Bitcoin Forum
June 14, 2026, 01:51:02 PM *
News: Latest Bitcoin Core release: 31.0 [Torrent]
 
   Home   Help Search Login Register More  
Pages: « 1 2 [3]  All
  Print  
Author Topic: [ANN] KERYX | Live AI Inference | 100% Fee Burn | GPU PoW BlockDAG | 10 BPS  (Read 1249 times)
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 21, 2026, 07:45:18 PM
Last edit: May 21, 2026, 08:05:23 PM by Keryx Labs
 #41

Here is the latest keryx-miner version v0.2.7. Available for Windows, Linux and HiveOS.

https://github.com/Keryx-Labs/keryx-miner/releases/tag/v0.2.7-OPoI

v0.2.7 — Multi-Model Inference & IPFS Distribution

What's new:
Multi-model support The miner now supports two inference models selectable at launch:
DeepSeek-R1-8B GGUF — default, higher quality responses
TinyLlama-1.1B — lightweight mode for lower-end GPUs, enabled with --light
IPFS model distribution
Models are now downloaded via the IPFS gateway instead of HuggingFace. No account, no rate limits, fully decentralized from day one.

Fixes:
Escrow recovery — streamed the HTTP response body to bypass the 10 MB ureq limit that was silently failing on large recovery payloads
Chat templates — DeepSeek now uses the LLaMA-3 chat template, TinyLlama uses Zephyr — fixes malformed inference outputs

Internal:
Escrow claiming logs demoted to debug to reduce noise in production output


EDIT: If your GPU isn't powerful enough to run DeepSeek R1, simply add the --light flag to run the miner with TinyLlama only
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 24, 2026, 06:02:10 PM
Last edit: May 24, 2026, 06:23:50 PM by Keryx Labs
 #42

Keryx Hardfork — DAA 15,550,000

Estimated activation: Thursday, May 28, 2026 — ~14:00 UTC

---
Why This Hardfork

While monitoring the live network following the integration of DeepSeek-R1-8B alongside TinyLlama, we identified a consensus-level vulnerability.

Miners who haven't updated their software are still attempting to answer DeepSeek queries using incorrect or mismatched models. The results they submit are not verifiable as coming from the declared model. We cannot rely on miners voluntarily updating — economic incentives alone are insufficient when bad data can still earn rewards.

The only clean, permanent fix is to enforce model capabilities strictly at the consensus level, with model identity cryptographically tied to the actual weight files. This requires a mandatory hardfork.

---
Why We're Launching Now

We are aware this delays exchange listings by a few days. We made that call deliberately.

Shipping a hardfork after listing creates far more friction — coordinating node operators, miners, and exchanges simultaneously under time pressure. Doing it now, while the network is growing but not yet listed, gives everyone a clean runway. A few days of delay is a reasonable price for a solid, trustworthy consensus from day one.

Network integrity comes first. Always.


---
Inference Rewards — Real KRX Utility
 
Every AiRequest transaction carries a fee in $KRX set by the requester. When a miner answers the request and the result passes the challenge window unchallenged, that fee is released to the miner.

Protocol inference prices:

- TinyLlama-1.1B — 0.5 KRX + 0.3 KRX network fee = 0.8 KRX minimum,
- DeepSeek-R1-8B — 2.0 KRX + 0.3 KRX network fee = 2.3 KRX minimum,

Requesters can set a higher inference fee to prioritize their request — miners will naturally serve the highest-paying requests first.

Miners earn $KRX in two ways:
- Block reward — standard PoW coinbase, same as before,
- Inference fee — paid per completed AiRequest, released after the challenge window,

This is real utility for $KRX: it is the currency you pay to run AI on the network. Not speculation — demand for inference creates direct, measurable buy pressure. Every agent, every dApp, every user that calls an AI on Keryx spends $KRX.

