sbintuitions/sarashina2.2-3b-instruct-v0.1

Text Generation • Updated Mar 5 • 13.9k • 22

Note 2025-03 This is a 3B model, but they claim it is comparable to a 7B class perf, so let's see: https://www.sbintuitions.co.jp/blog/entry/2025/03/07/093143

llm-jp/llm-jp-3-7.2b-instruct3

Text Generation • Updated Feb 4 • 6.24k • 3

Note 2025-02 ; 4096 context window, test with `VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve llm-jp/llm-jp-3-7.2b-instruct3 --max-model-len 8192 --rope-scaling '{"rope_type":"dynamic","factor":2.0}`

Qwen/Qwen2.5-7B-Instruct

Text Generation • Updated Jan 12 • 3.42M • • 665

Note Still has Chinese cross-lingual-token leakage

google/gemma-2-9b-it

Text Generation • Updated Aug 27, 2024 • 355k • • 708

Note No system prompt support breaks many evals

shisa-ai/shisa-v1-llama3-8b

Text Generation • Updated Mar 19 • 28 • 6

Note shisa-v1 SFT only

augmxnt/shisa-7b-v1

Text Generation • Updated Dec 20, 2023 • 65 • 28

Note Requires chat template or BOS token for proper responses