DCP Model Catalog

English model docs for all supported DCP launch models and templates.

DCP models at a glance

Use this index to pick the right model before submitting a job with `POST /api/jobs/submit`.

For Arabic-first and bilingual launches, start with these model pages:

10 of 10 models

Best in class Needs improvementClick rows to compare

#	Model⇅	Params⇅	Language support⇅	License⇅	Recommended GPU tier on DCP⇅	DCP template⇅
🥇	[JAIS 13B Chat](./jais-13b)	13BBEST	Arabic + English	JAIS research/commercial terms	Tier A/B (`>=24 GB`)	`vllm-serve`
🥈	[Llama 3 8B Instruct](./llama-3-8b-instruct)	8B#2	Strong multilingual incl. Arabic	Llama 3 community license	Tier A (`>=16 GB`)	`vllm-serve`
🥉	[ALLaM 7B Instruct](./allam-7b)	7B#3	Arabic-first, bilingual	Custom ALLaM terms	Tier A (`>=24 GB`)	`vllm-serve`
#4	[Falcon H1 7B Instruct](./falcon-h1-7b)	7B	Arabic + English	TII Falcon license	Tier A (`>=24 GB`)	`vllm-serve`
#5	[Mistral 7B Instruct v0.2](./mistral-7b-instruct-v0-2)	7B	Strong bilingual instruction use	Apache-2.0	Tier A (`>=16 GB`)	`vllm-serve`
#6	[Phi-2](./phi-2)	2.7B	English-first, limited Arabic utility	MIT	Entry tier (`>=6 GB`)	`llm-inference`
#7	[TinyLlama 1.1B Chat](./tinyllama-1-1b-chat)	1.1B	Mostly English, usable bilingual prototyping	Apache-2.0	Entry tier (`>=4 GB`, prefer 8 GB)	`ollama`
#10	[BGE-M3 Embeddings](./bge-m3)	Embedding encoder	100+ languages incl. Arabic	MIT	Tier B (`>=8 GB`)	`arabic-embeddings`
#10	[BGE Reranker v2 M3](./bge-reranker-v2-m3)	Cross-encoder reranker	Multilingual incl. Arabic	MIT	Tier B (`>=8 GB`)	`arabic-reranker`
#10	[Stable Diffusion XL Base 1.0](./sdxl-base-1-0)	Diffusion checkpoint	Prompt-driven, multilingual text prompts	CreativeML Open RAIL++-M	Tier B (`>=16 GB`)	`stable-diffusion`

Tier A: high-priority serving GPUs (`>=16 GB`, typically RTX 3090/4090 and above)
Tier B: add-on serving workloads (`8-16 GB`)
Tier C: frontier/on-demand large models (`>=80 GB`)

These tiers follow `infra/config/arabic-portfolio.json` and DCP benchmark defaults in `backend/src/db.js`.