With this hardfork, the AiResponse tx formally links the miner's identity, the committed result, and the reward claim in a single verifiable chain of transactions:

AiRequest tx  →  AiResponse tx  →  challenge window  →  fee released
   (fee locked)     (miner commits)    (1000 blocks)          (to miner)

---
What Changes
  
AiResponse as a Dedicated Transaction,

Miners no longer embed results in the coinbase. Instead, after completing an inference, they submit a standalone AiResponse transaction (subnetwork 0x04) with a fixed 78-byte payload:

[request_hash : 32 bytes]  — links back to the AiRequest tx
[challenge_window_end : 8] — DAA after which the result is final
[response_ipfs_cid : 34]   — IPFS CIDv0 of the result content
[response_length : 4]      — token count

  
Results Live on IPFS,

The actual text is uploaded to IPFS. Only the 34-byte CID is committed on-chain:

- Unlimited response length — no on-chain size constraint,
- Content-addressed — the CID cryptographically proves the content,
- Decentralized — any IPFS node can serve the result,

  
Model Identity Tied to Weight Files,

model_id is now sha2-256(primary_weight_file), extracted directly from the IPFS CID of the model weights. Anyone can verify:

download the CID from IPFS → sha256sum the file → compare to model_id

A miner cannot claim to run TinyLlama or DeepSeek-R1-8B with an arbitrary model_id. The identifier is now a cryptographic fact about the actual files.

OPoI Commitment Updated,

The on-chain commitment for fraud proofs is now response_ipfs_cid[2..34] — the SHA2-256 digest of the IPFS content. Challengers can point to exactly what content hash was committed.

Auto-managed IPFS Daemon,

Miners don't need to manually install IPFS. The miner binary automatically downloads kubo and starts it on first run. One new flag: --ipfs-url (default http://127.0.0.1:5001/).


---
What Miners Need to Do

1. Update to the new miner and node binaries before DAA block 15,550,000
2. Nothing else — kubo installs and starts automatically on first run,

---
What Doesn't Change

- Block structure, PoW, GHOSTDAG — untouched,
- Escrow / OPoI Phase 1–2 — fully compatible,
- All existing AiRequest transactions — still valid,
- Pre-hardfork blocks — still indexed correctly,

The updated binaries for both the node and the miner will be available on GitHub starting tomorrow.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 26, 2026, 04:36:10 PM
 #43

Keryx Node v1.2.1

https://github.com/Keryx-Labs/keryx-node/releases/tag/v1.2.1-OPoI

Hardfork release — This version contains consensus changes incompatible with v1.2.0. Update is mandatory to stay on the network.

  ---
  Highlights

  - UTXO Escrow for inference rewards — inference_reward is now flows through a consensus-validated UTXO escrow, ensuring only legitimate inferences get paid out.
  - Model Cap Enforcement — The consensus now rejects any AiResponse whose model_id is not declared in the coinbase ai:cap: field. Miners must advertise their supported models.
  - Fixed 78-byte AiResponse layout — AiResponse is now a fixed 78-byte structure with an encoded IPFS CIDv1. This breaks binary compatibility with v1.2.0.
  - New supported models — DeepSeek-R1-32B and LLaMA-3.3-70B added, with tiered rewards: 0.5 / 1.5 / 2.5 / 4.0 KRX based on model size.

  ---
  Consensus & OPoI

  - Token-proportional minimum reward: 0.05 KRX per 64 tokens generated
  - model_id = sha2-256(weight_file) derived from the IPFS CIDv1 — deterministic and verifiable off-chain
  - Per-model reward minimums + priority_fee payout to miners
  - Hardfork activation is now logged at the DAA score trigger

  Mempool

  - AiResponse and AiChallenge transactions are exempt from the minimum fee check — they carry their own inference_reward

  P2P / Network

  - Auto-ban phantom nodes after repeated ping timeouts
  - Ban peers on IBD spam and finality violations

  Fixes

  - UPnP: port mapping failure downgraded from WARN to INFO with an actionable message — UPnP is absent on most VPS routers and is not a critical error

  ---

The node will automatically detect the hardfork at the scheduled DAA score (15,550,000, estimated activation on Thursday, May 28, 2026, at ~14:00 UTC) and switch to the new consensus rule
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 27, 2026, 07:01:27 PM
 #44

keryx-miner v0.2.8 — GPU Inference OPoI (Full Pipeline)

https://github.com/Keryx-Labs/keryx-miner/releases/tag/v0.2.8-OPoI

This release delivers the complete Optimistic Proof of Inference pipeline: models now run natively on CUDA, results are published to IPFS, and the miner submits an on-chain AiResponse transaction — all automatically, on every mined block.

