Qwen/Qwen3.5-0.8B
Full fine-tune vs Qwen/Qwen3.5-0.8B-Base. Fine-tune: 90% of the weights changed, across 24 of 24 layers.
Tensors changed: Attention 148/220, Norms 87/135, Embeddings 3/5, MLP 79/127, other 1/1
306 derivatives checked
- ATH-MaaS/OvisOCR2: Full fine-tune
- Hcompany/Holo-3.1-35B-A3B-GGUF: Not comparable yet
- unsloth/Qwen3.5-0.8B-GGUF: Same weights, converted
- ggml-org/Qwen3.5-0.8B-GGUF: Same weights, converted
- unsloth/Qwen3.5-0.8B: Identical copy
- bartowski/Qwen_Qwen3.5-0.8B-GGUF: Same weights, converted
- lmstudio-community/Qwen3.5-0.8B-GGUF: Quantization
- Hcompany/Holo-3.1-0.8B: Full fine-tune
- SpeakoFlow/speakoflow-mini: LoRA-style
- unsloth/Qwen3.5-0.8B-MTP-GGUF: Same weights, converted
- Hcompany/Holo-3.1-4B: Different architecture
- reaperdoesntknow/Qwen3.5-0.8-Cyber-GGUF: Full fine-tune
- Jackrong/Qwen3.5-0.8B-Claude-4.6-Opus-Reasoning-Distilled-GGUF: Full fine-tune
- mlx-community/Qwen3.5-0.8B-MLX-4bit: Quantization
- dgpl/dgpl-linux-assistant-0.8b: Same weights, converted
- ML-Intern-lab/Qwen-Image-2.1-PE-T2I-Pocket-0.8B: Full fine-tune (fewer layers)
- AxionML/Qwen3.5-0.8B-NVFP4: Quantization
- Ninnix96/Qwengram-0.8B: Different architecture
- saidutta69/Qwen3.5-0.8B-heretic: Abliterated
- SZLHOLDINGS/chaski: LoRA-style
- lmstudio-community/Qwen3.5-0.8B-MLX-8bit: Quantization
- Mungert/Holo-3.1-4B-GGUF: Not comparable yet
- palette-lab/songgot-x-0.8b: Full fine-tune (fewer layers)
- jeff-legacy/Jeff-Qwen3.5-0.8B: Partial fine-tune
- vultr/VultronRetrieverFlash-Qwen3.5-0.8B: LoRA-style
- lmstudio-community/Qwen3.5-0.8B-MLX-4bit: Quantization
- NobodyWho/Qwen_Qwen3.5-0.8B-GGUF: Same weights, converted
- UT-IFML/dQwen3.5-0.8B-Base: Full fine-tune (fewer layers)
- matthewj-t6/Qwen3.5-0.8B-GGUF: Same weights, converted
- codecho/Qwen3.5-0.8B-text-only: Same weights, repacked (fewer layers)
52 other repos hold byte-for-byte identical weight files.