Cortensor (cortensor)

🛠️ DevLog – Testnet-1 Reset Check & L3 Prep Testnet-1 (COR L3 via RaaS) is now back online after the module reset, and both network + user task flows are running as before. A few node operators have already moved miners over and things look stable at first pass. 🔹 Testnet-1 – RaaS #L3 - Modules re-deployed and baseline network/user tasks are confirming end-to-end flow. - Miners are back on the pool, so we can start watching real traffic again. 🔹 Testnet-1a – Self-Managed #L3 - Still running clean for network tasks after the latest checks. - Basic user task flow is working with multiple miners connected. 🔹 What’s next - Run light stress/load tests on both Testnet-1 and Testnet-1a as prep work for Phase #2. - Use the data to compare RaaS vs self-managed L3 behavior before we scale up.

• 3 replies
• 2 recasts
• 3 reactions

🛠️ DevLog – SessionPaymentTable Live (Dynamic Weights Up Next) Quick follow-up on the new execution fee path: 🔹 The SessionPaymentTable → runtime unit cost wiring is now live and passing initial regression checks with the existing static/fixed unit cost model. 🔹 The dashboard's Unit Cost display is now reading from the same runtime table, so UI + contracts + router are all aligned on pricing. 🔹 Next step is to populate pricing parameters per session config – SLA tier, model class, execution count, and validator depth (PoI / PoUW) – so total cost can be computed as: base unit cost + SLA weight + model weight + execution weight + validator weights. 🔹 Over the coming tests we’ll try different unit prices per execution using this new path, while keeping behavior close to today’s fixed pricing until the dynamic weights are validated.

• 1 reply
• 2 recasts
• 3 reactions

🛠️ DevLog – New LLM & Embedding Models Surfacing in Dashboard 🔹 Dashboard model picker updated New inference models 47–55 and embedding models 56–64 are now wired into the dashboard model selector, with the UI detecting and displaying per-model capacity so it’s clear which models are actually runnable when you create a session. 🔹 Inference + validator stack alignment The additional LLMs expand general routing options, while the embedding models are aimed at upcoming Validator v2/v3 and /validate work, where we need lightweight, reliable vector backends for scoring and matrix calculations. 🔹 Rollout plan These changes are rolling out to both Testnet-0 and Testnet-1, and we’ll keep exercising the new models through real sessions, tuning which ones become defaults for validators and delegate/validate paths.

• 1 reply
• 2 recasts
• 3 reactions

🛠️ DevLog – Next Batch of Ollama Models (47–64, Planned) Lining up the next wave of #Ollama models for this week – nothing is built yet, but this is the target set we’ll start on. 🔹 47–55: New generative models (planning) - nemotron-3-nano:30b-gpu - olmo-3:7b-gpu, olmo-3:32b-gpu - olmo2:7b-gpu, olmo2:13b-gpu - orca-mini:3b-gpu, orca-mini:7b-gpu, orca-mini:13b-gpu, orca-mini:70b-gpu 🔹 56–64: Embedding models for Validator v2 experiments (planning) - nomic-embed-text:v1.5-gpu, mxbai-embed-large:335m-gpu, bge-m3:567m-gpu - all-minilm:22m-gpu, all-minilm:33m-gpu, embeddinggemma:300m-gpu - snowflake-arctic-embed:335m-gpu, snowflake-arctic-embed2:568m-gpu, granite-embedding:278m-gpu 🔹 What’s next - Start building images for 47–64. - Wire them into cortensord on dev-stable for dedicated-node smoke tests. - Use the embedding set to A/B test Validator v2’s similarity matrix and ranking behavior.

• 2 replies
• 2 recasts
• 2 reactions