What's new

Real GPU inference (CUDA 12.6)
Models are now loaded directly onto the GPU. The miner verifies VRAM budget at startup and places models accordingly — no more silent CPU fallback.

Full OPoI pipeline end-to-end
Mining template arrives → GPU generates a response → result is uploaded to IPFS → an AiResponse transaction (subnetwork 0x04) is broadcast on-chain with the IPFS CID → inference_reward is auto-claimed via escrow UTXO.

Multi-model tiers

--light — TinyLlama 1.1B only, any GPU with 6 GB+ VRAM
(default) — TinyLlama + DeepSeek-R1-8B, RTX 3060 12 GB / 3070 / 3080
--high — adds DeepSeek-R1-32B, RTX 3090 / 4090 (24 GB+)
--very-high — adds LLaMA-3.3-70B, RTX 5090 (32 GB+)
New models: DeepSeek-R1-32B and LLaMA-3.3-70B
Both distributed via IPFS, downloaded on demand at startup.

Model prefetch before mining
Model files are fetched from IPFS before the miner starts looking for blocks — no missed inference jobs on first launch.

Kubo auto-download
The IPFS node (Kubo) is downloaded automatically at runtime if absent — no manual setup required.

Capability announcement
The miner broadcasts its loaded models in the coinbase extra_data (/ai:cap:<model_id>/) so the network knows which models are available before routing requests.

Better response quality
Repeat-penalty sampler (1.15 / 64-token window), multi stop-token support, blocks stripped before publishing to IPFS, stronger Keryx identity system prompt.

Escrow fixes
Inference reward claims are prioritised over coinbase to prevent starvation. Seq-lock cooldown added for retry loops.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 28, 2026, 03:46:13 PM
 #45

🚀 HARDFORK SUCCESSFUL! 🚀

The OPoI Hardfork has been successfully activated at DAA 15550000!

🔥 What's live now:

UTXO escrow is fully functional.

Model cap enforcement is active—meaning you can now test out the 4 AI models!

Huge thanks to all the miners and node operators who upgraded on time. Everything is running smoothly, blocks are rolling in. Let's build! 🛠️


2026-05-28 17:41:14.483+02:00 [INFO ] Accepted 12 blocks ...69ca82046e5e3ef663ac009f3e3bf778fca93695d7b4ad2d6c076a8b9037fa83 via relay
2026-05-28 17:41:15.488+02:00 [INFO ] Accepted 15 blocks ...874149a31c38683907b17b5bc16d5169cf9efb4faa68f7fe98bc30f6da0b045f via relay
2026-05-28 17:41:16.838+02:00 [INFO ] Accepted 11 blocks ...a7ef89ca2385a6cc25b812b9a41942666f774e2a1dbff0b307adb1b658fe2597 via relay
2026-05-28 17:41:17.865+02:00 [INFO ] Accepted 7 blocks ...94c79ef96c0f2adaab8c42b393866b024a710e10e1bff32747e885b58dbd0d11 via relay
2026-05-28 17:41:18.665+02:00 [INFO ] === OPoI HARDFORK ACTIVATED at DAA 15550000 — UTXO escrow + model cap enforcement now live ===
2026-05-28 17:41:19.132+02:00 [INFO ] Accepted 13 blocks ...bb1c48c9e02d2b800a03e77b0508578d569b75a0240cdfa43409e03a1965a257 via relay
2026-05-28 17:41:20.166+02:00 [INFO ] Accepted 15 blocks ...69cad493972058ec73a531d4c8dd63c8b362ce339fb09803e69ee43ca66cdacc via relay
2026-05-28 17:41:21.333+02:00 [INFO ] Accepted 13 blocks ...f19e7469d155bf715e30eaff9db0a2f20dae115f6c37c8c5c2c90b023c65a9bb via relay
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 29, 2026, 08:28:26 PM
 #46

---
⚡️ Emergency Hardfork — KeryxHash SALT v2 | Activating at DAA 17,275,000 (~May 30, 15:00 UTC)

---
Why this hardfork?

