Jared Sulzdorf's picture

Jared Sulzdorf PRO

jsulz

·

https://www.jsulz.com/

AI & ML interests

Infrastructure, law, policy

Recent Activity

reacted to danielhanchen's post with 🚀 1 day ago

🦥 Introducing Unsloth Dynamic v2.0 GGUFs! Our v2.0 quants set new benchmarks on 5-shot MMLU and KL Divergence, meaning you can now run & fine-tune quantized LLMs while preserving as much accuracy as possible. Llama 4: https://huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF DeepSeek-R1: https://huggingface.co/unsloth/DeepSeek-R1-GGUF-UD Gemma 3: https://huggingface.co/unsloth/gemma-3-27b-it-GGUF We made selective layer quantization much smarter. Instead of modifying only a subset of layers, we now dynamically quantize all layers so every layer has a different bit. Now, our dynamic method can be applied to all LLM architectures, not just MoE's. Blog with Details: https://docs.unsloth.ai/basics/dynamic-v2.0 All our future GGUF uploads will leverage Dynamic 2.0 and our hand curated 300K–1.5M token calibration dataset to improve conversational chat performance. For accurate benchmarking, we built an evaluation framework to match the reported 5-shot MMLU scores of Llama 4 and Gemma 3. This allowed apples-to-apples comparisons between full-precision vs. Dynamic v2.0, QAT and standard iMatrix quants. Dynamic v2.0 aims to minimize the performance gap between full-precision models and their quantized counterparts.

reacted to danielhanchen's post with ❤️ 1 day ago

🦥 Introducing Unsloth Dynamic v2.0 GGUFs! Our v2.0 quants set new benchmarks on 5-shot MMLU and KL Divergence, meaning you can now run & fine-tune quantized LLMs while preserving as much accuracy as possible. Llama 4: https://huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF DeepSeek-R1: https://huggingface.co/unsloth/DeepSeek-R1-GGUF-UD Gemma 3: https://huggingface.co/unsloth/gemma-3-27b-it-GGUF We made selective layer quantization much smarter. Instead of modifying only a subset of layers, we now dynamically quantize all layers so every layer has a different bit. Now, our dynamic method can be applied to all LLM architectures, not just MoE's. Blog with Details: https://docs.unsloth.ai/basics/dynamic-v2.0 All our future GGUF uploads will leverage Dynamic 2.0 and our hand curated 300K–1.5M token calibration dataset to improve conversational chat performance. For accurate benchmarking, we built an evaluation framework to match the reported 5-shot MMLU scores of Llama 4 and Gemma 3. This allowed apples-to-apples comparisons between full-precision vs. Dynamic v2.0, QAT and standard iMatrix quants. Dynamic v2.0 aims to minimize the performance gap between full-precision models and their quantized counterparts.

reacted to danielhanchen's post with 🤗 1 day ago

🦥 Introducing Unsloth Dynamic v2.0 GGUFs! Our v2.0 quants set new benchmarks on 5-shot MMLU and KL Divergence, meaning you can now run & fine-tune quantized LLMs while preserving as much accuracy as possible. Llama 4: https://huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF DeepSeek-R1: https://huggingface.co/unsloth/DeepSeek-R1-GGUF-UD Gemma 3: https://huggingface.co/unsloth/gemma-3-27b-it-GGUF We made selective layer quantization much smarter. Instead of modifying only a subset of layers, we now dynamically quantize all layers so every layer has a different bit. Now, our dynamic method can be applied to all LLM architectures, not just MoE's. Blog with Details: https://docs.unsloth.ai/basics/dynamic-v2.0 All our future GGUF uploads will leverage Dynamic 2.0 and our hand curated 300K–1.5M token calibration dataset to improve conversational chat performance. For accurate benchmarking, we built an evaluation framework to match the reported 5-shot MMLU scores of Llama 4 and Gemma 3. This allowed apples-to-apples comparisons between full-precision vs. Dynamic v2.0, QAT and standard iMatrix quants. Dynamic v2.0 aims to minimize the performance gap between full-precision models and their quantized counterparts.

View all activity

Organizations

jsulz's activity

New activity in xet-team/README about 1 month ago

xet repositories

#1 opened 2 months ago by

New activity in jsulz/jsulz 5 months ago

[bot] Conversion to Parquet

#1 opened 5 months ago by

parquet-converter

New activity in xet-team/lfs-analysis 7 months ago

The monthly addition figure has empty extension?

#5 opened 7 months ago by

Compressed -> Deduped column header

#4 opened 7 months ago by

LFS Analysis Roadmap

#3 opened 7 months ago by

Suggested text changes

#1 opened 7 months ago by

Filter by file extension chart potential bug

#2 opened 7 months ago by

New activity in jsulz/space-stats 8 months ago

[bot] Conversion to Parquet

#1 opened 8 months ago by

parquet-converter

New activity in jsulz/state-of-the-union-addresses 8 months ago

[bot] Conversion to Parquet

#1 opened 8 months ago by

parquet-converter