README.md
8.7 KB · 166 lines · markdown Raw
1 ---
2 license: apache-2.0
3 language:
4 - en
5 base_model:
6 - google/gemma-4-31B-it
7 tags:
8 - telecom
9 - telecommunications
10 - gsma
11 - at-t
12 - microsoft
13 - dell
14 - amd
15 - red-hat
16 - open-telco-ai
17 - rag
18 - instruction-following
19 - tool-calling
20 - domain-adaptation
21 - post-trained
22 - transformers
23 pipeline_tag: text-generation
24 ---
25
26 # OTel-2.0-LLM-31B-IT
27
28 > **Checkpoint update notice:** The current checkpoint is expected to be updated within the next few hours. After that initial refresh, OTel 2.0 checkpoints are expected to continue receiving weekly weight updates. For reproducible evaluation or production deployment, pin a specific model revision, checkpoint hash, or release tag.
29
30 **OTel-2.0-LLM-31B-IT** is a telecom-specialized instruction model post-trained from **Gemma 4 31B-IT** on approximately **440 billion telecom training tokens**. It is the first release in the OTel 2.0 family and is designed to support telco-grade AI workflows across network operations, standards interpretation, product development, network configuration assistance, RAG, and telecom-specific question answering.
31
32 OTel 2.0 extends the original OTel effort from a RAG-oriented telecom fine-tuning release into a larger domain-adapted training program. The model was trained from a much larger standards and telecom corpus, with new data preparation coverage for agentic tool calling, direct telecom QnA, instruction following, abstention, RAG, and base-model-style training data.
33
34 ## Release Status
35
36 OTel 2.0 models are expected to continue training after release, and weights may be updated weekly. For reproducible evaluation or production deployment, pin a specific model revision, checkpoint hash, or release tag rather than relying on the floating latest weights.
37
38 ## Model Details
39
40 | Attribute | Value |
41 |---|---|
42 | Base model | Gemma 4 31B-IT |
43 | Parameters | 31B |
44 | Model family | OTel 2.0 |
45 | Training method | Telecom domain post-training / instruction tuning |
46 | Raw telecom corpus | ~15B tokens from GSMA through Open Telco AI |
47 | Processed data volume | ~1T tokens processed using Red Hat SDG Hub |
48 | Training tokens | ~440B |
49 | Compute for data processing | ~530 GPUs through Azure Managed Compute, primarily MI300X |
50 | Model training infrastructure | On-prem AMD MI355X with Dell servers |
51 | Day 0 inference availability | Microsoft Foundry, Featherless AI, and Red Hat |
52 | Language | English |
53
54
55 ## Model Lineage
56
57 `Gemma 4 31B-IT` -> OTel 2.0 telecom data processing and post-training -> `OTel-2.0-LLM-31B-IT`
58
59 ## What Changed From OTel 1.0
60
61 | Category | Metric | OTel 1.0 | OTel 2.0 |
62 |---|---|---:|---:|
63 | Tokens | Raw documents | ~600M | ~15B |
64 | Tokens | Training | ~1B | ~440B |
65 | Data preparation | Agentic tool calling | No | Yes |
66 | Data preparation | Knowledge / fact direct QnA | No | Yes |
67 | Data preparation | RAG | Yes | Yes |
68 | Data preparation | Abstention | Yes | Yes |
69 | Data preparation | Instruction following | No | Yes |
70 | Data preparation | Base model training set | No | Yes |
71
72 Relative to OTel 1.0, OTel 2.0 increases raw source coverage by roughly 25x and training-token volume by roughly 440x. More importantly, the data mixture is broader: OTel 1.0 focused primarily on context-grounded RAG and abstention, while OTel 2.0 adds direct telecom knowledge QnA, instruction following, tool-use-oriented examples, and base-model-style telecom training data.
73
74 ## Training Data
75
76 The starting corpus contains approximately **15 billion raw tokens** provided by GSMA through Open Telco AI. The corpus includes telecom standards and technical material from seven standards development and industry organizations:
77
78 | Source family | Examples |
79 |---|---|
80 | Cellular standards | 3GPP |
81 | Telecom standards and specifications | ETSI, ITU |
82 | Industry and operator materials | GSMA |
83 | Network API specifications | CAMARA |
84 | Open RAN specifications | O-RAN |
85 | Telecom business and operations frameworks | TM Forum |
86
87 The raw corpus was processed into approximately **1 trillion tokens** using **Red Hat SDG Hub** on **Azure Managed Compute**, using approximately **530 GPUs**, primarily **AMD MI300X**. From this processed pool, approximately **440 billion training tokens** were generated for OTel 2.0 post-training.