We've identified a critical threat to the Keryx protocol: a subset of miners have been modifying their software to bypass OPoI — participating in PoW mining while intentionally not loading any AI model.

The consequences are real and visible:
- AI inference requests on-chain receive invalid responses from these miners
- The live feed stays stuck on PENDING indefinitely when a cheating miner "answers" a request without actually running inference
- The economic incentive of inference_reward is being captured without doing the work

Since protocol integrity is our #1 rule, we are responding immediately with an emergency hardfork.

---
What's new in this hardfork

1. PoW SALT v2 — Forced binary update
The KeryxHash matrix generation now uses a new domain-separation salt (KeryxHash-v2:2026-05-29). Any miner running an old or modified binary will produce invalid PoW hashes — their blocks will be rejected by the network. There is no way around this without updating.

2. OPoI Challenge system
The node now actively challenges miners at random intervals. When you request a block template, the node may send back a challenge: run inference on a specific prompt with a specific nonce, and include the result in your next request. Miners that can't answer (no model loaded) stay is_synced=false and cannot mine.

3. Model integrity enforcement
The miner now verifies that model files are fully downloaded and intact before allowing mining to start. Partial downloads or missing models block the miner at launch — not silently at inference time.

4. Capability announcement
Every coinbase now declares which AI models the miner supports (ai:cap:<model_id>). The network can see in real-time which miners are honest participants. This data feeds the dApp's miner capability check before submitting inference requests.

5. Challenge anti-replay
Challenge responses must include the exact nonce issued by the node. Replaying old responses or faking results is detected and the miner stays blocked.

---
What you need to do

Update both before DAA 17,275,000 :

- Miners → v0.2.9
- Node → v1.2.2

New releases will be available tomorrow morning.

After activation, any node or miner running an older version will be rejected by the network — old nodes won't validate SALT v2 blocks correctly and will end up on a dead fork.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
May 30, 2026, 05:04:50 PM
Last edit: June 01, 2026, 06:19:36 PM by Welsh
 #47

Hardfork objective 100% accomplished! Ai Inference is working perfectly, all miners on the network are now running inference!

▸ LIVE INFERENCE FEED                 auto-refresh 5s

-----------------------------------------------------------------------------------------------------------


e6d3b2acd8…ecc87b     DeepSeek-R1-32B     ✓ RESPONDED     DAA 17,289,121

Can you give me a detailed introduction to the Keryx blockchain?

reward 2.9 KRX     fee 0.3 KRX

SLM →     block f7a1cb…33ee7b

The Keryx blockchain is an innovative platform designed to integrate decentralized AI with advanced consensus mechanisms. Here's a structured overview:

1. **Decentralized AI Inference**:
   - Keryx enables distributed participation where miners contribute computational power for AI tasks, enhancing security by eliminating central points of control.

2. **BlockDAG Consensus Mechanism**:
   - Utilizing Directed Acyclic Graphs (DAG), BlockDAG allows parallel processing and transactions, potentially increasing speed compared to traditional linear blockchains.

3. **Optimistic Proof of Inference (OPoI)**:
   - Replaces traditional proof-of-work with AI inference tasks, where miners' computational efforts contribute both to security and network functionality, enhancing efficiency.

4. **On-Chain Governance**:
   - Decisions are made transparently via on-chain votes, fostering a community-driven approach without intermediaries.

5. **Interoperability**:
   - Designed to work with other blockchains, facilitating asset transfers and data sharing across platforms, reducing fragmentation.

6. **Sustainability**:
   - Emphasizes energy efficiency by integrating AI tasks, potentially lowering energy consumption compared to traditional chains like Bitcoin or pre-transition Ethereum.

7. **Potential Considerations**:
   - Clarification on how AI tasks are rewarded (accuracy vs effort) and BlockDAG's conflict resolution mechanisms.
   - Details on interoperability protocols and applications that leverage Keryx's unique features.
   - Comparative analysis of energy efficiency with other chains to validate sustainability claims.

