Skip to content

Command-R / Command-R+ + Granite 3: RAG-Native + Citation FT + Enterprise Tier

Cohere Command-R (35B) / Command-R+ (104B) — RAG-tuned baseline, native citation token training. IBM Granite 3 — Apache 2.0 enterprise tier, governance-focused. RAG-FT dataset format, citation accuracy measurement, tool-calling, Command-R+ QLoRA recipe on 4×H100 80GB.

Şükrü Yusuf KAYA
28 min read
Advanced
Command-R / Command-R+ + Granite 3: RAG-Native + Citation FT + Enterprise Tier

1. Command-R Ailesi — Cohere#

ModelParamsContextLisansRAG-Native
Command-R35B128KCC-BY-NC (research)✅
Command-R+104B128KCC-BY-NC✅
Command-R7B7B128KCC-BY-NC✅
RAG-Native ne demek? Pre-train + SFT'de citation token format öğretilmiş:
<results>
Document [1]: İstanbul nüfusu 15 milyondur.
Document [2]: Bursa nüfusu 3 milyondur.
</results>

Soru: İki büyük şehrin nüfusu?
Cevap: İstanbul 15 milyon [1], Bursa 3 milyon [2].
Lisans dikkati: CC-BY-NC = research only. Commercial deploy için OCİ Endpoints veya Bedrock üzerinden.

2. IBM Granite 3 — Enterprise Tier#

ModelParamsLisansNotlar
Granite 3.1 2B2BApache 2.0edge / cost-effective
Granite 3.1 8B8BApache 2.0general
Granite 3.1 MoE 1B (3B active)3B activeApache 2.0efficient
Granite 3.1 MoE 3B (8B active)8B activeApache 2.0balanced
IBM'in pozisyonu: Enterprise governance, Apache 2.0 (kısıtsız), audit trail için curated data.
Cookbook tavsiyesi: Enterprise compliance ihtiyacın varsa (telecom, banking, healthcare) — Granite 3.1 8B + custom domain FT.
✅ Teslim
  1. Command-R7B (open weights) ile mini RAG-FT lab. 2) Citation accuracy ölç (citation/[1] grep). 3) Sonraki ders: 4.10 — Hybrid SSM (Falcon-Mamba, Zamba2).

Yorumlar & Soru-Cevap

(0)
Yorum yazmak için giriş yap.
Yorumlar yükleniyor...

Related Content

Part 0 — Engineering Foundations

Welcome to the Fine-Tuning Cookbook: System, Stage Taxonomy, and the Reproducibility Contract

Start Learning
Part 0 — Engineering Foundations

Reproducibility Stack: Seeds, cuDNN Flags, and Deterministic CUDA — End the 'Works on My Machine' Problem

Start Learning
Part 0 — Engineering Foundations

Environment Pinning: uv + pyproject.toml, CUDA Version Matrix, and Container Recipes

Start Learning