Resourcing

Recommended CPU / RAM / Disk to run Hymalaia

Running Locally

When running locally through Docker, we recommend making at least 4 vCPU cores and 10GB of RAM available to Docker (16GB is preferred).
This can be configured in the Resources section of the Docker Desktop settings menu.

Single Cloud Instance

For small- to mid-scale deployments, setting everything up on a single instance (e.g., AWS EC2, GCP Compute Engine, Azure VM) via Docker Compose is the simplest way to get started.
Check out our EC2 deployment guide for a step-by-step walkthrough.

We recommend the following for a single instance:

• CPU: ≥ 4 vCPU cores (≥ 8 vCPU cores recommended for larger document volumes)
• Memory: ≥ 16 GB RAM
• Disk: ≥ 50 GB + ~2.5× the size of the indexed documents

💡 Note: Vespa does not allow writes when disk usage exceeds 75%, so always allocate extra disk space.

🧹 Tip: Old/unused Docker images can consume significant space. Run docker system prune --all regularly to free space.

⚠️ Compatibility: Vespa requires Haswell (2013) or later CPUs. For older CPUs, use the vespaengine/vespa-generic-intel-x86_64 image. This version is slower but compatible.

Kubernetes / AWS ECS

For deployments where each component gets its own resources, allocate at least:

Component	CPU	Memory
api_server	1	2Gi
background	2	8Gi
indexing_model_server	2	4Gi
inference_model_server	2	4Gi
postgres	2	2Gi
vespa	≥4	≥8Gi
nginx	250m	128Mi

⚠️ Important: The vespa component’s needs scale with the number of indexed documents.
For production, we recommend more than the minimum (e.g., 10 CPU, 20Gi+ RAM for 50GB of documents).

Minimum Node Size: 14 vCPU / 23GB RAM

How Resource Requirements Scale

The primary factor for scaling is the size of the indexed documents.

For every 1GB of raw text documents, expect:

• +3GB RAM
• +0.5 CPU core

Baseline (without document indexing):
4 vCPU and 16GB RAM

📘 A “1GB document” means raw text. PDFs with images may be large in file size, but the actual text may be small.

🧠 Dimensionality of embedding models and techniques like quantization can reduce requirements.

Example: 10GB of Text Documents

Index Component Requirements:

• CPU: 4 + 10 × 0.5 = 9 cores
• Memory: 16 + 10 × 3 = 46GB

Single Instance Total Recommendation:

• CPU: ≥ 13 cores
• Memory: ≥ 50GB

🖥️ Suggested instance types: m7g.4xlarge, c5.9xlarge

Kubernetes / AWS ECS Breakdown:

Component	CPU	Memory
api_server	1	2Gi
background	2	8Gi
indexing_model_server	2	4Gi
inference_model_server	2	4Gi
postgres	2	4Gi
vespa	10	34Gi

Total Recommended Node Size: 17 CPU / 66Gi RAM