In summary, Keryx offers a forward-thinking approach by combining AI with DAG structures for scalability, efficiency, and sustainability. While its concepts are promising, further technical insights would clarify operational mechanics and potential challenges.

block 3a29a84b…e0cd47     max_tokens: 512


-----------------------------------------------------------------------------------------------------------


fc46355c58…d7555f     DeepSeek-R1-8B           ✓ RESPONDED     DAA 17,286,647

Can you give me a detailed introduction to the Keryx blockchain?

reward 1.7 KRX     fee 0.3 KRX

SLM →     block 01bbdf…bfcbc6

I am Keryx Network AI, a decentralized assistant built on the Keryx BlockDAG protocol. My operations are performed by GPU miners as proof-of-work, with results secured via OPoI (Optimistic Proof of Inference). I rely solely on training knowledge and cannot access live data or external resources.

block 21f1741e…03a91a     max_tokens: 256


-----------------------------------------------------------------------------------------------------------


fd9fe337d0…93451a     DeepSeek-R1-32B     ✓ RESPONDED     DAA 17,278,550

1+x=2; x=?

reward 2.7 KRX     fee 0.3 KRX

SLM →     block ac15d2…7e5b6a

To solve for \( x \) in the equation \( 1 + x = 2 \):

1. Subtract 1 from both sides:
   \[
   1 + x - 1 = 2 - 1
   \]
2. Simplify:
   \[
   x = 1
   \]

**Answer:**  
\( x = 1 \)
block 7c46ffdc…8c603f     max_tokens: 256


-----------------------------------------------------------------------------------------------------------

2bd8c862c2…42da67     TinyLlama-1.1B     ✓ RESPONDED     DAA 17,278,483

1+x=2; x=?

reward 0.7 KRX     fee 0.3 KRX

SLM →     block 98870c…2e2f43

No, the given equation doesn't have an integer solution for x = 2 because it is not satisfied by any values of x that satisfy both 1 + x and x.

block c718467e…e045fa     max_tokens: 256
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 03, 2026, 02:45:03 PM
 #48

keryx-miner v0.3.0 — Full Pool OPoI Integration

https://github.com/Keryx-Labs/keryx-miner/releases/tag/v0.3.0-OPoI

This release brings complete Optimistic Proof of Inference (OPoI) support to pool mining (stratum), achieving full parity with the solo miner. Pool miners now declare their loaded models, receive capability challenges, and submit inference CIDs alongside every share — no exceptions.

---
What's New

Pool Mining — OPoI Full Parity

- mining.declare_capabilities — miner declares loaded model IDs after authorize; the bridge uses this for AiTask routing and capability challenges
- mining.challenge handler — pool miners respond to periodic inference challenges (60 min interval, aligned with the solo node); PoW is paused during the challenge so the GPU is fully available for inference
- Anti-replay nonce — challenge response now echoes the nonce issued by the bridge, preventing replay attacks (mirrors solo behavior)
- OPoI tag on every share — tag_fixed(nonce) is submitted with each share as proof of inference
- MiningSubmitWithCID — inference result CID is attached to share submissions for on-chain AiResponse TX
- No models = no mining — mining is suspended until at least one model is loaded and ready (mirrors the solo gate, no exceptions)
- PoW paused during AiTask inference — running kHeavyHash and SLM inference simultaneously crashes the GPU; PoW is now correctly suspended for the duration of any AiTask

Stability & Bug Fixes

- Rejected share is now non-fatal — miner logs a warning and continues (mirrors solo behavior; previously caused disconnection)
- Hardened GPU inference engine — improved robustness under concurrent load

CUDA / PTX

- All PTX kernels (sm61, sm75, sm86, sm89) recompiled with CUDA 12.2 — ensures driver compatibility ≥ 535 across all supported architectures, including HiveOS rigs
- sm100 (Blackwell / RTX 50xx) PTX retained from CUDA 13.2 (requires driver ≥ 570)

Stratum

- DAA score pool support (contributed by @BaikalMine)

---
If your miner is working fine, this update isn't necessary. It is only useful for pool miners or anyone running SM75-supported GPUs, such as the GTX 1660 Super, 1660, 1660 Ti, or RTX 20-series cards
chafer99
Member
**
Offline

Activity: 102
Merit: 10


View Profile
June 03, 2026, 09:14:43 PM
Last edit: June 04, 2026, 07:08:24 AM by chafer99
 #49

Miners - 0.2.9, 0.3.0 cannot be run on AMD GPU. Displayed that missing Cuda toolkit 12.6, but i have only AMD cards.  Command --cuda-disabled has been removed.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 04, 2026, 09:38:20 PM
 #50

To be honest, I thought AMD cards wouldn't be able to mine on our network because performing inference is mandatory to mine, and our inference runs through a candle-core module that is only compatible with CUDA 12.6. It’s even still written in our whitepaper that AMD GPUs couldn't participate.

