Llama-3.1-8B-Instruct lineage
According to the model card, it is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct, with hub tags listing SFT, TRL, and generated_from_trainer markers.
Open Source Model Profile · barc0
Llama-3.1-ARC-Heavy-Induction-8B
Llama-3.1-ARC-Heavy-Induction-8B is an 8.03B-parameter Llama fine-tune from barc0. According to the model card, it adapts Llama-3.1-8B-Instruct for ARC grid-induction prompts.
- Publisher
- barc0
- Task
- text-generation
- Model type
- llama
- License
- llama3.1
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Llama-3.1-ARC-Heavy-Induction-8B is published by barc0 as a llama-based text-generation fine-tune. The captured configuration identifies LlamaForCausalLM and Safetensors metadata reports 8030261248 parameters. According to the model card, it fine-tunes Meta-Llama-3.1-8B-Instruct for ARC induction work, with card data recording llama3.1.
Recorded capabilities
ARC induction formatting
According to the model card, ARC public evaluation problems are converted into input-output grid prompts that ask for output-grid prediction from reference pairs.
Reported 0.2765 validation loss
According to the model card, the evaluation set reports loss 0.2765 after 2 epochs and 2956 steps, with epoch-1 validation at 0.2865.
Documented hyperparameters
According to the model card, training lists learning rate 1e-05, seed 42, Adam betas 0.9 and 0.999, cosine scheduling, 0.1 warmup ratio, and multi-GPU execution on 8 devices.
Llama3.1 licensing
Card data records llama3.1 for this repository.
Use cases in the source record
- ARC-style grid-induction experiments that use the documented input-output grid prompt conversion.
- Llama-3.1 instruct-template conversational workflows, as the card states it follows that template.
Limitations and unknowns
- According to the model card, the training dataset field shows None, so the fine-tuning data source is unclear from this record.
- The only reported result is publisher-listed training and validation loss; no independent benchmark scores were extracted.
- Provider state is historical snapshot data, not independently refreshed current availability.
- No context-window value was extracted from this record.
Source and provenance
Source: barc0/Llama-3.1-ARC-Heavy-Induction-8B
Captured: Unknown. Processed: 2026-09-07T19:34:40.775885+00:00.
l3.1-8b-inst-fft-induction-barc-heavy-200k-lr1e-5-ep2 This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the None dataset. It achieves the following results on the evaluation set: Loss: 0.2765 prompt example We follow Llama-3.1 instruct template. For example, the ARC public evaluation problem 62ab2642 is converted to [{"role": "system", "content": "You are a world-class puzzle solver with exceptional pattern recognition skills and expertise in Python programming. Your task is to analyze puzzles and provide Python solutions."}, {"role": "user", "content": "Given input-output grid pairs as reference example…
F001F002F003F004F005F006F007F009F010F011F012F013F014F015