88
89 ## Data Preparation Capabilities
90
91 OTel 2.0 includes a broader supervised and synthetic-data mixture than OTel 1.0:
92
93 | Capability | Description |
94 |---|---|
95 | RAG | Context-grounded telecom answer generation from retrieved standards and technical documents |
96 | Abstention | Training examples that teach the model to avoid answering when context is missing, irrelevant, or insufficient |
97 | Direct telecom QnA | Knowledge and factual question-answer pairs for standards, protocols, services, and network concepts |
98 | Instruction following | Telecom-specific instructions for analysis, summarization, transformation, and operational support tasks |
99 | Agentic tool calling | Examples intended to improve tool-use behavior in telecom AI workflows |
100
101 ## Intended Use
102
103 OTel-2.0-LLM-31B-IT is intended for telecom-focused applications where domain knowledge, standards familiarity, and deployment control matter. Suitable use cases include:
104
105 - Retrieval-Augmented Generation over telecom standards, specifications, and internal technical documentation.
106 - Standards interpretation and summarization for 3GPP, ETSI, GSMA, CAMARA, ITU, O-RAN, and TM Forum materials.
107 - Product development, network configuration assistance, and engineering support tasks.
108 - Telecom-specific direct QnA where the model has been separately evaluated for the target benchmark or application.
109 - Agentic workflows where a larger system provides verified tools, retrieval, validation, and audit logging.
110
111 For high-impact operational use, the model should be deployed with retrieval, source attribution, validation checks, and human review appropriate to the system risk.
112
113 ## Limitations and Responsible Use
114
115 - OTel 2.0 is domain-specific to telecommunications and should not be treated as a general-purpose model for unrelated fields.
116 - Telecom standards evolve over time; answers should be checked against the relevant document version and release.
117 - Agentic tool calling requires an external tool runtime, tool schema validation, permissioning, logging, and safeguards. The model alone does not guarantee safe tool execution.
118 - RAG quality depends on document ingestion, chunking, retrieval, reranking, prompt design, and source freshness.
119 - Direct QnA behavior should be evaluated separately from RAG behavior; strong performance in one setting does not imply strong performance in the other.
120
121 ## Related Models
122
123 - [OTel LLM Collection](https://huggingface.co/collections/farbodtavakkoli/otel-llm)
124 - [OTel Embedding Collection](https://huggingface.co/collections/farbodtavakkoli/otel-embedding)
125 - [OTel Reranker Collection](https://huggingface.co/collections/farbodtavakkoli/otel-reranker)
126
127 ## Project Resources
128
129 - Project page: https://huggingface.co/farbodtavakkoli
130 - Code: https://github.com/farbodtavakkoli/OTel
131 - Media coverage list: https://github.com/farbodtavakkoli/OTel/blob/main/docs/media_coverage.md
132
133 ## Contributors and Organizations
134
135 Contributors to the OTel 2.0 release and supporting infrastructure include:
136
137 | Organization | Contributors |
138 |---|---|
139 | AT&T | Farbod Tavakkoli, Jorden Terrazas, Roderic Paulk, Sharath Japa, Tzvi Chumash, Miguel Armenta, Pavan Tagirisa, Kostikey Mustakas, Mark Austin, Andy Markus |
140 | MLCommons | Gregory Diamos, David Kanter, Kenneth Church |
141 | Microsoft | Chunyu Li, Gulsimo Osimi, Inayat Wali, Manoj Bableshwar, Marie-Louise Onga Nana, Naomi Moneypenny, Osi Otugo, Rahul Kumar, SeokJin Han, Steve Sweetman, Trinidad Salazar, Ven Kumar, Vivek Ramaswamy |
142 | AMD | Alexander Finn, Andy Allred, Andrey Ivannikov, Antti-Ville Suni, Mark van Heeswijk, Kumaran Siva, Curt Wortman, Mehrvash Poole, Eric Lynn |
143 | Dell | Brian Sweeney, Suzanne Randall, Patrick Allen, Matt Currie, Justin Wilson, Jason Kane, Keith Napoleon, Sarah Lake, Mike Hess, Randy Tornes |
144 | GSMA | Louis Powell, Zeinab Nezami, Enrique Molero |
145 | Red Hat | Aditi Saluja, William Caban, Shivchander Sudalairaj, Kai Xu, Ravi Sharma, Hanen Garcia, Joe Crispo, Eshwar Sivaramakrishnan |
146 | Pleias | Anastasia Stasenko, Pierre-Carl Langlais, Mohamed Chenene, Carlos Rosas, Yannick Detrois |
147
148 Organizations involved in the OTel 2.0 release and supporting infrastructure include AT&T, MLCommons, Microsoft, AMD, Dell, GSMA, Red Hat, and Pleias.
149
150 ## Citation
151
152 ```bibtex
153 @misc{otel_2_models_2026,
154 title = {OTel 2.0: Open Telco AI Datasets, Benchmarks, and Models},
155 author = {Tavakkoli, Farbod and others},
156 year = {2026},
157 note = {Open Telco (OTel 2.0) model release},
158 url = {https://huggingface.co/farbodtavakkoli},
159 organization = {AT\&T, MLCommons, Microsoft, AMD, Dell, GSMA, Red Hat, Pleias}
160 }
161 ```
162
163 ## Contact
164
165 For technical questions, contact farbod.tavakkoli@att.com or farbodtavakoli@gmail.com.
166