Unverified50% confidenceBenchmarkExact time
The unified model's recall@1 on Russian and English dropped by 3.0 and 4.2 percentage points respectively, a common cost of multilingual fine-tuning.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/10/2026
First Seen
Valid until: 2/8/2027
Sources
Related Entities
Related Claims
UnverifiedOn the Tajik legal retrieval task, the unified model achieved a recall@10 of 92.2, essentially on par with the starting model's 92.4, but recall@1 dropped by 4.4 percentage points.68% similarUnverifiedOn the Belebele dataset, Saidzoda-CA-EmbeddingGemma-300M improved recall@1 for the four languages by 14.4 to 26.0 percentage points, reaching 75.2% to 77.7%.67% similarUnverifiedThe new engine reduces the proportion of complete values visible to 0.3% [0.1; 0.9] in Russian and 0.2% [0.1; 0.7] in British English, respectively.64% similarUnverifiedGPT-5.4-mini's estimates were within 2 percentage points of zero, showing no clear component language assignment effect.63% similarUnverifiedIn the controlled GPT-2 experiments, the unanchored projection update group achieved a target string hit rate of 28.7% under natural language questioning, which dropped to 22.0% after adding module residual anchoring.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153284API
curl https://kongchang.com/api/v1/knowledge/claims/1153284MCP
get_claim(id=1153284)