But a Discord member found a workaround. He installed CUDA 12.6 even though he only has AMD GPUs, and then launched his miner using these commands:

./keryx-miner --mining-address keryx:YOUR_ADDRESS_HERE --light --cuda-disable --opencl-no-amd-binary

This way, he manages to pass the OPoI challenge because his CPU handles the inference using only TinyLlama
HdCe
Newbie
*
Offline

Activity: 1
Merit: 0


View Profile
June 04, 2026, 10:17:02 PM
 #51

I have two windows machines that I used to use for mining. Both have RTX50 series cards and runninf win11 pro.

One is mining in pool and another one has own node running. Both worked on 0.2.8 miner.

The one with own node (node updated) and working. Gave some missing cublas error... when tried to run miner 0.2.9 and I was not able to figure out what was the problem.
Today I noticed 0.3.0 miner and tested it with the machine without node and using pool. 0.3.0 miner starts to load models but allways fails to download that 4gb model. Smaler model was downloaded successufully.

Any ideas what might be wrong?
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 04, 2026, 10:29:24 PM
 #52

I think the issue comes from CUDA 12.6. Since version 0.2.9, inference has become mandatory on the network. The node challenges the miner at startup to verify if it meets the prerequisites for the declared models (fully downloaded models, enough VRAM, etc.). As I mentioned earlier, we use candle-core for inference, which is only compatible with CUDA 12.6, whereas the 50xx series cards are not compatible with CUDA 12.6...

I'll look into this and see if I can find a solution or a fix for the 50xx series GPUs
chafer99
Member
**
Offline

Activity: 102
Merit: 10


View Profile
June 05, 2026, 12:14:51 AM
 #53

To be honest, I thought AMD cards wouldn't be able to mine on our network because performing inference is mandatory to mine, and our inference runs through a candle-core module that is only compatible with CUDA 12.6. It’s even still written in our whitepaper that AMD GPUs couldn't participate.

But a Discord member found a workaround. He installed CUDA 12.6 even though he only has AMD GPUs, and then launched his miner using these commands:

./keryx-miner --mining-address keryx:YOUR_ADDRESS_HERE --light --cuda-disable --opencl-no-amd-binary

This way, he manages to pass the OPoI challenge because his CPU handles the inference using only TinyLlama

Hi, i installed Cuda toolkit 12.6, but the miner still reports "Missing cuda toolkit 12.6, cuda runtime libs missing...". Command  --cuda-disable, has been removed from Miners - 0.2.9, 0.3.0. Previous miners perfectly worked on AMD GPU.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 05, 2026, 05:36:41 PM
 #54

Keryx Node v1.2.4 — Coordinated chain relaunch (hardfork)

https://github.com/Keryx-Labs/keryx-node/releases/tag/v1.2.4-OPoI

MANDATORY UPGRADE — activates at DAA score 21,932,751

All node operators, pools and miners must upgrade to v1.2.4. This release relaunches the chain through a coordinated, height-gated hardfork after the network froze following v1.2.3.

Why this is mandatory

v1.2.3 correctly removed the non-deterministic OPoI slashing path, but it was not enough to heal an already-fragmented network. By the time slashing was disabled, nodes had diverged onto per-node branches and the chain had effectively frozen: no new UTXO-validated blocks, difficulty stuck at its pre-incident level, and no common tip for a permissive upgrade to converge back to.

A permissive change can only keep agreeing nodes in sync — it cannot force diverged nodes back onto a single chain. Restoring one canonical chain requires a shared restart point, agreed by everyone at the same height. That is what v1.2.4 does.

