Unverified50% confidenceFactExact time
The 280.1B parameters retained by Coder average 2.06 bits, no single weight is stored at less than 1 bit, and the pruned-away parameters lower the overall average by being counted as zero bits.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedCompressing DeepSeek Coder to 2-bit loses 67% of passing solutions, whereas the 4-bit version sacrifices only 2% for a 70% reduction in size.70% similarUnverifiedThe official algorithm for "1 bit per parameter" is: 72.12 GB divided by the original 551.9B transformer parameters gives 1.05 bits, rounded to 1; the removed experts are counted as zero bits.69% similarUnverifiedUnder 2-bit quantization, each weight has only about 4 discrete values.64% similarUnverified据B站UP主XCommand的实测，Claude Code单个会话占用386MB内存，jcode仅占用27.8MB，相差约13.9倍63% similarUnverifiedThe Coder version retains 192 experts per layer (half pruned), with a transformer weight size of 72.1 GB, compared with 138.2 GB for the unpruned IQ2_XXS build.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1127278API
curl https://kongchang.com/api/v1/knowledge/claims/1127278MCP
get_claim(id=1127278)