Indonesia-focused instruct tuning
According to the model card, Sahabat-AI models were pretrained and instruct-tuned for Indonesia, co-initiated by GoTo and Indosat and developed by GoTo with AI Singapore.
Open Source Model Profile · GoToCompany
Llama-Sahabat-AI-v2-70B-IT
Llama-Sahabat-AI-v2-70B-IT is a 70.55B-parameter Llama-family instruct model from GoToCompany. According to the model card, it was pretrained and instruct-tuned for Indonesia.
- Publisher
- GoToCompany
- Task
- text-generation
- Model type
- llama
- License
- llama3.1
- Library
- transformers
- Publication status
- Approved for indexing
Model overview
Llama-Sahabat-AI-v2-70B-IT is published by GoToCompany as a text-generation model. The captured configuration identifies LlamaForCausalLM with model type llama, and Safetensors metadata reports about 70.55B parameters under llama3.1. According to the model card, it belongs to Sahabat-AI, a collection pretrained and instruct-tuned for Indonesia.
Recorded capabilities
Six documented languages
According to the model card, the model supports English, Indonesian, Javanese, Sundanese, Batak Toba, and Balinese with a 128k context length.
Documented multilingual evaluation
According to the model card, evaluation covered QA, sentiment, toxicity, bidirectional translation, summarization, causal reasoning, NLI, LINDSEA, and SEA-IFEval using zero-shot native prompts.
Documented deployment setups
According to the model card, recommended setups include four NVIDIA L40s or two NVIDIA H100 GPUs with bfloat16 Transformers pipeline loading.
Use cases in the source record
- Indonesian and regional-language QA, sentiment, toxicity detection, translation, summarization, and NLI workflows named in the card's evaluation list.
- Instruction-following conversational use with constraint adherence and language-correctness checks described under SEA-IFEval.
Limitations and unknowns
- No evaluation scores were extracted from this record; task coverage is described without published figures.
- No VRAM or latency figures were extracted beyond the two recommended GPU configurations.
- Provider state is historical snapshot data, not independently refreshed current availability.
- Capability, language, context, and hardware statements are publisher claims and were not independently verified.
Source and provenance
Source: GoToCompany/Llama-Sahabat-AI-v2-70B-IT
Captured: Unknown. Processed: 2026-09-07T19:35:39.600258+00:00.
Llama-Sahabat-AI-v2-70B-IT Sahabat-AI is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for Indonesia. Co-initiated by: PT GoTo Gojek Tokopedia Tbk, Indosat Ooredoo Hutchison Developed by: PT GoTo Gojek Tokopedia Tbk, AI Singapore Model type: Decoder Languages supported: English, Indonesian, Javanese, Sundanese, Batak Toba, Balinese License: Llama 3.1 Community License Model Details Model Description For tokenisation, the model employs the default tokenizer used in Llama 3.1 70B Instruct. The model has a context length of 128k. Benchmark Performance We evaluated Llama-Sahabat-AI-v2-70B-IT…
F001F002F003F004F005F006F007F010F011F012F013F014F015F016F017F018