metadata
datasets:
- HuggingFaceH4/ultrachat_200k
base_model:
- meta-llama/Llama-3.2-1B
library_name: transformers, deltazip
meta-llama/Llama-3.2-1B - 4b_2n4m_128bs Compression
This is a compressed model using deltazip.
Paper, Compression Tool, Inference Engine (Soon).
Compression Configuration
- Base Model: meta-llama/Llama-3.2-1B
- Compression Scheme: 4b_2n4m_128bs
- Dataset: HuggingFaceH4/ultrachat_200k
- Dataset Split: train_sft
- Max Sequence Length: 2048
- Number of Samples: 256
Sample Output
Prompt:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
Cutting Knowledge Date: December 2023
Today Date: 05 Dec 2024
<|eot_id|><|start_header_id|>user<|end_header_id|>
Who is Alan Turing?<|eot_id|>
Output:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
Cutting Knowledge Date: December 2023
Today Date: 05 Dec 2024
<|eot_id|><|start_header_id|>user<|end_header_id|>
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
Cutting Knowledge Date: December 2023
Today Date: 05 Dec 2024
<|eot_id|><|start_header_id|>user<|end_header_id|>
Who is Alan Turing?<|eot_id|><|eot_id|>!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!