Falcon H1 7B Instruct

1. What it is

Falcon H1 7B Instruct (`tiiuae/Falcon-H1-7B-Instruct`) is a 7B instruction-tuned model from TII UAE.

2. What it does

It provides balanced Arabic and English chat/instruction performance with strong throughput for API serving.

3. How it compares

• Versus ALLaM 7B: Falcon H1 can be a good balance choice when you need bilingual coverage with strong speed.
• Versus JAIS 13B: lower memory cost and generally lower latency, but JAIS may outperform on high-precision Arabic enterprise prompts.

4. Best for on DCP

• Bilingual assistants with high request volume
• Regional support bots
• Arabic-first copilots where 24 GB GPUs are available

5. Hardware requirements on DCP

• DCP floor: `min_vram_gb: 24` (Tier A)
• Recommended providers: RTX 4090, A100 40GB+
• Template: `vllm-serve`

6. How to run on DCP

1. Use `job_type: "vllm_serve"` with `params.model: "tiiuae/Falcon-H1-7B-Instruct"`.
2. Set `gpu_requirements.min_vram_gb` to `24`.
3. Consume the OpenAI-compatible endpoint returned by job output.

7. Licensing and commercial-use notes

Falcon models use TII model license terms. Validate your intended commercial use (hosting, redistribution, derivatives) against the card license text.

Sources:

• https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct
• /home/node/dc1-platform/infra/config/arabic-portfolio.json
• /home/node/dc1-platform/backend/src/db.js