What changed (this is a hardfork)

Activated at DAA score 21,932,751 (the last common valid tip of the frozen chain):

- New PoW matrix salt (KeryxHash v3). Blocks at or above the activation height must use the v3 salt. This cleanly isolates the relaunched chain from any residual old-salt blocks and from the stalled v2 branches — old-salt blocks above the activation height are rejected.
- Difficulty reset at the fork point. The difficulty window resets to genesis_bits at activation, so mining can restart from a clean baseline instead of being pinned to the pre-freeze hashrate (which would otherwise make the chain unmineable).
- Slashing remains disabled (inherited from v1.2.3). UTXO computation stays deterministic; escrows remain always-spendable.

Blocks below the activation height keep their inherited salt (v1/v2) and difficulty — history is preserved, only the relaunch point changes.

Important notes

- This IS a hardfork with an activation height. Unlike v1.2.3, upgrading is not optional and not permissive: nodes and miners that do not run v1.2.4 cannot produce or accept valid blocks above DAA 21,932,751. Coordinate your upgrade.
- Miners must also update to keryx-miner v0.3.1, which implements the v3 PoW salt. Older miners will mine invalid (rejected) blocks past the activation height.
- Slashing is still off. A correct, deterministic slashing mechanism (verifiable commitment + first-writer-wins + challenger deposit) will return in a future coordinated release (OPoI v2).
- Pre-existing slashed escrows remain spendable by their owners.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 06, 2026, 08:48:39 PM
 #55

Keryx v1.2.5 — Difficulty Re-Anchor Recovery

https://github.com/Keryx-Labs/keryx-node/releases/tag/v1.2.5-OPoI

MANDATORY UPGRADEbefore DAA 22035000

Every node — including solo miners' and pools' nodes — must be running v1.2.5 before the chain reaches DAA 22035000. Any node still on v1.2.4 will diverge (fork) at that point.

---

The bug being fixed

Since the v1.2.4 relaunch, difficulty stayed stuck at its floor (genesis). At that level the PoW is trivial: the network produces a flood of blocks, the DAG blows up, and the nodes' (single-threaded) virtual resolution can no longer keep up.

Visible symptoms:
- the validated chain advanced at only ~0.3 blocks/s (far below the 10 BPS target);
- miners saw their hashrate collapse (e.g. ~6 MH/s on an RTX 3090) — the GPU was actually idle waiting for templates, not slowed down;
- the difficulty auto-adjustment could not recover on its own (validated production < target → difficulty stays at the floor → the flood continues).

Confirmed by profiling: the bottleneck was the nodes' virtual resolution choked by an abnormal DAG — not he miner, not the algorithm, not OPoI.

The fix

v1.2.5 introduces a difficulty re-anchor gated at a future DAA:
- at DAA 22035000, the difficulty window restarts from a sane value (`0x1d080000`, calibrated for ~10 BPS) instead of the floor;
- the standard auto-adjustment then takes over and converges to the network's real hashrate within ~1–2h.

The spiral is broken cleanly, at a known height announced in advance.

Action required

1. Update `keryxd` to v1.2.5 (build / provided binary).
2. Restart your node before DAA 22035000.
3. That's it — the mining binary does not change, this only affects the node.

Compatibility & safety

- Before the gate, v1.2.5 behaves identically to current nodes → you can upgrade right now with no fork risk.
- Divergence happens only at DAA 22035000, and only for nodes still on v1.2.4.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 09, 2026, 09:52:30 AM
Last edit: June 12, 2026, 08:32:02 PM by Welsh
 #56

Hey guys, I recently came across SharkWallet — a clean multi-currency desktop wallet that lets you easily add custom coins and tokens.
Thought it might be useful for the team or community once KRX gets more widely available. Looks promising for holding and managing different assets in one place.
What do you think?

Thanks for the suggestion! I'm not familiar with SharkWallet, but I'll check it out and see how it fits. Thank you for your interest in KRX!

Keryx Node v1.2.6 — SALT v4 & Virtual Processor Fix

https://github.com/Keryx-Labs/keryx-node/releases/tag/v1.2.6-OPoI

Mandatory update for ALL node operators and miners. v1.2.6 activates a new PoW salt (SALT v4)

Critical fix: resolves the virtual-processor stall that was throttling the chain to ~0.3 BPS and starving miners.
Clean relaunch on stock Kaspa difficulty (SALT v4) — the experimental difficulty re-anchoring from v1.2.4/v1.2.5 is gone. Difficulty is back to vanilla Kaspa auto-adjustment.
No data loss: full history, balances and archival state are preserved.

