## Model 3: Gemini Nano (Google)
+
+### Model Name & Provider (Gemini Nano)
+
+**Gemini Nano**, developed by **Google DeepMind**.
+Smallest member of the Gemini model family (Nano, Pro, Ultra).
+
+### Hosting / Deployment (Gemini Nano)
+
+Runs **on-device** through **Android AICore** (subsystem introduced in 2023).
+Designed for mobile hardware such as Pixel 8 Pro and Pixel 9 series.
+**Reduces energy use by eliminating cloud compute and network transmission.**
+
+Sources: [Google AI Blog – Introducing Gemini][google-blog],
+[Android Developers – Gemini Nano Overview][android-dev],
+[The Verge – Gemini Nano arrives on Pixel 8 Pro][verge-gemini]
+
+### Estimated Energy (Nano Inference)
+
+No official values published for per-query energy.
+Independent device benchmarks indicate $\approx 0.01$ Wh ($0.00001$ kWh) per query.
+This is **$10–30\times$ more efficient** than cloud-hosted models such as GPT-4.
+
+Sources: [Google Pixel AI Benchmarks (2024)],
+[Epoch AI – How Much Energy Does ChatGPT Use][epoch-ai]
+
+### Training Energy (Gemini Nano)
+
+Gemini Nano was **distilled** from larger Gemini models trained on **TPU v5e**
+clusters.
+Training energy for Nano $\approx 200 – 1,200$ MWh ($\approx 1–5\%$ of Gemini Ultra’s
+training compute).
+
+Sources: [Google Research – Efficient TPU Training (2024)],
+ [Google Cloud Sustainability Report (2024)]
+
+### Water Usage (Gemini Nano)
+
+Inference uses **no data-center water** since it runs locally on devices.
+Training used Google data centers with **Water Usage Effectiveness (WUE)**
+$\approx 0.18$ L/kWh.
+Google targets **net-positive water impact by 2030**.
+
+Sources: [Google Environmental Report (2024)],
+[Bloomberg – Google AI Water Consumption (2024)]
+
+### PUE / CI Context (Gemini Nano)
+
+* **Google Data Centers** report average **PUE** $\approx 1.10–1.12$.
+* **Carbon Intensity (CI)**
+* $\approx 0.15$ kg $\text{CO}_2\text{e}$ / kWh due to $70\%+$ renewable
+  energy mix.
+* **On-device execution** uses $< 5$ W of mobile power per inference.
+
+Sources: [Google Data Center Efficiency Overview (2024)],
+[Google TPU v5e Efficiency Blog (2024)]
+
+---
+[azure-blog]: https://azure.microsoft.com/en-us/blog/introducing-gpt4-in-azure-openai-service/
+[epoch-ai]: https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use
+[sustainability-numbers]: https://www.sustainabilitybynumbers.com/p/carbon-footprint-chatgpt
+[verge-gpt]: https://www.theverge.com/2023/1/18/energy-water-chatgpt
+[anthropic-blog]: https://www.anthropic.com/blog/claude3-overview
+[aws-report]: https://aws.amazon.com/about-aws/sustainability/
+[anthropic-policy]: https://www.anthropic.com/responsible-scaling
+[aws-water]: https://aws.amazon.com/about-aws/sustainability/#water
+[google-blog]: https://blog.google/technology/ai/google-gemini-ai/
+[android-dev]: https://developer.android.com/ai/gemini-nano
+[verge-gemini]: https://www.theverge.com/2023/12/6/23990823/google-gemini-ai-models-nano-pro-ultra
+[AWS Bedrock Claude Integration]: https://aws.amazon.com/bedrock/
+[Anthropic Claude 3 Announcement]: https://www.anthropic.com/news/claude-3-models
+[Epoch AI – AI Training Compute & Energy Scaling]: https://epoch.ai/gradient-updates/ai-training-compute-energy-scaling
+[Anthropic Sustainability Commitments]: https://www.anthropic.com/sustainability
+[AWS Global Infrastructure Efficiency Data]: https://aws.amazon.com/about-aws/sustainability/
+[Google Pixel AI Benchmarks (2024)]: https://ai.google/discover/pixel-ai/
+[Google Research – Efficient TPU Training (2024)]: https://arxiv.org/abs/2408.15734
+[Google Cloud Sustainability Report (2024)]: https://sustainability.google/reports/environmental-report-2024/
+[Bloomberg – Google AI Water Consumption (2024)]: https://www.bloomberg.com/news/articles/2024-02-13/google-ai-water-consumption-analysis
+[Google Data Center Efficiency Overview (2024)]: https://cloud.google.com/sustainability/data-centers
+[Google TPU v5e Efficiency Blog (2024)]: https://cloud.google.com/blog/products/ai-machine-learning/introducing-tpu-v5e