sbintuitions/sarashina2.2-3b-instruct-v0.1
Full fine-tune vs sbintuitions/sarashina2.2-3b. Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 65/65, MLP 96/96, Attention 128/128, lm_head 1/1
23 derivatives checked
- mmnga/sarashina2.2-3b-instruct-v0.1-gguf: Quantization
- sbintuitions/sarashina2.2-ocr: Full fine-tune (fewer layers)
- tatsuyaaaaaaa/sarashina2.2-3b-instruct-v0.1-gguf: Same weights, converted
- DevQuasar-9/sbintuitions.sarashina2.2-3b-instruct-v0.1-GGUF: Same weights, converted
- neody/sarashina2.2-3b-instruct-v0.1-gguf: Same weights, converted
- mradermacher/sarashina2.2-3b-instruct-v0.1-GGUF: Same weights, converted
- mlx-community/sarashina2.2-3b-instruct-v0.1-4bit: Quantization
- FiShota/yamato-3b-v1-legal-gguf: LoRA-style
- cyberagent/ca-reward-3b-ja: Full fine-tune
- DataPilot/ArrowNeo-AME-4x3B-v0.1-MoE: MoE built from dense models
- Aratako/sarashina2.2-3b-RP-v0.2: Full fine-tune
- Aratako/sarashina2.2-3b-RP-v0.1: Full fine-tune
- SousiOmine/sarashina2.2-3b-instruct-v0.1-Pythonic-FunctionCall: LoRA-style
- kurogane/sarashina2.2-3b-OpenMathReasoning-random-100k: LoRA-style
- OsakanaTeishoku/sarashina2.2-3b-cot-sft-20250920: LoRA-style
- kurogane/sarashina2.2-3B-OpenScience-NoReasoning-300k: LoRA-style
- DataPilot/sarashina2.2-3Bx4-moe: MoE built from dense models
- kurogane/sarashina2.2-3b-OpenCodeReasoning-random-100k: LoRA-style
- DataPilot/sarashina2.2-3Bx8-moe: MoE built from dense models
- kurogane/sarashina2.2-5.48B-instruct-no-tuning: Partial fine-tune (fewer layers)
- DataPilot/ArrowNeo-Neko-3B-instruct-v0.2: LoRA-style
- OsakanaTeishoku/sarashina-toolcall-exp-20251213: LoRA-style
- SousiOmine/star-kira-instruct-250426: Full fine-tune
1 other repos hold byte-for-byte identical weight files.