Llama-3-70B-Instruct lineage
According to the model card, it builds on Meta Llama-3-70B-Instruct with expanded context work from Gradient AI.
Open Source Model Profile · xiangxinai
Xiangxin-2XL-Chat-1048k-Chinese-Llama3-70B
Xiangxin-2XL-Chat-1048k-Chinese-Llama3-70B is a 70.55B-parameter Llama-3 chat model from xiangxinai. According to the model card, it offers 1048k context with ORPO Chinese alignment.
- Publisher
- xiangxinai
- Task
- text-generation
- Model type
- llama
- License
- llama3
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Xiangxin-2XL-Chat-1048k-Chinese-Llama3-70B is published by xiangxinai as a Llama-based text-generation chat model. The captured configuration identifies LlamaForCausalLM and Safetensors metadata reports 70553706496 parameters. According to the model card, it is based on Meta Llama-3-70B-Instruct with Gradient AI expanded context and ORPO Chinese value alignment, with card data recording llama3.
Recorded capabilities
1048k Chinese chat context
According to the model card, context reaches 1048k, described as up to 1 million words, with stronger Chinese capability.
ORPO value alignment
According to the model card, training used ORPO on a proprietary Chinese value-aligned dataset.
Reported 70.22 average
According to the model card, it averaged 70.22 across eight listed benchmarks, above the compared Gradient 1048k model.
400GB inference note
According to the model card, running inference with Transformers requires approximately 400GB of GPU memory.
Use cases in the source record
- Long-context Chinese and English conversational work using the publisher Transformers pipeline and chat-template flow.
- Chinese-value-aligned chat experiments where the publisher describes stronger Chinese capability from ORPO training.
Limitations and unknowns
- According to the model card, benchmark and training-data claims come from the publisher and have not been independently verified by Ethen.
- According to the model card, the alignment dataset cannot be publicly disclosed, limiting independent review.
- Provider state is historical snapshot data, not independently refreshed current availability.
- No independent evaluation results were extracted beyond the publisher-reported benchmark table.
Source and provenance
Source: xiangxinai/Xiangxin-2XL-Chat-1048k-Chinese-Llama3-70B
Captured: Unknown. Processed: 2026-09-07T19:35:01.029335+00:00.
Xiangxin-2XL-Chat-1048k 我们提供私有化模型训练服务，如果您需要训练行业模型、领域模型或者私有模型，请联系我们: wanglei@xiangxinai.cn We offer customized model training services. If you need to train industry-specific models, domain-specific models, or private models, please contact us at: wanglei@xiangxinai.cn . 模型介绍/Introduction Xiangxin-2XL-Chat-1048k是 象信AI 基于Meta Llama-3-70B-Instruct模型和 Gradient AI的扩充上下文的工作 ，利用自行研发的中文价值观对齐数据集进行ORPO训练而形成的Chat模型。该模型具备更强的中文能力和中文价值观，其上下文长度达到100万字。在模型性能方面，该模型在ARC、HellaSwag、MMLU、TruthfulQA_mc2、Winogrande、GSM8K_flex、CMMLU、CEVAL-VALID等八项测评中，取得了平均分70.22分的成绩，超过了Gradientai-Llama-3-70B-Instruct-Gradient-1048k。我们的训练数据并不包含任何测评数据集。 Xiangxin-2XL-Chat-104…
F001F002F003F004F005F006F007F010F012F013F014F015