Unverified50% confidenceFactExact time
Ollama downloads Q4 quantized models by default, which have small VRAM usage but lower precision and more pronounced hallucination behavior.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/16/2026
First Seen
Valid until: 12/15/2026
Sources
Related Entities
Related Claims
UnverifiedThe lower the quantization bit depth, the more likely the model is to show degraded performance on detailed reasoning, but for most tasks at quantization levels above 4-bit, the experiential difference is not obvious to ordinary users.65% similarUnverifiedGGUF量化后的Q4模型在大多数任务上仍能保留原始模型90%以上的能力64% similarUnverifiedMultimodal large models have extremely low factual recall accuracy in video question answering, which is difficult to improve even with supervised fine-tuning.64% similarVerifiedA 2023 Stanford University study showed that multimodal models such as GPT-4V, when parsing flowcharts with nested conditional logic, had accuracy approximately 15–25 percentage points lower than equivalent textual descriptions.64% similarUnverifiedOLLAMA_KV_CACHE_TYPE=q8_0 occupies approximately only half the memory of the f16 default, with minimal quality loss63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924992API
curl https://kongchang.com/api/v1/knowledge/claims/924992MCP
get_claim(id=924992)