What was wrong
Since the relaunch, the node's single-threaded virtual processor was spending ~3.8 seconds per block inside sink_search_algorithm. It was repeatedly trying to reorg onto an abandoned, higher-blue-work branch left in the datadir (the old salt-v3 chain, stored header-only), walking the entire dead branch on every block — ~425,000 iterations per resolve. The GPU sat idle waiting for templates, the chain crawled at ~0.3 BPS, and difficulty never recovered.

This was not a difficulty problem — every previous diff re-anchor only masked it cosmetically.

The fix
sink_search_algorithm now skips candidates that can never be the sink — header-only blocks (no UTXO state) and blocks already disqualified from the chain — without walking their parents. The valid sink is always reachable through the body tips already in the heap, so the result is identical (same sink, same candidate set) — it just no longer wastes the walk.

Impact:
- Sink_search iterations: 425,047 > 22 iterations/block
- Sink_search time: 3.78s > ~2ms/block
- Block rate: ~0.3 BPS > ~10 BPS
- GPU Hashrate: 6 MH/s > 761 Mh/s

This is a correct, fork-safe change (it never alters which sink is selected) and it generalizes: any node will now shrug off abandoned high-blue-work branches instead of choking on them.
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 10, 2026, 08:45:15 PM
Last edit: June 11, 2026, 01:21:09 PM by Welsh
 #57

Suprnova just opened a KRX pool!

https://krx.suprnova.cc/

Keryx Miner v0.3.2 — SALT v4

https://github.com/Keryx-Labs/keryx-miner/releases/tag/v0.3.2-OPoI

Mandatory update. Any miner still on v0.3.1 or earlier will produce invalid shares.

What's changed

SALT v4 — new PoW domain

The kHeavyHash matrix salt switches to KERYX:KeryxHash-v4:2026-06-07:xx, blocks mined with the old salt will be rejected by the network after this point.

Escrow — I/O performance

- O(1) deduplication via HashSet (replaces O(N) scan)
- Per-block HashMap index for O(reds) lookup (replaces O(chain) walk)
- 2-second save_state debounce — eliminates redundant disk writes per block
- Periodic state file compaction

gRPC — larger channel buffer

Channel capacity 2 → 1024 — removes back-pressure stalls during block bursts.

Build

CUDA 12.2 — PTX loadable on driver ≥ 535 (HiveOS compatible)
Architecture: sm75 (RTX 16xx / 20xx / Tesla T4)
For RTX 30xx/40xx: use the sm86 package
Keryx Labs (OP)
Newbie
*
Offline

Activity: 34
Merit: 0


View Profile
June 13, 2026, 07:35:44 PM
 #58

The first tests are successful!

The new feature --vram-pool that utilizes the entire rig's VRAM is working. I managed to run and answer a DeepSeek-R1 32B inference using my rig with 5x 3060 Ti

[2026-06-12T21:17:29Z INFO  keryx_miner::slm] SlmEngine: --vram-pool — splitting 'deepseek-r1-32b' layers evenly across 5 GPUs
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] OPoI challenge in progress — PoW paused, stand by
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] Device #4 (NVIDIA GeForce RTX 3060 Ti): OPoI challenge in progress — stand by
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] Device #1 (NVIDIA GeForce RTX 3060 Ti): OPoI challenge in progress — stand by
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] Device #2 (NVIDIA GeForce RTX 3060 Ti): OPoI challenge in progress — stand by
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] Device #0 (NVIDIA GeForce RTX 3060 Ti): OPoI challenge in progress — stand by
[2026-06-12T21:17:23Z INFO  keryx_miner::miner] Device #3 (NVIDIA GeForce RTX 3060 Ti): OPoI challenge in progress — stand by
[2026-06-12T21:17:28Z INFO  keryx_miner::client::grpc] OPoI: sending challenge response model=94296733


Thanks to this, I'm going to add a massive model option for miners with big rigs (like 6x 3090s), for example Qwen3-235B-A22B
Pages: « 1 2 [3]  All
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!