Qwen/Qwen2-VL-7B-Instruct
Full fine-tune vs Qwen/Qwen2-VL-7B. Fine-tune: 92% of the weights changed, across 28 of 32 layers.
Tensors changed: Embeddings 1/2, Norms 57/187, MLP 88/216, Attention 196/324, lm_head 1/1
61 derivatives checked
- Qwen/Qwen2-VL-7B-Instruct-AWQ: Quantization
- allenai/olmOCR-7B-0225-preview: Full fine-tune
- bartowski/Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- CraftJarvis/minecraft-openha-qwen2vl-7b-2509: Full fine-tune
- unsloth/Qwen2-VL-7B-Instruct-unsloth-bnb-4bit: Quantization
- unsloth/Qwen2-VL-7B-Instruct: Same weights, repacked
- mradermacher/Qwen2-VL-7B-Instruct-i1-GGUF: Quantization (fewer layers)
- ggml-org/Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- second-state/Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- Mungert/olmOCR-7B-0225-preview-GGUF: Full fine-tune (fewer layers)
- lmstudio-community/Qwen2-VL-7B-Instruct-GGUF: Quantization (fewer layers)
- Alibaba-NLP/gme-Qwen2-VL-7B-Instruct: LoRA-style
- unsloth/Qwen2-VL-7B-Instruct-bnb-4bit: Quantization
- mradermacher/Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- allenai/olmOCR-7B-0225-preview-FP8: Quantization
- OS-Copilot/OS-Atlas-Base-7B: Partial fine-tune
- OpenSearch-AI/Ops-MM-embedding-v1-7B: LoRA-style
- Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4: Quantization
- yujiepan/qwen2-vl-tiny-random: Different architecture
- Michael4933/Migician: Partial fine-tune
- DevQuasar-2/Qwen.Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- tensorblock/Qwen2-VL-7B-Instruct-GGUF: Quantization (fewer layers)
- prithivMLmods/coreOCR-7B-050325-preview: Full fine-tune
- Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8: Quantization
- microsoft/GUI-Actor-7B-Qwen2-VL: Full fine-tune
- OS-Copilot/OS-Atlas-Pro-7B: Partial fine-tune
- gaianet/Qwen2-VL-7B-Instruct-GGUF: Same weights, converted (fewer layers)
- mlx-community/olmOCR-7B-0225-preview-bf16: Full fine-tune
- chaoyinshe/EchoVLM: Partial fine-tune (fewer layers)
- erax-ai/EraX-VL-7B-V1.5: LoRA-style
15 other repos hold byte-for-byte identical weight files.