YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Private Model Card Draft โ€” v11

Prepared on: 2026-04-11

Status:

  • Draft for private artifact management only.
  • Not approved for public publication.

Model Identity

  • Display name: Telecom Expert Q&A โ€” v11 alpha candidate
  • Runtime model name: telecom-expert
  • Serving image validated for alpha: 10.127.251.228:32000/telecom-rag:canary-retrieval-audit-v11
  • Base model lineage: IBM Granite 3.3 2B Instruct derivative, fine-tuned internally
  • Deployment context: internal RAG-backed telecom engineering assistant

Overview

This artifact supports an internal telecom-domain engineering assistant that answers questions using retrieval grounding over public 3GPP and Cisco packet-core documentation. The model is intended to help engineers explain standards behavior, compare interfaces, review call-flow concepts, and retrieve relevant source material with citations.

The current v11 candidate is a serving-path hardening and retrieval-improvement milestone rather than a broad new model family. The validated improvement comes from the deployed retrieval/query path plus guardrails, not only from model weights in isolation.

Intended Use

Allowed use:

  • internal telecom standards and Cisco packet-core Q&A
  • explanation of 3GPP concepts, interfaces, and procedures
  • grounded citation-oriented assistance using retrieved documentation
  • engineer productivity support where a human reviews the answer before acting

Out Of Scope

Do not use this artifact for:

  • autonomous network configuration changes
  • direct execution of generated commands against production systems
  • legal, HR, finance, medical, customer-eligibility, or person-impacting decisions
  • answers that require authoritative source-of-record treatment without human verification

Validation Snapshot

Current exact-image candidate evidence:

  • Day 1 full benchmark on deployed v11 image:
    • 96/96 responses
    • response_rate = 1.0
    • timeouts_or_errors = 0
    • mean_duration = 52.84s
    • p95_duration = 115.56s
    • cli_safety_pass_rate = 1.0
  • Day 3 retrieval audit:
    • spec_hit_top_8 = 0.8769
    • request_errors = 0
  • Day 7 readiness:
    • GO

Safety And Controls

Controls currently relied on:

  • retrieval grounding over indexed public 3GPP and Cisco documentation
  • bounded generation settings and trimmed prompt context
  • CLI safety validation and refusal behavior when verified CLI evidence is absent
  • internal-only deployment and controlled alpha scope
  • stable rollback path preserved separately from the candidate

Known Limitations

  • Section-level citation exactness still lags spec-family retrieval exactness.
  • The system remains an internal alpha candidate, not a broad production release.
  • Output quality depends strongly on retrieval quality and runtime configuration.
  • Human validation remains mandatory for technical conclusions and CLI use.

Distribution Position

Recommended packaging position:

  • private artifact storage only
  • internal governance and reproducibility support
  • no public release at this stage

Open Items Before Private Registry Publication

  • suggested private repository slug: telecom-expert-v11-alpha-candidate
  • recommended primary upload payload: remote merged-HF directory at /mnt/data/models/granite-telecom-merged-hf/
  • optional runtime payload: remote GGUF artifact at /mnt/data/models/granite-telecom-q4km.gguf
  • attach sanitized evaluation summary and governance checklist
  • confirm Granite derivative redistribution obligations
  • confirm internal approval for storing derivative artifacts in the selected private repository
Downloads last month
10
GGUF
Model size
3B params
Architecture
granite
Hardware compatibility
Log In to add your hardware

We're not able to determine the quantization variants.

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support