Byte-level analysis shows the alleged leaked DeepSeek V4.1 weights are fake
teortaxesTex · x · 2026-11-01
A claimed leak of DeepSeek V4.1 1017 weights on S3 has been debunked by user toriset via byte-level forensics: every file size and byte exactly matches the publicly known DS V4.1 1017 shards 00013–00042; eu-north-1 doesn't host "Fable"; the math doesn't add up (640 × 7.39GB ≈ 4.73TB, not the claimed 8TB); and you can't distill from 1/640th of a model. Follow-up hex dump analysis shows the first tensor (ffn.gate.biasvl, F32, shape [384]) is precisely DeepSeek's 384-expert VLM MoE router bias, each shard is exactly 7,389,761,368 bytes, and layer dimensions match hiddensize=5120, qlorarank=1280, etc. — meaning the faker simply reused official weight bytes.
More from Models
- Claude Code rolls out bankable usage resets; $200-tier user still says quota not enough — lxfater · 2026-11-01
- Analyst maps OpenAI vs Anthropic coding race: Opus 5.5 sparks Anthropic revenue rebound — firstadopter · 2026-11-01
- DeepSeek V4.1 Flash clocks ~1038 tok/s live via Baseten inference — gajesh · 2026-11-01
- Guess which Halloween sign was one-shot by Gemini and which by Claude — tristanbob · 2026-11-01
- Banteg jokes Claude's 'sex-negative' refusals are 'deeply misaligned behavior' — banteg · 2026-11-01
- Is There a Modality-Agnostic Taxonomy of ML Techniques? Reddit Asks Which Patterns Repeat Across Text, Image and Audio — OkCampaign8400 · 2026-11-01