Unverified50% confidenceFactExact time
The study examined the performance of three stages—domain adaptation, quantization, and knowledge distillation—on three models: FANAR-9B, JAIS-8B, and Qwen2.5-7B.
1
Sources
50%
Confidence
Long-term
Relevance
11/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe research team conducted fine-tuning experiments on two models, Qwen3.5-9B and Gemma4-12B-it.77% similarUnverifiedOn Gemma-2-9B and Qwen3-14B, the researchers successfully switched the facts read by the model by modifying early internal activation vectors while keeping the written instructions unchanged.71% similarUnverifiedTesting on Qwen2.5-7B, Llama-3.1-8B, and two DeepSeek models found that selecting memory by relevance reintroduces decoy facts.71% similarUnverifiedThe research team evaluated six KEE techniques across three model families (OLMo2, Qwen2.5, Llama3), covering the three major categories of decoding adjustments, activation steering, and prompt engineering.68% similarUnverifiedThe study used the Qwen2.5-Coder-7B-Instruct model, supplemented by the 1.5B model for sensitivity analysis, completing a total of 13,638 model requests.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1141683API
curl https://kongchang.com/api/v1/knowledge/claims/1141683MCP
get_claim(id=1141683)