[KongchangAI]
Unverified50% confidenceFactExact time

The entire pipeline from data to running is reproducible, with corpus construction taking about 80 seconds, including Gutenberg corpus cleaning, synthetic task stream generation, BPE, training, and QAT-lite int8 export to the C runtime.

1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027

Sources

MIT Author Builds a 2MB Model That Fits in CPU L3 Cache, Hits 3405 tok/s on a Single Core

Related Entities

CoreFlowproductGutenbergconcept

Related Claims

UnverifiedRestoring an ultra-dense flowchart took approximately 6 minutes and 5 seconds, with the model identifying 20 table flowchart boxes and connecting lines.66% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/1006048
API
curl https://kongchang.com/api/v1/knowledge/claims/1006048
MCP
get_claim(id=1006048)