Claude-token training claim
According to the model card, Llama 3.1 8B Instruct was trained on 9,000,000 Claude Opus and Sonnet tokens.
Open Source Model Profile · Undi95
Meta-Llama-3.1-8B-Claude
Meta-Llama-3.1-8B-Claude is an 8.03B-parameter Llama text-generation model from Undi95. According to the model card, it trains Llama 3.1 8B Instruct on 9 million Claude Opus/Sonnet tokens with a Llama3 prompt template.
- Publisher
- Undi95
- Task
- text-generation
- Model type
- llama
- License
- Unknown
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Meta-Llama-3.1-8B-Claude is published by Undi95 as a Llama text-generation model. The captured configuration identifies LlamaForCausalLM with model type llama, and Safetensors metadata reports 8,030,261,248 parameters. According to the model card, it trains Llama 3.1 8B Instruct on nine million Claude Opus and Sonnet tokens over two epochs.
Recorded capabilities
Documented training setup
According to the model card, training ran for two epochs over six hours on eight H100 NVL GPUs.
Listed chat datasets
According to the model card, training data spans seven Claude-style chat datasets plus a Llama3 prompt template.
Claude-style system prompt
According to the model card, the system prompt presents the assistant as Claude, Opus 20240306, created by Anthropic.
Use cases in the source record
- Claude-style assistant chat generation using the Claude-token training and Claude Opus system prompt described in the model card.
- Conversational text generation with the Llama3 prompt template documented in the model card.
Limitations and unknowns
- No license value was extracted from this record.
- No evaluation results were extracted from this record.
- No context-window value was extracted from this record.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: Undi95/Meta-Llama-3.1-8B-Claude
Captured: Unknown. Processed: 2026-09-07T19:34:38.157680+00:00.
Details Llama 3.1 8B Instruct trained on 9 000 000 Claude Opus/Sonnet tokens. Trained on 2 epoch, for 6 hours on 8x H100 NVL. Training data used Norquinal/claude_multiround_chat_30k kalomaze/Opus_Instruct_3k mahiatlinux/Claude3-Opus-Instruct-ShareGPT-14k kalomaze/Opus_Instruct_25k meseca/opus-instruct-9k Gryphe/Sonnet3.5-SlimOrcaDedupCleaned Gryphe/Opus-WritingPrompts Prompt template: Llama3 <|begin_of_text|><|start_header_id|>system<|end_header_id|> {system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|> {input}<|eot_id|><|start_header_id|>assistant<|end_header_id|> {output}<|eot_id|> System Prompt Claude Opus 20240306 Th…
F001F002F003F004F005F006F008F009F010F011F012F013