质疑：量化可能损害了 GLM 的视觉性能
Sentdex · x · 2026-08-31
Sentdex questions the evaluation results of GLM 5.3 Flash, suspecting the use of a quantized version which may have hindered vision performance. He notes a discrepancy between the评测 data and real-world usage, and is seeking unlabeled source images for independent verification.
More from Models
- PhoneLLM benchmarks: P95 latency under 600ms, ~$0.0025 per minute — altryne · 2026-08-31
- Challenge: make LLMs draw a plane in 3D using a single brush tool — SeesawGullible398 · 2026-08-31
- Stop conflating bad AI detection tools with Pangram — TuhinChakr · 2026-08-31
- LTX-2.5 forces voiceover generation, issue resolved by reverting to LTX-2.3 — zeddinh707 · 2026-08-31
- Tests show GLM 5.3 Flash is weak at vision tasks — bhq322 · 2026-08-30
- Leak: Google Gemini 3.8 Flash coming soon with major quality boost — mark_k · 2026-08-30